THE COMPLETE TECHNICAL PAPER PROCEEDINGS FROM:

AN EVALUATION OF ALTERNATIVE TECHNOLOGIES FOR INCREASING NETWORK INFORMATION CAPACITY Ron Shani, Xtend Networks David Large, Consultant

Abstract interchangeably with “information capacity” and “RF bandwidth” will be used when the HDTV, VOD, ITV and other applications historical meaning is intended). The are placing ever-greater pressure on increasing bandwidth demands fall into three operators to transport more information – broad categories: that which is widely distributed as well as communications with individual customers. 1) Common downstream (“broadcast”) Choosing how to create adequate capacity is bandwidth; that is, bandwidth occupied by difficult; driven by financial and regulatory signals that are transmitted throughout the constraints, capital costs and ongoing network (irrespective of whether or not operating considerations. individual customers are enabled to receive them). An example of common signals would This paper will evaluate some of the be a high-definition stream from HBO that technical options against those factors. would be continuously transmitted system- Evaluated technologies will include wide, but for which only certain subscribers bandwidth expansion to 1 GHz, more efficient would be authorized. modulation, more efficient video encoding, elimination of analog video carriage, splitting 2) Interactive downstream (“unicast”) of existing nodes, switched digital video and a bandwidth; that is, bandwidth occupied by proposed use of frequencies above 1 GHz that signals that are transmitted to individual offers the greatest bi-directional bandwidth customers. VOD, Internet communications expansion and the greatest benefit/cost ratio. and telephone are all examples of such signals.

INTRODUCTION 3) Upstream bandwidth; that is, bandwidth occupied by signals that are transmitted from Bandwidth Pressures individual customers towards the headend. With the exception of a small amount of The history of cable television is one of bandwidth occupied by network element ever-increasing need for information capacity, management systems (NEMS), all upstream initially driven by the expansion of over-air signals fall in the same category as interactive broadcasting, then premium and ad-supported downstream bandwidth. satellite networks, followed by pay-per-view and high-speed data services. Today, The Case for Dramatic Bandwidth Increase operators are launching bandwidth-intensive high-definition television (HDTV) channels, Historically, manufacturers have offered various flavors of video on demand (VOD), cable operators increases in upper higher Internet access data rates, and downstream RF bandwidth limits in steps of telephone services. Each of these increases 50 MHz or so from an upper frequency limit the need for an increase in system information of 220 MHz to 860 MHz, with the upstream capacity (for purposes of this paper, unless bandwidth remaining fixed, except for one otherwise specified, “bandwidth” will by used step from 30 to 42 MHz. By contrast, in the data world, speeds have increased 4) Direct broadcast satellite operators will be exponentially over several orders of taking advantage of new spectrum, closer magnitude. As more content carried over satellite spacing, higher power and spot beam cable systems is digital in nature, more technology to realize greatly increased communications are directed to and from throughput – as much as 18,000 MB, or individuals, and competitors greatly increase enough to carry 2800 high-definition both video and non-video capacity, the programs at 6.5 Mb/s/program using question is whether operators will need to advanced codecs.iv significantly increase bandwidth, especially upstream bandwidth, to take advantage of 5) In general, television is moving from pre- opportunities and meet competition. scheduled broadcast of standard-resolution programs to on-demand presentation of high- A few points to consider: definition, with a 4X increase in bits per stream and the need to send programming to 1) On the competitive data front, SBC and (and receive communications from) individual Verizon, among other telcos, have launched a subscribers. Competitively, one satellite major fiber-to-the-curb/home push. Typical operator expects to offer its customers 150 of the technology to be deployed is Wave7’s national and 500 local HDTV channels by equipment which provides 500 Mb/s 2007 symmetrical data, shared among 16 passings, in addition to 860 MHz of RF downstream 6) Finally, upstream data communications bandwidth.i Verizon is offering data rates to rates from subscribers are increasing rapidly. 30 Mb/s downstream/5 Mb/s upstream in its VoIP is a symmetrical service; file sharing fibered markets, with the capability to offer can be symmetrical or even asymmetrical in rates of hundreds of megabits per second.ii the upstream direction; and near-future Some overbuild competitors in the US have services such as video telephony will require already offered 100 Mb/s service options to multiples of the bandwidth required for voice. customers and speeds of between 10 and 100 Comcast recently announced plans to offer Mb/s are commonly available in Asia. video instant messaging. RCN now offers a Finally, the capability of copper plant video surveillance service that allows continues to improve and now supports high- customers to stream video from up to four definition digital video. cameras through their broadband connection.v

2) Cable operators are already being pushed In summary, there is significant evidence to significantly increase rates – Comcast that cable operators will need major increases announced a standard rate of 4 Mb/s and an in bi-directional information capacity in the available 6-Mb/s downstream/768-kb/s near future, and that the upstream in upstream rate; Cox increased its standard rate particular, with a current capacity of only to 4 Mb/siii and RCN has upgraded its rates to about 100 Mb/s/node, is a major bottleneck 10 Mb/s. that will need to be addressed.

3) On the telephone side, the number of VoIP Operators can realize this increased residential and small business lines is information capacity through an increase in predicted to hit almost 11 million by 2008, RF bandwidth, through more efficient use of with a significant amount of that traffic existing bandwidth, or through more efficient carried over cable systems. sharing of existing bandwidth. Additionally, increased interactive bandwidth can be realized by sharing of the bandwidth devoted to interactive services among fewer fixing basic problems, while other systems customers. The various upgrade technologies may require little incidental preparation. that will be considered differ in their effects on broadcast verses interactive and Methodology downstream verses upstream information capacity, as will be seen. For ease of comparison, each technology was evaluated as a candidate for upgrading a Candidate Technologies hypothetical 100,000 home cable system which currently has 500-home nodes and an There are many approaches to generating average density of 100 homes per plant mile. more information capacity in a cable system. It is assumed to be 80% aerial plant. The This paper will evaluate the following connected household penetration is assumed possibilities: to be 70%, with 35% of connected homes equipped for digital video reception. The 1) An increase in downstream upper RF system is assumed to currently carry 80 bandwidth limit from 550, 750 or 870 MHz to channels of analog video, 136 total standard- 1 GHz. resolution broadcast digital video streams, 12 2) An increase in digital modulation density high-definition broadcast digital video from 256 QAM to 1024 QAM. streams, VOD, high-speed data, and VoIP. 3) Utilization of more effective digital video Unless otherwise stated, the system is compression technologies, such as MPEG-4. assumed to have been upgraded to 750 MHz 4) Subdivision of existing optical nodes. within the previous ten years. Other 5) Elimination of analog video carriage, with assumptions regarding the system will be the formerly-analog signals transmitted only discussed when relevant to each individual in digital form. candidate technology. 6) Use of switched digital video to avoid sending low-usage channels to subscriber Technologies were evaluated with respect groups except when requested. to their effect on both downstream and 7) Use of RF bandwidth above 1 GHz to upstream capacities and with respect to both expand both downstream and upstream commonly delivered (broadcast) and capacity. interactive services. In each case, the technologies were also evaluated qualitatively This is obviously not a comprehensive list, with respect to future enhancement options. and the choices are not mutually exclusive. Finally, conformance of each alternative to For example, an operator may choose to current regulatory requirements is noted. simultaneously increase modulation density and also use advanced digital compression INCREASE TO 1 GHz BANDWIDTH algorithms. For keep the matrix manageable, however, we evaluated each option separately. An increase in the upper downstream frequency limit to 1 GHz follows the When it comes to discussing quantitative traditional pattern of cable RF bandwidth results, we used what we felt were reasonable expansion. While it offers additional assumptions for an average cable system. For downstream capacity, it does not address the every possible upgrade scenario, however, the upstream bottleneck and does not offer a results will vary depending on the assumed straightforward path to future increases, as condition of the unmodified plant. For discussed below. example, a marginal system may not be able to take advantage of 1024 QAM without Distribution Network Issues standards. Furthermore, since the FCC has adopted SCTE 40 into its rules, operators are The cost of coaxial equipment upgrade will forbidden from offering one-way digital video depend on the starting bandwidth and on the services above 864 MHz. condition of the original plant. Variables include: the percentage of passive devices We therefore assumed that, while the which are already rated at 1 GHz, whether upgrade would create additional capacity upgrade modules are available for actives, between the existing upper limit and 864 whether the gain of the new actives will be MHz, the space above that would be limited sufficient to avoid re-spacing, the condition to services that need be received only on CPE and type of original coaxial cable and provided by cable operators. Of the available connectors, and whether the increase in drop choices, the most logical seemed to be cable loss is such as to require replacement. simulcasting of the existing analog programming (the first step to an eventual all- The tradeoffs in a bandwidth increase are digital plant and recovery of the spectrum well known. If the amplifier spacing does not now used for analog transmission) to digital- change, each amplifier must have higher gain only converters. We estimated the cost of and either the input levels will be lower simulcasting from a report on Charter’s Long (degrading C/N), the output levels must be Beach, CA conversionvi and estimated the cost higher (degrading distortions), or the of the digital-only converters at $85vii, the amplifier must have higher power output recovery value of the old converters at $25, hybrids (increasing power consumption and and the labor cost to make the change at $10. heat). Furthermore, the number of signals Thus, the estimated cost includes the CPE carried will presumably increase, further changes necessary before the expanded increasing intermodulation products. bandwidth can be used, but not the cost of Alternately, amplifier spacing can be adding any new services. decreased, but then the number of cascaded amplifiers increases, degrading both noise and Using these assumptions, the total cost and distortion. Thus, this technology is self- gained downstream bandwidth (in equivalent limiting and does not offer a solution to future 6-MHz channels) is as follows: expansion. Original Bandwidth 550 750 860 Our estimates are based on figures Cost/HP $274 $116 $81 developed by a major MSO for their current Added DS Chans 75 42 23 mix of cable systems of various bandwidths, conditions and original parentage. Added to This upgrade, of course, does nothing to these costs are estimates of the replacement address the upstream issue. optical equipment required at headend or hub and node to feed the upgraded plant. UPGRADE TO 1024 QAM

Consumer Premises Equipment (CPE) and The highest existing digital modulation is Regulatory Issues 256 QAM, which transmits 8 bits of information per symbol, for an effective The entire cost is not in the distribution transmission rate of about 38 Mb/s in a 6- system upgrade – the bandwidth must be used MHz RF channel. One proposal for for something. No existing CPE tunes above increasing information capacity is to use the 870 MHz, nor is it required to do so to meet next logical increment of modulation density, current DOCSIS (data) or SCTE 40 (video) 1024 QAM, to increase the bandwidth efficiency of networks by transmitting 10 bits slightly, we estimated a cost of $850/mile in per symbol, a theoretical increase of 25%. distribution system “fixes”.

The practical network issue with this Headend Costs upgrade is existing network noise and distortion performance. SCTE 40 mandates end-of-line C/(noise + interference) of 33 dB Existing headend modulators must be for 256 QAM.viii To maintain the same replaced to prepare the system to utilize the headroom, a 1024 QAM signals would need expanded throughput. To minimize the cost, to be received with a C/(noise+interference) we assumed that only digital video of 39 dB. modulators are replaced (leaving data and VoIP unchanged). Additionally, we added the In most cable systems, data signals are cost of re-multiplexers for those signals carried at the same average power level as currently received from satellite and passed analog video signals (typically referred to as 6 through the headend unchanged. This dB lower only because analog video signals prepares the system for adding 2-3 additional are referenced to sync peak level and digital video streams per multiplex in the future. signals to average power level). Thus, raising the power level of 1024 QAM signals is Customer Premises Equipment Costs and probably not a practical option. Regulatory Issues

Typical cable systems are designed for an No existing CPE is capable of receiving end-of-line ideal analog video C/N (thermal 1024 QAM signals. Furthermore, current noise only) of 48 dB. With normal variations, FCC rules mandate that one-way digital video aging and maintenance tolerance, 46 dB is services use only 64 QAM or 256 QAM. about all that can practically be assured – just Although operators may approach the enough to pass the FCC’s 43 dB requirement technical issue in various ways, we assumed after passing through a typical converter (with that existing converters would be replaced 0 dBmV input and a 13 dB noise figure). with hybrid analog/digital converters Taking into account the difference in noise enhanced to receive 1024 QAM that would susceptibility bandwidth between video (4 cost $175, with a value of $25 assigned to the MHz) and data (5.3 MHz) and the 6 dB retrieved converters they replace. We have difference in how their levels are referenced, assumed that all existing converters are the expected carrier-to-thermal-noise of a replaced, which enables the efficiency received data signal may be as low as 38.8 improvement to be applied across all digital dB, to which must be added the effects of video channels, but means that converter composite beat products among analog replacement dominates the other costs. signals, composite intermodulation products among digital signals and crosstalk in multi- Using these assumptions, the total cost of wavelength optical links.ix Otherwise stated, an upgrade to 1024 QAM is $60 per home a system that just meets FCC specifications passed for an effective downstream bandwidth for analog video will not be adequate to carry increase of 6.75 6-MHz RF channels. As with 1024 QAM signals. the 1 GHz upgrade, converting to 1024 QAM does not address the upstream bandwidth Distribution Network Costs constraint, nor provide a path for future upgrades. Unlike, a 1 GHz upgrade, our 1024 To account for solving the inevitable QAM scenario is in conflict with current FCC system problems and increasing performance regulations, however applying it to only interactive services would greatly reduce the In summary, an upgrade to advanced video throughput gain. encoding is less expensive than an upgrade to 1024 QAM because no plant changes are ADVANCED VIDEO COMPRESSION required, and results in a larger effective capacity increase. Specifically, the estimated All cable digital video services today are cost, using our assumptions, is $53 per home compressed using MPEG-2. While this was a passed and results in an effective bandwidth breakthrough technology when introduced, increase of 12.2 downstream RF channels. It more efficient algorithms have since been does not address the upstream bottleneck. introduced, of which the dominant contenders are MPEG-4 AVC and Windows Media NODE SUBDIVISION (SMPTE VC-1). Either offers roughly a 2:1 increase in streams/channel compared with Subdividing optical node serving areas MPEG-2. Since a large use of downstream does not increase the instantaneous system bandwidth in a typical cable system is for information capacity to any network segment, digital video, adopting a more efficient but does share that capacity among fewer compression algorithm will increase overall customers. Thus, to the extent that the sub- effective throughput. areas are fed separately, an effective capacity

increase is realized for those services which Because no new modulation is involved, are delivered to individual customers. To be the upgrade imposes no increased demand on precise, the capacity is increased in proportion the distribution network. to the bandwidth allocated to those services and multiplied by the number of downstream Headend Costs or upstream segments created. Furthermore, since nothing is changed except for effective The cost of adopting advanced node size, no regulatory or CPE technical compression is dependent on how widely it is issues are created. adopted. We assumed that most digital video would arrive at the upconverted system in the Plant Costs new format, either from the original program source or from the MSO’s regional center. We assumed that the previous upgrade was not a total rebuild – that is, it utilized as much Customer Equipment Costs and Regulatory of the then-existing plant as possible -- and Issues that the cost was further minimized by “dropping” non-scaleable nodes into the The CPE situation for advanced encoding coaxial distribution system to create the is essentially the same as for use of 1024 required 500-home serving areas. Thus, the QAM – existing boxes do not receive and cost of node subdivision included the cost of cannot be upgraded to receive the new-format replacing the node itself with a segmented signals, and thus require replacement. model (2:1 downstream and 4:1 upstream) and re-routing the coaxial distribution plant to The regulatory issues are also similar, as create four roughly-equal-sized segments the FCC limits one-way digital services to (requiring, on average, 1,000 ft of new cable MPEG-2 encoding. As with 1024 QAM, we plus splicing). It does not include any calculated the efficiency gain across all digital service-specific hardware. video channels and did not address the regulatory issues. Headend Costs carriage of at least Basic channels in analog format absent individually-granted In order to activate the expanded exceptions, though that requirement will cease bandwidth, we included the cost of one when broadcasting transitions to digitalx. additional downstream transmitter and three additional upstream receivers to communicate We evaluated two versions of an all-digital with the new sub-nodes and thus activate the conversion – a downstream-only version and additional capacity. a further option in which a portion of the formerly-downstream bandwidth is allocated In summary, we estimated that the division to upstream usage. of 500-home nodes into two downstream segments and four upstream segments, would Plant Costs cost approximately $30 per home passed. We assumed that eight downstream channels were Since standard 256 QAM signals are used for individual subscriber, interactive assumed, analog channels are converted to services, resulting in a net effective bandwidth digital at approximately the same total RF gain of four channels in each of the two power per channel, and thus no additional downstream sub-nodes, equivalent to a loading is placed on the distribution system. doubling of the downstream interactive As discussed above, end-of-line digital signal service throughput capability. We assumed C/N should be slightly below 39 dB at worst, that 30 MHz of the upstream bandwidth was and thus have a significant margin above the usable for interactive services and therefore SCTE 40 and FCC minimum of 33 dB, even the 4:1 split creates an effective bandwidth when distortion parameters are included. gain 22.5 MHz in each of the four upstream Thus, no plant changes are required to make sub-nodes, equivalent to a quadrupling of the analog to video conversion in the upstream interactive service throughput downstream-only option. capability. Expanding the upstream bandwidth, CONVERSION TO ALL-DIGITAL however, requires changing every diplex filter in the system, the upstream amplifiers (wider In the future, all television, whether over- bandwidth and higher gain), upstream optical air broadcast, satellite or locally originated, transmitter modules in nodes, and optical will be in digital form. One option for receivers in the headend. We assumed that operators is to accelerate that process by the new upstream spectrum would extend converting all current analog video signals to from 10 to 85 MHz and that the downstream digital form and providing digital converters spectrum would start at 105 MHz to preserve at every connected television receiver. use of equipment that operates in or near the FM band. We estimated the total of plant and The advantages include at least a 10:1 optical headend cost to make the frequency increased usage of former-analog bandwidth, change to be $10,180 per 500 HP node, lower cost receivers, uniform transport including the cost to realign the plant. protocols across all services and breaking DBS operators claim to be the only “all Headend Processing Costs digital” network. Disadvantages include the cost of providing converters to current analog As with advanced compression techniques, subscribers and defeating the features of some we assumed that most (75%) of signals would basic subscriber’s video equipment. arrive at the headend in digital form from Additionally, current FCC regulations require broadcasters, cable networks or MSO regional centers, but that the remainder would require Switched digital video (SDV) gains conversion at the headend. We scaled effective network throughput by offering less Charter’s reported cost to upgrade their popular programs to service groups only on California systemxi by the required number of demand, using technology that is transparent locally-converted channels and estimated the to users – that is, the viewer should ideally be total headend cost to be $250,000. unaware when selecting a program that it might not be delivered until the virtual Consumer Premise Costs channel is selected. Use of SDV does not increase the information capacity of the We assumed the same $75 digital-only network, but rather shares it more efficiently. converter cost for this option as for the 1024 QAM case. The difference is that, rather than When a switched channel is selected, a replacing existing digital converters because small resident application in the user’s box of incompatibility, additional converters are sends a request to the headend SDV server, required for every television outlet in the which, if the requested stream is not already system what did not previously have one. being viewed in the service group, adds it to an appropriate multiplex. Then it directs the The cost of the downstream-only (“low- box to the correct channel and program split”) version and the version that includes identifier. When the channel is no longer expanding the upstream spectrum (“mid- being viewed within the group, the stream is split”) is summarized in the table below. The dropped.xii mid-split version more than doubles upstream capacity. Trials of SDV are still in an early stage, with widely varying results. One operator Option DS Only DS + US estimated a potential savings of 26% of total Cost/HP $157 $175 video channelsxiii, while a larger and more Added DS Chans 72 63 recent trial conducted in a Cox system Added US MHz 0 38 suggests that as many as 41 programs can share an RF channel on a switched basis and SWITCHED DIGITAL VIDEO that trials with 28 programs per channel xiv resulted in no instances of blocked access . With the exception of server-based on- demand programming, cable operators The regulatory problem with switched currently transmit all available programming video is that operators are required to deliver choices simultaneously and continuously all non-interactive digital video services in a throughout their networks. However, given way that is compatible with one-way digital the widely different popularity of different cable-ready receivers. Those receivers are programming among any given group of obviously not capable of sending message to subscribers, viewing is concentrated among a the headend to request streams. Thus, until few channels and many of the hundreds that hurdle is overcome, only interactive video offered are not simultaneously viewed. Thus, services can be offered on a switched basis. even though there are good reasons for offering a wide choice of programming, it is Given the state of development of SDV an inefficient use of bandwidth to send signals and regulatory constraints that limit which to sections of the network except when at channels can be offered on a switched basis, least one subscriber wishes to access them. we assumed that 100 current two-way digital program offerings (premium and pay-per- view) would be delivered over five statistically-shared RF channels – a savings of expansion. While various proposals for use of 50% over the spectrum formerly required. In this spectrum have been proposed for many other words, we assumed that an operator years, none have been widely deployed as an would choose in this option to comply with upgrade strategy. The system described current regulations. below, however, has been successfully used to implement selective overlays to service Headend Costs commercial customersxv, so the viability of the technology was not at question, but rather There are no distribution plant costs its applicability to a system-wide upgrade to associated with the addition of SDV, since the serve the entire customer base. distributed signals are identical to non- switched signals. In the headend, a SDV The version we evaluated is based on the manager is required to manage the addition creation of two sub-octave transmission bands and deletion of streams and an MPEG -- 1250-1950 MHz downstream and 2250- switch/mux is required to create the required 2750 MHz upstream -- with nodes and multiplexes feeding each service group. We amplification equipment paralleling legacy assumed that bandwidth, multiplexers and equipment in the field, as shown in Figure 1. modulators were dedicated to the SDV Splitting amplification between legacy and service, rather than being shared with on- extended amplifiers greatly simplifies demand services. The cost estimate assumes amplifier design. Passives are replaced by 4 nodes and 5 RF channels (80 streams total) equivalent units passing the entire 5-2750 per service group. MHz band. Tests have shown that current- generation hard cables used by the industry Customer Premise Costs will support these frequencies, while the first non-TEM mode does not occur until about 4.6 SDV is completely compatible with GHz for the largest cable sizes in use today.xvi current-generation digital two-way boxes. The required software module is much Triplexer US Amp. Triplexer smaller than that required to implement VOD. 2250- 2250- 2750 2750 Therefore, there is no cost to implement SDV DS Amp. In 1250- 1250- Out among existing digital video subscribers. 1950 1950 5- 5- With the above assumptions, 860 860 implementation of a limited SDV service to existing digital video customers is estimated to cost only $5175 per node, but to free up only the equivalent of 5 downstream RF Legacy Amp. channels. It has no effect, of course, on Figure 1: Extended Frequency Amplifier upstream congestion and, in fact, adds traffic In order to assure compatibility with from video set-top boxes for which low existing headend and CPE and to ensure latency is very important. Much greater gains compliance with FCC regulations, the are possible, of course, but only if the extended downstream frequency band is regulatory issues are resolved. converted to/from 100-800 MHz and, for residential applications, each upstream sub- EXTENDED BANDWIDTH (> 1 GHz) block of 12-42 MHz could be converted to one of ten “slots” in the 2250-2750 upstream A final choice is to activate the spectrum distribution band and back in the headend. above 1 GHz for bi-direction bandwidth This scheme would allow both headend RF Plant Costs equipment and consumer premises equipment to operate at normal levels and frequencies. The plant upgrade effort is comparable to a Since the entire 500 MHz upstream spectrum typical rebuild effort. Because the required is transported to the headend without any techniques to parallel active equipment are channelization, it is also possible to allocate a different, we based our estimated labor costs wider portion of the spectrum to applications on actual field trials and on evaluations by which require greater data rates than can be experienced contractors in cable construction. transported through a standard cable The material costs include powering upgrades television upstream path. Only a terminal required for the additional actives. As with equipment change would be required to re- node subdivision, we assumed that the allocate the spectrum for such applications. required two fibers (or at least one wavelength As with any block segment conversion on each of two fibers) between headend or scheme (such as those used for years for hub and node were available. upstream node segmentation) low phase noise, frequency accurate converters are Headend Costs required. The premise equipment diagram is shown in Figure 2. Headend costs include the equipment required to activate the additional bandwidth, Triplexer Diplexer including frequency conversion and optical 2250- Tunable Up 10- transmitters and receivers. 2750 Converter 40

Drop 1250- Down 100- 1950 Converter 800 Out Premise Costs 5- 860 As with other options, the cost of activating frequencies above 1 GHz is Figure 2: Premises Terminal dependent on how those frequencies are utilized. For purposes of this study, we In addition to a 2:1 downstream and up to assumed that analog video, cable modem and 17:1 upstream bandwidth increase (11:1 if all VoIP signals remained on the legacy channelized in 30-MHz sub-bands), a bandwidth, while digital video and a digital significant advantage of this scheme is that simulcast of analog video signals were placed the expanded upstream communications on the expanded downstream band, with STB capacity is not “locked” to a sub-node group upstream signals on one of the expanded (as in node subdivision), but rather can be upstream slots. This scenario is compatible assigned on a subscriber-by-subscriber basis with all existing digital and analog equipment, anywhere in the node serving area, simply by including subscriber-owned modems and assigning which slot upstream frequencies are television receivers, while allowing the cable converted to. A second significant advantage operator to purchase digital-only STBs going is that the ingress into any upstream slot is forward and freeing bandwidth for advanced limited to that occurring in the residences (or video and data services. The model included businesses) whose upstream communications the cost of installing residential block are converted to that slot, thus improving the converters for every customer who subscribes usability of the added spectrum. Ingress in to digital video services. We estimated the the drop has no effect on the extended labor cost of this to be comparable to spectrum. installing a standard drop amplifier.

We evaluated the cost of such an upgrade Instantaneous vs Virtual Capacity Increases under two scenarios -- activation of two or eight of the ten possible upstream slots – in Some evaluated technologies increase the order to determine how sensitive the peak information-carrying capacity of the technology is to the degree of upstream network, while others realize the effective bandwidth expansion. The results are throughput increase by other means. Both are summarized in the following table. important: Peak capacity limits the amount of information that can be transmitted to any Option 2 Blocks 8 Blocks given subscriber group, while virtual capacity Cost/HP $121 $129 increases are dependent on how services are Added DS Chans 117 117 divided between those which are broadcast Added US MHz 60 240 and those which are directed to specific customers or customer groups. Today, peak In summary, the use of frequencies above upstream capacity, using 16 QAM, is limited 1 GHz for expansion of both down and to about 100 Mb/s (ten 3.2 Mb/s channels, upstream bandwidth offers the greatest each with a capacity of 10 Mb/s). Even if 64 information capacity of any of the evaluated QAM were usable across the entire 9-41 MHz options, with the possibility of further band, the potential increase would only be expansion of the upstream at very low cost. about 25% to 125 Mb/s.

SUMMARY All the evaluated technologies increase downstream effective capacity. The Quantitative comparisons will depend on following table shows which also increase assumptions and intended use the expanded peak downstream information rates and which capacity. This summary is based on the increase upstream effective and/or peak rates. assumptions stated previously. Technology DS Upstream Regulatory Issues Peak Peak Virtual 1 GHz Yes No No The use of frequencies above 862 MHz, 1024 QAM Yes No No 1024 QAM or advanced encoding for digital AVC* Yes No No video all violate provisions of Paragraph Node Split No No Yes 76.640 of the FCC’s rules, if applied to one- All digital Yes No No way digital video services. Until and unless + US expand Yes Yes Yes those provisions are modified, the gain from Switched No No No use of these techniques will be constrained Extended BW Yes Yes Yes because of that. Our summary results take *Advanced video compression those restrictions into account. Only the elimination of analog video Secondly, the FCC requires that basic combined with expansion of the upstream television service be carried in an analog band or the use of two-way extended form. Thus, the conversion to all-digital bandwidths provides an increase in both video is dependent on obtaining a waiver or upstream and downstream effective and waiting until all VHF over-air transmission instantaneous information rates. ceases.

Comparisons of Capacity and Cost throughput gains for a couple of reasons. 1024QAM, which carries 10 bits per symbol, Figure 3 illustrates the increase in effective only offers a theoretical 25% gain over 256 downstream channels for each of the QAM. Additionally, all three technologies technologies, while Figure 4 shows the are currently constrained the FCC regulations increase in effective upstream bandwidth. and which are derived from SCTE40. Of the Figure 5 shows the cost effectiveness of each, three, SDV is the most efficient because it is which we calculated by taking the ratio of compatible with existing set-tops. Absent per-node capital cost to the total downstream- regulatory restrictions, SDV holds the plus upstream effective bandwidth increase. promise for major downstream effective throughput gain. Assuming our assumptions are reasonable, it appears that the most efficient 1 GHz Splitting of existing nodes offers upgrade is from a 750 MHz system. While a significant gains in effective downstream and 550 MHz system will gain more DS upstream bandwidth for moderate cost and bandwidth, it will also require much more without causing any regulatory problems or cable, passive and drop replacement work. equipment compatibility issues. It is second On the other hand, while an 860 to 1 GHz only to extended bandwidth in cost efficiency. upgrade is the least costly of the three, the lower incremental bandwidth makes it less Use of extended bandwidths, as described efficient. earlier, is comparable in cost to a 1 GHz upgrade and less expensive than an all-digital As expected, converting to all-digital video conversion. It offers the greatest incremental gains a lot of DS bandwidth due to the 10:1 bandwidth improvement -- effective and improvement in program streams per channel. instantaneous; upstream as well as Looking at Figure 5, however, it is not one of downstream -- of any of the options. As a the most capital-efficient upgrades simply result its cost effectiveness is greater than any because of the cost of placing one or more of the alternatives. Furthermore, the digital converters in every Basic subscriber’s incremental cost to activate 8 upstream blocks house. Converting to a mid-split is slight compared with activating just two, so configuration is slightly less cost-efficient but that it is very economically scalable to future is one of only three options to improve the expansion needs. We suggest that it should be critical upstream throughput bottleneck. seriously considered for future major throughput upgrades. 1024 QAM, advanced video compression and switched video offer only moderate

250

200

150

100 Equivalent Bandwidth Gain (MHz) Gain Bandwidth Equivalent

50

0 550 MHz 750 MHz 860 MHz 1024QAM AVC Node Split Low Split Mid Split Switched 2 US 8 US Increase DS Limit to 1 GHz From Eliminate Analog Video Video Extended BW

Figure 4: Effective Upstream Bandwidth Gain

800

700

600

500

400

Dollars/Node/MHz 300

200

100

0 550 MHz 750 MHz 860 MHz 1024QAM AVC Node Split Low Split Mid Split Switched 2 US 8 US Increase DS Limit to 1 GHz From Eliminate Analog Video Video Extended BW

Figure 5: Cost-Benefit Ratio REFERENCES Broadcasting and Cable magazines, September 2004. 1. Report on SCTE’s 2005 Emerging 12. “Coneco offers all-digital option,” CED Technology conference by Laura Broadband Direct, 9/20/04. Also, Hamilton and Johathan Tombes in “Bargain boxes, is the cable industry Pipeline. within reach of the $50 all-digital set- 2. “Verizon idenfifies more ‘FiOS’ markets,” top,” CED Magazine, September 2004. CED Magazine online, 1/19/05. 13. SCTE 40 2004 Digital Cable Network 3. “Linksys adds ADSL2+ support,” Interface Standard, Table B. (describing 25 Mb/s technology) Jeff 14.See Chapter 13 of Modern Cable Baumbartner, CED Broadband Direct, Television Technology, Second Edition, 2/3/05. by Ciciora, Farmer, Large and Adams for 4. “Comcast raises broadband speed,” Jim a fuller explanation of fiber crosstalk Hu, c/net News.com. effects. 5. “RCN pushes cable-modem speed to 10 15. “Congressional leaders want ‘hard date’ mbps,” CTAM SmartBrief, 1/20/05. end for analog broadcasts,” (suggesting 6. “What’s ahead for Vonage in 2005?” 2006 deadline) CTAM SmartBrief 2/3/05 Net.Worker column in Network World, 16. See footnote 8. 1/10/05. 17. “Getting from here to anything, 7. “Satellite Technology and Platforms,” anytime,” David Large, CED magazine, Steven Osman, presented at SCTE’s September, 2004. This article explores Emerging Technologies 2005 conference. and advantages and limitations of SDV. 18. “IP-based switched broadcast,” Joachim Vanhaecke, Proceedings Manual, 2003 8. “Cable could rule if it plays it cards right,” SCTE Cable-Tec Expo. David Lieberman, USA Today.com, 19. “The statistics of switched broadcast,” 1/24/05 Nishith Sinha and Ran Oz, Proceedings 9. “Comcast puts video IM on deck,” Jeff of SCTE 2005 Conference on Emerging Baumgartner, CED Broadband Direct, Technologies. 1/20/05. 20. “Arris, Xtend test out-of-band DOCSIS,” 10.“RCN offers Net-connected surveillance,” Jeff Baumbartner, CED Broadband Boston Globe, January 27, 2005. 1/31/05 11. “Convergence, California Style,” IP 21. Modern Cable Television Technology, Solutions Summit, Supplement to 2nd edition, equation 10.11 (which is in MultiChannel News, CED and error by a factor of 2).

BANDWIDTH MANAGEMENT FOR THE UNIVERSAL EDGE

By Bruce Thompson, Xiaomei Liu Cisco Systems, Inc.

Abstract deployment scenario where QAMs are dedicated to each service. For VOD services, To offer more services to end users in the VOD servers send content over gigabit cable network, bandwidth needs at the edge of Ethernet (GE) to video capable downstream HFC network are growing rapidly. The (DS) QAMs which reach Setop boxes (STB) industry is moving from current architecture in the home. VOD servers are in control of where each service has its own edge allocating both video pumps and QAMs when resources to a multi-service universal edge VoD session requests are made by STBs. architecture to use the RF bandwidth more efficiently. MPEGoIP

This paper will go into detail on the benefits of dynamically sharing resources across services. It will also describe a VOD GE Server QAMs STBs control/data plane architecture that can be used to share resources between these services. Lastly, this paper will provide examples of standardized protocol suites that CMTS can be used to implement the interface IP Aggregation DS between the components of a distributed M A C Network US architecture that supports resource sharing CMs across services.

Figure 1. VOD and DOCSIS services in OVERVIEW today’s cable network

The HFC plant is the point of convergence For DOCSIS Internet Access services, a for all of the services that MSOs provide. In CMTS serves a group of cable modems. The addition to Docsis based Internet Access and CMTS consists of a Docsis MAC layer Broadcast Video, new services such as video processor, downstream (DS QAM) channels, on demand (VoD), network PVR and and upstream (US) channels bundled into a switched broadcast are now being offered. In single platform. The downstream QAMs current deployments, each service has a embedded in the CMTS use a different statically allocated portion of the RF portion of the RF spectrum than the QAMs spectrum. Dedicated QAM pools are allocated dedicated to the VoD service. for broadcast, VOD, switched broadcast and DOCSIS based Internet Access services. With the increasing popularity of VoD services, high definition television, and Using VOD and DOCSIS Internet Access DOCSIS Internet Access, the demand for services as an example, Figure 1 shows the HFC network bandwidth is ever increasing. MSOs must use the bandwidth of the existing of the modular CMTS are connected using HFC infrastructure more efficiently to avoid Gigabit Ethernet. This architecture makes it having to upgrade plants as the need for possible for the downstream QAM to accept bandwidth increases. both VOD and DOCSIS traffic. In later section of this paper, more details will be The demand for bandwidth has motivated given on the data plane and control plane MSOs to find ways to improve the efficiency architecture which has the promise of sharing of bandwidth utilization. The downstream for QAM resources. DOCSIS and other video services use QAM modulated MPEG-2 transport streams to carry the data. Because of the common transport Cat45PHYxx SUP VI D D D D encapsulation and modulation technique, a S S S S single set of MPEG-2 based QAM devices GE and associated RF bandwidth can potentially VOD be shared across all of these services for more Server RF Switch efficient bandwidth utilization. In later section CMTS of this paper, quantative analysis will be given Cat45PHYxx SUP VI M A C to show the potential saving by sharing U U U U S S S S QAMs. IP/MPLS M A C Core M A C

However, the current network architecture M A C show in Figure 1 makes the QAM resource RF Switch sharing impossible. First, each service is responsible for managing the QAMs Figure 2. New architecture with universal dedicated to that service. Thus it is not edge QAM possible to dynamically share HFC bandwidth between services. Secondly, the DOCSIS BENEFIT OF UNIVERSAL EDGE CMTS bundles both upstream and downstream together in the single logical How much RF bandwidth associated device. This makes it difficult to share QAMs can be saved when multiple service downstream RF resource between DOCSIS share the universal edge QAM? This section and video services. will do quantative analysis to show the potential significant savings. We will use A new architecture is now evolving which switched broadcast, VOD and DOCSIS data addresses the problems of the existing services as examples. architecture. Figure 2 shows the new architecture that is capable of dynamically The first factor that allows bandwidth sharing downstream QAMs between VOD savings is the fact that the busy hour and DOCSIS services. In this new associated with each of these services may not architecture, the downstream QAM is capable be at exactly at the same time or day. For of both video and DOCSIS processing. In example, while the busy hour associated with addition, the function of the DOCSIS CMTS broadcast and on demand services are is now broken into 3 separate components. typically during the broadcast network prime They are the Docsis MAC processor, an time period (8:00 PM to 10:00 PM), the busy upstream QAM, and a downstream QAM. hour associated with internet access services These 3 components together are called the may occur later in the evening after children modular CMTS (M-CMTS). The components have gone to bed. The second factor that allows bandwidth it is with telephony. Another factor that makes savings is the savings associated with the the Erlang calculation applicable to video is semi random behavior of individual that the calculation is independent of call hold subscribers and the savings that can be time. In telephony, the call hold time is the obtained by taking advantage of the amount of time an average call lasts. It is probability distribution of the behavior of a typically a couple minutes. In VoD, the population of subscribers across services. The equivalent of call hold time is the amount of telephone industry has long used probabilistic time the typical subscriber spends watching a models based on subscriber behavior to movie. While this time is likely much longer determine how much bandwidth needs to be for video than for telephony, the Erlang deployed to allow a group of subscribers to calculation is independent of this factor. gain access to the network with a high probability of obtaining network service. A Since Erlang B calculators perform well known model for this type of calculations in the context of telephony telecommunication traffic design and analysis requirement, the variables of a typical Erlang is called the Erlang model. B calculator must be translated into appropriate units relevant to other services Multi Service Erlang Analysis such as video and Internet Access. Erlang B calculators are typically used for call center From mathematical point of view, Erlang analysis and are readily available from many model has provided further evidence that the sources including the Internet. An Erlang B second factor mentioned above achieves calculator has 3 variables associated with it. bandwidth savings. The Erlang model shows The calculator typically allows the user to that the efficiency level of a resource such as specify 2 of the variables and it calculates the telephony trunks or RF bandwidth in a cable third variable. plant increases as the number of subscribers that share that resource goes up. When RF Erlang B Analysis for Video Services resources are shared across multiple services for a given population of users, the effect is to The 3 variables in an Erlang calculator are: essentially increase the number of subscribers busy hour traffic (or Erlangs), blocking factor, that are sharing that RF bandwidth. This more and capacity measured in number of lines. efficient usage of the RF bandwidth allows Busy hour traffic (BHT) is the number of less RF spectrum to be allocated to the hours of call traffic during the busiest hour of combined set of services that would be the operation in the system. For VoD services, case if resources were not shared. this can be determined by multiplying the number of homes in a service group, the While most Erlang calculators are specific percentage of homes subscribed to the service, to telephony, the calculation itself is and the engineered peak usage rate for the applicable to any form of service where the service. The blocking factor for VoD services arrival rate of new requests during the period specifies the percentage of time that VoD over which the calculation is run (the busy requests will be allowed to fail due to lack of hour) can be assumed to be random. Since the QAM bandwidth. Note that the blocking exact timing of how subscribers make VoD factor for VoD is usually specified to be very requests is not synchronized to external events low since it is undesirable to disallow service (such as the advertised beginning of a to a subscriber. The number of lines is the television show) the arrival rate for VoD value that we are solving for in the Erlang requests can be assumed to be random just as calculations shown in this paper. For telephony, this is the number of telephone bandwidth can be considered the blocking lines that must be installed to support the factor for the Erlang calculation. The blocking specified traffic at the given blocking rate. For factor for Internet Access can be quite high video, the number of lines can be translated to since the effect of a “blocked” user is that his the number of video streams that you need Internet Access service appears slower than QAM bandwidth for. To turn video streams the minimum rate. Another factor that must be into a bandwidth value, we assume that each taken into account is that when a subscriber is video stream requires 3.75 Mbps of using their Internet connection, they are not bandwidth. To determine the number of always making requests that require QAMs required, we then divide the resulting bandwidth. We must take this factor into bandwidth by the bandwidth per QAM (38 account when calculating BHT. We call this Mbps) and round to the next higher integer. factor the Internet usage factor. The Erlang calculations for Internet Access in this paper Given the above factors, the full formula will use an Internet usage factor of 20% or .2. for determine the number of QAMs required Given the above factors, the full formula for in a service group for VoD services is: determine the number of QAMs required in a service group for Internet Access services is: BStream = BW per VoD Stream = 3.75 Mbps BQAM = BW per QAM = 38 Mbps IUsage= Internet Usage Factor = .2 Homes = homes per service group BSub = BW per Internet Subscriber SR = Subscription Rate PR = Peak Usage Rate BHT = Homes * SR * PR * IUsage BF = Blocking Factor # of QAMs = BHT = Homes * SR * PR roundup( ErlangB( BHT, BF) * BSub / BQAM), 1) # of QAMs = roundup( Multi Service Erlang B Example ErlangB( BHT, BF) * BStream / BQAM), 1) In the following example, we apply the Note that the above Erlang analysis can also Erlang analysis to a cable plant with service be used for switched broadcast services. usage patterns that are typical in today’s network. For VoD services, the example Erlang B Analysis for Internet Access shows with 500 homes per service group, a 20% subscription rate, a 10% peak usage rate, Erlang B analysis can also be used to a 0.001% blocking factor and peak usage time model traffic associated with an Internet of 8:00PM. Access service. While the traffic patterns associated with Internet Access are different For the internet access service, we assume than telephony or video, you can still model 2000 homes per service group, a 30% the Internet Access service as one where subscription rate, a 20% peak usage rate, a subscribers are randomly making requests and blocking factor of 1% and a peak usage time the service provider is trying to provide a user of 10:00PM. Finally, to calculate the number experience where the subscriber gets a of QAMs needed for Internet Access we will minimum bandwidth for a certain percentage assume a minimum rate per subscriber of 1 of the time. The percentage of time that the Mbps. subscriber does not get this minimum The non-peak hour usage rate is assumed service group, a 30% subscription rate, a 20% to be half of the peak usage rate for each peak usage rate, a blocking factor of 1% and a service. peak usage time of 10:00PM. From table 2, it is clear that savings can be achieved if the Table 1 shows the RF bandwidth peak usage times for Docsis and VoD services requirement and QAM resources needed per are not the same. 2000 subscribers if these services use statically allocated QAMs. The RF bandwidth Table 2. Example future RF bandwidth calculation must take into account the sum of requirement without resource sharing the peaks of each service. The calculations were done using the Erlang analysis described Service Usage Blocking BHT BW QAM above. (%) (%) (hour (mbps) ) DOCSIS 20 1 6 52 2 Table 1. Current RF bandwidth requirement VOD 10 0.001 10 101.25 3 without resource sharing Total 5

Service Usage Blocking BHT BW QAM (%) (%) (hour (mbps) If QAMs are dynamically allocated ) between Docsis and VoD, the combined DOCSIS 20 1 24 35 1 service group can be provisioned for each VOD 10 0.001 10 101.25 12 service peak independently. Dynamic Total 13 allocation will ensure that the correct number of QAMs is allocated to each service is it Note from Table 1 that the amount of reaches its peak usage. QAMs required for VoD and broadcast are much greater than those needed for Internet Table 3 and Table 4 shows the bandwidth Access services. Because of this, dynamic savings that can be obtained in the using the resource sharing does not provide much usage data above if RF bandwidth is benefit with this type of usage pattern. dynamically allocated to each service.

Table 2 shows a likely future usage pattern Table 3 shows the RF bandwidth that will become common as the need for requirement at 8:00 PM when the VoD Docsis bandwidth grows. The basic service is running at its peak rate while the assumption here is that the amount of DOCSIS service are running at its non-peak bandwidth that the MSO sells the subscriber hour rate. for Internet Access service will increase from 1Mbps to 4 Mbps. An example change that Table 3. RF bandwidth requirement at 8:00 will drive the need for higher Docsis PM with dynamic resource allocation bandwidth is the evolution of Web based Video over IP to higher screen resolutions. In Service Usage Blocking BHT BW QAM this scenario, the video usage is the same as in (%) (%) (hour) (mbps) Table 1, but the following Docsis usage DOCSIS 10 1 3 32 1 patterns apply. The increased use of Docsis VOD 10 .0001 10 101.25 3 bandwidth will drive down the size of the Total 4 serving group for Docsis to be identical to that of VoD. In this future example, we assume an Internet Access service with 500 homes per Table 4 shows the RF bandwidth control plane architecture that can be used for requirement at 10:00 PM when the Docsis dynamic resource sharing. service is running at its peak rate while the VoD service is running at its non-peak hour Data plane architecture rate. There are multiple ways to achieve the Table 4. RF bandwidth requirement at 10:00 resource sharing among different services. PM with dynamic resource allocation Figure 3 shows an example data plane architecture. In this architecture, a VOD Service Usage Blocking BHT BW QAM server, a real time broadcast encoder, a CMTS (%) (%) (hour) (mbps) core, Downstream QAMs, and Upstream DOCSIS 20 1 6 52 2 QAMs are all inter-connected through a VOD 5 .0001 5 105 2 Gigabit Ethernet network. The GE can switch Total 142.5 4 traffic from any components to any components. With this architecture, DS QAM From the above tables, we can see that resources are shared among all three services. dynamic resource sharing between VoD and In other words, DS QAM is capable of both DOCSIS services requires 4 vs. 5 QAMs to be video processing and DOCSIS data deployed in the serving group which results in processing. a 20% reduction of plant bandwidth used by Video these services. The 20% saving comes from Encoder the difference in peak hours between the different services. QAM STB Gigabit VOD Ethernet QAM CM While not shown in the above example, Servers Network additional savings can be obtained by sharing Upstream the same service group with switched broadcast services. In this case, additional CMTS Core savings can be obtained by using a single

QAM pool for both VoD and Switched Figure 3. Data plane architecture Broadcast services. These additional savings occur because the number of effective users This architecture is highly scalable. As sharing the same pool of QAM resources is more services are added, only the server increased. The savings is essentially due to related to the service needs to be connected to the law of large numbers which is what is the Gigabit Ethernet. If the QAM bandwidth represented through Erlang analysis. needs to be increased, additional QAM resources can be shared among all existing DATA AND CONTROL ARCHITECTURE services.

The clear separation of data plane and Control plane architecture control plane components makes it easier to put a common resource manager to manage To achieve resource sharing, a common the resources associated with multiple logical resource management unit needs to services. In this section, a data plane exist to coordinate the resource allocation of architecture that can be used for resource different services. A component called the sharing will be discussed first followed by a session manager is then responsible for determining the classes of resources required The control plane architecture provides for a session request and communicating with two additional functions to the system. The the resource managers responsible for first is QAM discovery while the second is allocating those resources.Figure 4 shows an dynamic QAM allocation. example control plane architecture that can be used for dynamic QAM allocation. In Figure Service discovery protocol allows the Edge 4, a control component called edge resource Resource Manager to dynamically detect manager is introduced. Edge resource when a QAM comes in or goes out of service. manager is responsible for monitoring the DS When a new QAM is added or taken out of QAM resource and allocating QAMs for each service, the resource manage will be notified service. immediately about the resource change. The edge resource manager also maintains a Switched database maintains the mapping of QAMs to Broadcast VOD Session Session service groups. Manager Manager The second function added is dynamic QAM STB Edge resource allocation signaling. Each session Resource Gig E Switch QAM CM manager signals to the Edge Resource Manager Manager to allocate or deallocate QAM Upstream bandwidth. The Edge Resource Manager the

CMTS returns information of allocated QAMs to Core each session manager.

Figure 4. Control plane architecture This control plane architecture introduces several benefits for the system. First, it The control QAM resources the edge simplifies provisioning and management, resource manager must communicate with which in turn reduces the operational expense. session plane components from each service. In addition, it can improve availability by In Figure 4, these components are the VOD dynamically reallocating QAMs when a QAM session manager, and the switched broadcast failure is detected. Finally, the separation of session manager. The VoD session managers session management and resource is responsible for accepting user requests from management make it possible to dynamically Set Tops for VoD sessions while the Switched allocate QAM bandwidth across services. Broadcast Session Manager is responsible for This provides for more efficient use of accepting channel change requests from Set existing HFC plant bandwidth. Tops. Each of these session managers request bandwidth from the edge resource manager as CONTROL PROTOCOLS part of the process of instantiating a session. As mentioned in previous section, the The CMTS core is responsible for the control plane supports both service discovery managing a DOCSIS mac domain. It will and session signaling. Based on the different request QAM bandwidth from the edge requirements for these two functionalities, resource manager as part of the process of different control protocols can be selected. setting up or modifying the bandwidth associated with a Docsis MAC domain. For service discovery, RFC 3219 (TRIP) can be used with minor modifications to suit the needs of cable networks. TRIP is Telephony Routing over IP protocol which available, the resource manager will notify deals the problem of translating telephone the session manager about the QAM that was numbers into session signaling address of a selected for the session. If no resources are telephony gateway in VOIP system. When available to satisfy this session request, the modified for an HFC plant, TRIP allows a session manager will get an RTSP response QAM to dynamically announce properties with an indication of why the request failed. about itself to an edge resource manager. These properties include attributes such as the CONCLUSION frequency the QAM has been configured for, the HFC service group the QAM is connected This paper describes the trend in the cable to, the amount of bandwidth that is available industry to move to a distributed architecture for the edge resource manager to allocate where RF resources for different services can from, etc. be shared. From the quantative analysis of this paper, it is clear that resource savings can be Dynamic resource signaling could be achieved by dynamically sharing resources implemented with a protocol such as RTSP. among different services. RTSP is an HTTP based client / server protocol that provide a simple state machine This paper further describes a possible that can be used for resource allocation. RTSP architecture to achieve resource sharing and can be used by a session manager to request related control plane supports. At the time of qam bandwidth from the edge resource writing this paper, the cable industry is manager. The request for bandwidth is actively working on standardizing this encoded in an RTSP Setup message. It architecture and related data / control plane includes information such as the amount for interfaces. bandwidth required for the session / service and the HFC serving group that the bandwidth REFERENCES needs to be allocated from. 1. Erlang calculator: After getting the SETUP request, edge www.erlang.com/calculator resource manager uses its QAM selection 2. Telephony Routing Over IP (TRIP), RFC algorithm to search for a best QAM to use for 3219, IETF, Jan 2002 this session request. If QAM resources are 3. Real Time Streaming Protocol (RTSP), RFC 2326, IETF, April, 1998

CABLE IP VIDEO DISTRIBUTION

Timothy O'Keefe, Robert Sayko and C.J. Liu; Edited by Sean Carolan AT&T Labs

Abstract demands for quality video distribution, and will explore how these technologies can be Advances in IP technology are used to move nationwide video content providing Cable Operators with the distribution to a national IP-based network. opportunity to offer innovative interactive services while sharing transmission capacity WHY DISTRIBUTE VIDEO OVER IP between IP and traditional video services. NETWORKS? This paper will examine how QoS, MPLS, VPN, and Multicast standards for IP When cable operators offered only networks fit against the demands for quality traditional video service, satellite access video distribution, and will explore how linked their local distribution network access these technologies can be used to move to nationally distributed video content. New nationwide video content distribution to a Internet Access and VoIP services are national IP-based network. fundamentally different than traditional video service; they are more effectively delivered using terrestrially based IP networks, rather than latency-encumbered INTRODUCTION satellite delivery. Because Internet Access and Voice services send traffic throughout Cable operators are connecting their the world, operators need to connect their IP local networks with IP services to facilitate networks to the national and worldwide Internet Access and Voice over IP services. networks that form the greater Internet. Within their local networks, operators are Cable operators are servicing this need by using Ethernet to distribute VoD streams. connecting their headends to national The Next Generation Network Architecture terrestrially based IP networks to deliver (NGNA) is introducing new applications for these services. IP technologies in the delivery of services. NGNA proposes the delivery of a greater For IP networks, scale offers significant volume of video services, and the routing of cost advantages. A fiber connection into a video service control signaling, over IP national network can expand from 100Mbps based connections. to 10 Gbps just by upgrading the equipment used at the connection's endpoints. While Advances in IP technology are enabling the bandwidth can increase 100 fold, the these new services, and we envision more cost difference between the 100 Mbps innovative services and service integration electronics and the 10Gbps electronics is using IP capabilities for the future. relatively small. A company could Implementations of QoS, MPLS, VPNs and effectively reduce the unit cost of their new codec schemes are enabling video Internet Access and Voice services for a services over IP. This paper will examine given headend by expanding their existing how QoS, MPLS, VPN, and Multicast IP network capacity, and by using the standards for IP networks fit against the expanded capacity to receive video services REQUIREMENTS FOR A QUALITY IP as well. DISTRIBUTION OF VIDEO CONTENT

The fiber medium provides much easier Video Delivery Requirements capacity expansion than a satellite-based system. Fiber path construction typically The requirements for distribution of installs a bundle of fibers. As the capacity Video-over-IP are guided by two realities. of an individual fiber fills, lighting another The first of these are the technical fiber in the bundle can easily activate new capabilities that must be achieved to present capacity. Dense Wave Division an acceptable video experience. The second Multiplexing (DWDM) also enables each are the business drivers that then impose single fiber to multiply its capacity. This additional technical requirements. easy access to capacity creates the opportunity to apply less compression of the At the present time, formal technical digital video during transmission, thus standards that support requirements of real receiving higher quality video . time Video-over-IP, as promoted through a recognized standards organization such as Additional efficiencies are gained by IETF and compliant with MPEG standards using a common technology across all for video encoding, are being developed services. When Voice, Video and Internet based on needs that are still emerging from Access are combined into the IP network, a the industry. We recognize the IP transport common set of IP network equipment must facilitate transmission of the MPEG (routers, switches, etc) can be used across all stream such that the technical requirements the services. Technical engineering and that apply at the endpoint are met. The goal operations staffs then have fewer is to guarantee a delivery across the national technologies to master. An Operations IP network that will support MPEG center can use common IP based network requirements for video performance at the monitoring and management tools. The endpoint, and will also offer a guarantee of common use of IP technology across all the reliability and availability as it relates to services creates efficiencies throughout the uninterrupted service. business operation. The standard measures for video quality Video content suppliers are now more likely include the following: to have access to IP networks. A well- established national IP network likely passes • Inter-packet jitter as close to content producers as it does to • Packet loss content distributors. The content suppliers • Packet arrival order can input their content to the IP network at • Availability any point along the path. Just as the cable operator’s IP connection links their Internet Latency, while critical for real-time Access and VoIP customers to the world, interactive applications like telephony, is not their IP network connection can link them to as critical for video distribution. Latency of the world for access to new, interesting a well run terrestrial IP network will be content. substantially less than that of satellite television. Inter-packet jitter - the variance in latency from packet to packet – can, for any reasonable jitter that is expected in an IP significant difference in cost of network to network, be accommodated by buffering of design, build, and operate. Given the option IP packets at the receiver. Packet arrival to make the availability/distribution-price order can also be overcome through decision on a per channel basis, a cable buffering and packet reordering in the IP operator might elect to receive some endpoints. channels on the lower cost connection and others on the higher cost connection. If Packet loss is a critical measure of satellite is a third, more reliable service performance for IP video distribution. delivery, the cable operator might elect to While transport protocols, like TCP, can move some lower value channels from request retransmission of lost packets, this is satellite deliver to IP network delivery in not practical in a video distribution order to create satellite bandwidth for more application where a single source is sending high value content. a multicast stream of packets to a large number of endpoints. The stream must While no ubiquitous industry standard support transmission, and associated IP currently exists for this, it is expected that a overhead, of up to 3.75Mb/s of MPEG2 data video delivery system should employ an for a Standard Definition ATSC architecture that would meet very high level transmission; 19.3Mb/s for a High of availability. While a 99.999% Definition ATSC transmission. A single availability metric represents 5.3 minutes of lost IP packet can translate into a loss of 7 outage in a year, cable operators expect MPEG2 packets. interruptions to be few, of short duration and restoration must be seamlessly engineered Network Reliability so that continuity of the video is preserved for the user. Based on these criteria, a Service availability is another key successful re-convergence of an interrupted technical capability required to offer video stream should occur within 1 second broadcast quality content delivery to a cable or less of detection. headend. Since satellite is typically the delivery method for all channels, satellite Performance benchmarks and seamless reliability can be considered a benchmark end-user reliability guarantees are not all for video service delivery. Operators don’t that drive the IP video distribution technical think about availability on an individual requirements. In fact there are several other channel basis when all channels arrive via important business criteria that drive the same transmission path. addition technical requirements that must be met in order to offer a viable service. These However, when various delivery options criteria are: exist and slight changes in availability performance can be traded against economic • Costs benefits, a range of acceptable availability • Competition might become part of the delivery system • Functionality decision. For example, IP networks can be • Efficiency designed for 99.99% availability and 99.999% availability. These two metrics The costs of converting to and using represent a difference of 48 minutes of Video-over-IP distribution must meet certain availability in a year. Yet they represent a thresholds for initial investment (e.g. CapEx) and the ongoing cost of running the mentioned, the IP connectivity is two-way. business. The first threshold is met when It can already support VoIP and Internet the reliability and performance of the IP Access service. The same IP connection can distribution reduces the investment in be used for the CA connection. The use of expensive terminal equipment to control and IP for video service delivery and CAS may groom the received video product. The create new opportunities for CA ongoing costs are even more important to mechanisms. the MSO and are also tightly coupled with the three other criteria: competition, The final criterion, efficiency, is very functionality and efficiency. Satellite important to the MSO as video content distribution is not the only competition for options expand, consumers become more terrestrially-based Video-over-IP; other sensitive to picture quality, and new competition comes from satellite broadcast innovative services create interaction providers (e.g. Dish Network, DirectTV). between video, voice, data, and wireless The level of service offered by these services. The video service delivery must competitors is what actually establishes the be capable of expanding overall capacity market's benchmarks for the Video-over-IP rapidly, support varying levels of video distribution service. The bottom line is that program compression to ensure high quality IP video distribution must at least content, and provide delivery protocols that marginally beat the cost of alternate easily inter-work with the other services. providers for equivalent services or provide Finally, it should be easy to add new significantly better service(s) at only endpoints to the video service distribution to incremental cost increases facilitate easily adding content suppliers and local content distribution networks. One key advantage that Video-over-IP must provide over these competitors is the AN OVERVIEW OF IP TECHNOLOGIES ability to deliver a significantly larger (almost limitless) amount of content. A Listed below are the various IP protocols starting point for the service should begin that can be applied to allow video streams where the competition leaves off. For on an IP network meet the requirements example: stated above. These brief descriptions are here just to provide a high level review of • 200 – 400 channels of SD the terms, as they will be used throughout programming the paper. More detailed explanations can • 25 channels of HD programming be obtained from the IETF and vendor web • 100 channels of CD quality music sites.

Functionality is another important IP Encapsulation of Video Frames consideration for the MSO. Frequently, a Conditional Access connection is required to MPEG2 frames can be encapsulated in support the video service delivery, and the an IP packet. Seven MPEG2 frames are CA authority is located away from the typically combined into one IP packet since headend. These implementations today need this creates a packet size within the 1500 to coordinate access to the satellite video byte limit of Ethernet and enables the packet services through a separate terrestrial CA to be easily moved between layer two network connection. As previously transmission protocols. While MPEG2 is currently the most common video stream Virtual Private Network (VPN) protocol, other encoding protocols are also easily placed into IP packets. These IP This segregates traffic on the network packets are typically sent as UDP frames. such that the network operator can keep The UDP protocol does not include the some traffic flows separate from each other ability for the receiving end point to request and from the public Internet. Public Internet the retransmission of a lost packet. A traffic just reacts to the IP destination, and national network with 50msec latency might will pass any packet to any requested actually allow for 100msec round trip to be destination. VPN allows the operator to used to retransmit a packet. However, establish additional rules for traffic flow. Forward Error Correction (FEC) and These rules can include restricting interleaving packets are more common participation in the traffic flows, encryption methods for correcting for lost packets. The of the packets in a particular flow, and Real-Time Protocol (RTP) is also utilized to packet routing based on VPN identification help sequence packets on the receiving end. instead of IP address, If two packets arrive out of order, the sequence numbers in the RTP protocol will Multicast allow the receiving end to assemble the video stream in the correct sequence. Most traffic flows in an IP network are point-to-point transmissions. A single Quality of Service (QoS) source wants to deliver a packet to a single destination, a Unicast flow. Multicast is QoS standards allow traffic to be marked used when one source wants to send the for specific handling when the network is same information simultaneously to multiple congested. The most basic handling of IP recipients. The multicast routing protocol traffic is called “best effort”. There is no builds a tree distribution map for all the special handling of this class of traffic. The recipients on the network. A single copy of network will try its best to get the packet the packet moves through the network until through as fast as the network will allow. it arrives at a branch in the distribution tree. Packets are processed in the order they At the branch, the network duplicates the arrives, first in/first out. The highest quality packet and sends one copy of the packet for traffic handling is called Real-time class. down each branch of the tree. This class gets the top priority from the network equipment. This class of traffic MPLS Traffic Engineering (MPLS-TE) will only be dropped if all the capacity allocated to this class is consumed. Network Traffic Engineering is an extension of operators typically allocate enough capacity the MPLS standard that provides the to this level to support all Real-time traffic network operator with more control over the they’ve agreed to accept to avoid any path packets take through the network. This dropped packets once it has entered their control serves to aid capacity management network. Other classes between Best effort as well as fast failover recovery. The and Real-time define specific behavior operator can define specific paths for during network congestion periods to defined MPLS flows. TE also allows the prioritize and drop packets based on the operator to define a specific failover path for needs of that traffic class. an MPLS flow. This pre-defined failover has become known as Fast ReRoute (FRR). FRR enables the network to recover a data Model B: A VPN Multicast. stream in < 100msec because the alternate path is already known. Since it was An IP network configured with VPN and predefined, the network doesn’t need to take Multicast makes a significant improvement time to discover alternate routes. over the basic implementation. Implementing the VPN protocol helps POTENTIAL IP IMPLEMENTATIONS protect the video content from being FOR IP VIDEO CONTENT intercepted by unauthorized parties during DISTRIBUTION transmission. The VPN protocol restricts the traffic to specific end delivery points. The most basic implementation would The network operator controls access to this be to Unicast an IP video stream across a traffic stream to authorized recipients. wide area IP network. This implementation would require the video stream to be The application of QoS markings to replicated at the source for each headend Multicast IP packets is currently not destination. It provides little if any available across all vendors and routers, so protection of the content, implies a best the reliable delivery of these streams will effort delivery and does not scale very well usually be dependent on the network when there are multiple destinations. It’s operator allocating sufficient bandwidth easy to see that this doesn’t fit very well throughout the network to avoid congestion against requirements for quality video delay. In addition, a route failure in a programming distribution. The following is traditional IP implementation can cause up a discussion of how other protocols can be to a 10 second outage while the IP network applied to improve upon this basic recalculates routes around the failed implementation. connection. (An advanced network can provide SLAs that are significantly shorter.) Model A: A Multicast solution. Once the route is reestablished, it can take many more seconds for the video stream end We can make the basic implementation point to resynchronize, re-establish buffers, more efficient by implementing multicast. and return the video stream to a stable flow. With video services, we expect a single This amount of video stream loss far source and many recipients. Multicast exceeds our requirements for availability. provides the ability for the network to take in one video stream and distribute it to Model C: MPLS-TE for Fast Network multiple recipients by replicating the stream Recovery and QoS only when necessary. This greatly reduces the capacity demands from the basic It is desirable to have sub-second implementation and enables the operator to restoration capability for video distribution, transmit many channels to many end-points which can pose problems when required of within a reasonable network capacity an IP network. For conventional IP allocation. For example, one allocation of 2 networks, it typically takes five to ten Gigabytes could distribute 300 channels for seconds to have traffic rerouted around SDTV and 40 channels of HDTV to any connectivity failures; either failed links or recipient connected to the network. failed nodes. It’s possible to tune nodes on the IP network (using "hello interval", "dead timer", and by leveraging a calculation of the hold time between two consecutive SPF of the incoming real time traffic stream is calculations) to improve IGP convergence higher than the available bandwidth for the time. However, it’s still not easy to shave real time traffic. The network operator the recovery time down to a level below one needs to know how much video traffic will second. FRR (FastReroute) in MPLS TE be coming into the network so allocate the technology enables fail over time of less necessary bandwidth through the network. than 50 ms; an interval that matches the link DiffServ-aware Traffic Engineering (TE) is restoration capabilities of SONET. Fast a tool for network operators to implement Reroute is initiated for a Label-Switched the appropriate bandwidth allocations. Path (LSP) when the feature is enabled for DS-TE is meant to enable computing path the associated LSP tunnel as a result of a per class with different bandwidth configuration command on the head-end. constraints, and perform admission control The head-end router is responsible for over different bandwidth pools. OSPF informing all routers along the LSP's path extensions for DS TE allow advertisement that the LSP is requesting protection. The of unreserved TE bandwidth, at each LSP tunnel head-end control module will preemption level, for each class type. In keep RSVP informed of the status of the DS aware TE tunnels setup time, LSP Fast Reroute attribute for all active LSPs. signaling includes class type as a tunnel When the RSVP module in a Label Switch parameter, in addition to bandwidth, label, Router (LSR) [other than tail end] along the explicit route, affinity, preemption, LSP's path learns that the LSP should be adaptability and resilience. Class-type protected, it will initiate local Fast Reroute aware call admission control will be protection procedure to protect the LSP performed at each LSR during the DS TE against possible failure of the immediate tunnel setup. Rate limiting at the head end downstream link. Upon link failure, all of the DS TE tunnel can be configured to protected LSPs switch to the backup path. ensure the traffic into the tunnel does not FRR performs the operations to prevent the exceed the provisioned tunnel bandwidth. downstream routers (still along the path in use by the LSP) from tearing down the LSP, Unfortunately, MPLS does not support if the failure is also detected downstream. multicast. MPLS tagging assumes a packet coming into the network can be mapped to Content delivery services require more one exit point on the network. MPLS can significant guarantees for bandwidth rates accept a packet coming into the network and for Quality-of-Service (QoS) from the from one of many possible entry points, IP network than conventional IP services do. assign a tag representing the appropriate exit As currently formulated, the leading IETF- point, and efficiently direct that packet to the endorsed architecture for QoS maintenance correct single exit point. Multicast wants to of differentiated services, "Diffserv", is do the opposite. Multipoint processing strong on simplicity and weak on bandwidth assumes a packet entry at a single exit point guarantees. In the case of network should be distributed to many exit points. congestion events, different services would Relevant IETF working groups are compete for the available link bandwidth. discussing changes to MPLS that could A strict priority queue that includes support multicast traffic, but it may be a bandwidth policing for real-time traffic year or more before those changes begin to could be enforced, but packet loss and appear in network equipment. latency still cannot be guaranteed if the rate As it happens, MPLS-TE no longer has a the standard FRR TE mechanism, therefore, monopoly on Fast Reroute; standards bodies ensure recovery within 50 ms in case of and the vendor community are working on link/node failure events. Fast Reroute on native IP connections. This approach must also be evaluated and A caveat is: even though there are compared to MPLS-TE as standards emerge. obvious benefits of deploying TE tunnels in IP network, there are concerns about its Model D: We need a Solution that scalability and the complexity it adds to incorporates QoS, Multicast, Fast Reroute, network operation. For a facility based ISP and VPN. that owns the physical links and infrastructure of its IP network, capacity Possibly the best solution is a constraint is a relatively minor issue combination of all these technologies. We compared to other ISPs which have to need multicast to make efficient use of purchase or lease capacity from other network capacity. We need QoS to ensure providers. It is hard to justify sending all consistent, on-time delivery of the packet IP traffic into fully meshed TE tunnels stream. We need VPN to enable access ubiquitously deployed for a facility based control. Finally, we need Fast Reroute ISP. Instead, only special traffic, such as capability to minimize interruptions to the VoIP, broadcast video, video conferences, or video stream caused by network failure VoD transported in IP network, are events. Standards bodies and vendors are candidates to be carried in MPLS TE working to make the whole combination tunnels. This requires the traffic that enters available. a configured MPLS TE tunnel get preferential treatment over all other traffic Point to Multipoint Traffic Engineering by all routers’ queuing and congestion Label Switched Path (P2MP TE LSP) is avoiding mechanism along the path. TE currently being proposed in support of the queues for configured MPLS TE tunnels in construction of a Point-to-Multipoint every router the tunnel traverses had been (P2MP) backbone network for multicast proposed. The proposed TE queues for services. In such a scheme, a P2MP Label P2P unicast tunnels can be extended to Switched Path (LSP) will be set up between P2MP multicast tunnels. an ingress Provider Edge (PE) and multiple egress PEs; the ingress PE would IN CONCLUSION accommodate a multicast source, and the multiple egress PEs would accommodate Cable operators have made IP protocols multicast receivers. Ingress/egress PEs at an important part of their network services the edge of the multicast network will for Internet Access, VoD, and VoIP. NGNA handle subsequent multicast routing. The is creating additional opportunities for IP P2MP LSP will be set up with TE based services in the network. Advances in constraints and will allow efficient packet IP technologies provide are creating the replication at various branching points in the opportunity to move national broadcast network. The proposed P2MP TE LSP video distribution to IP networks. The would be established by setting up multiple application of Multicast and VPN with standard P2P TE LSPs. If each P2P sub- sufficient bandwidth allocation can provide LSP is protected by its backup-tunnel, the very reliable video distribution and could be multicast video traffic can be protected by used for some channels today. The application of Fast Reroute capabilities can mpls-rsvp-p2mp-04.txt by S. Yasukawa, bring recovery from link outages to around A. Kullberg, and L. Berger 100msec and make IP video delivery even 2. Requirements for Point to Multipoint more reliable. Mixing national video extension to RSVP-TE, draft-yasukawa- distribution with Internet Access and VoIP mpls-p2mp-requirement-01.txt traffic creates economies of scale throughout 3. Performance Analysis of MPLS TE the business. Putting all service delivery on Queues for QoS Routing, Yihan Li, IP enables new opportunities for delivery to Shivendra Panwar, and C.J. (Charlie) end consumers and innovative service Liu, Proceedings of Applied integration. Moving national video content Telecommunication Symposium/ASTC, distribution to IP networks could be the next 2004 (pp.170-174), April 18-22, big service breakthrough for Cable Arlington, Virginia. operators. 4. MPLS TE Queue Creation as a Mechanism for QoS Routing, C.J. REFERENCES: (Charlie) Liu, Proceedings of Applied Telecommunicatio Symposium/ASTC 1. Extended RSVP-TE for Point-to- March 30-April 3, 2003, Orlando, Multipoint LSP Tunnels, draft-yasukawa Florida

CONTENT PROTECTION CONSIDERATIONS FOR DIGITAL CABLE READY PRODUCTS AND SECURE HOME NETWORKING

Brad Hunt1 and Jim Williams2 1Sr. VP, Chief Technology Officer, 2VP, TV & Video Systems

Abstract them when they move. With the separation of the security functions, content owners and This technical paper highlights several MSOs no longer have a direct relationship important content protection considerations or voice in the construction of cable for Digital Cable Ready products including receiving products. secure digital outputs, steps to address the “Analog Hole”, secure integrated personal Content owners continue to have a vital digital recorders and secure home interest in ensuring that all content networking. distribution platforms are secure not only for existing services but also for future Adding new features in a secure manner envisioned service offerings. Their views will help maintain the viability of cable concerning the security aspects of a content television in the competitive and expanding acquisition device should be incorporated market of digital content distribution. It will into the device’s technical specifications and also better position the cable industry to the content protection related licensing launch new innovative programming terms. In addition, the process for approving services that can increase revenue, control new protected digital output and secure churn, and expand the subscriber base. recording technologies must also include a role for content owners.

INTRODUCTION As directed by the FCC, the Cable and Consumer Electronics industries have begun In the recent past, marketplace solutions working with content owners to define the for content protection and security were content protection requirements for next- developed by Cable MSOs and other generation bidirectional “Digital Cable MVPDs through independent negotiations Ready” products. These ongoing and contractual obligations with content discussions regarding the bidirectional providers and receiver manufacturers. As framework have provided content owners part of its implementation of Section 304 of with an opportunity to express and discuss the Telecommunications Act of 1996, the their views on content protection. As FCC issued its Second Report and Order in recognized in the Broadcast Flag regulation, October 2003 that outlined rules and the ability of distribution channels to attract standards for unidirectional digital cable high value content is enhanced by due ready products. Through this ruling the recognition of the security needs of content FCC committed to the principle of owners. separation of security functions from the base customer premises equipment to This technical paper will highlight support the retail availability of digital cable several important content protection set-top boxes that consumers could take with

considerations related to Digital Cable SECURE HOME NETWORKING – Ready products. EVOLUTION FROM COPY PROTECTION TO CONTENT APPROVAL OF EFFECTIVE DIGITAL PROTECTION CONTENT PROTECTION TECHNOLOGIES The nature of customer premises equipment is changing -- evolving from one As important partners in enabling content or more independent receivers with analog distribution over cable, content owners have video outputs to a suite of networked digital a legitimate interest and should have a devices that have access to shared resources meaningful role in approving new digital including tuners, mass storage devices, optical content protection technologies in digital media burners, computers and Internet cable products. These include new protected connections. With continually increasing digital output technologies and secure processing power, Internet connection speed, recording methods. compression algorithm performance and storage capacity, the customer premises Currently, CableLabs has the authority to equipment suite is becoming a digital approve or disapprove new digital output processing, communications, storage, and protection technologies and secure recording consumption powerhouse ripe with new methods for digital cable products content usage possibilities for consumers. manufactured under the CableCARD Host Interface License Agreement (CHILA) and Digital content protection technologies the unidirectional DFAST license. It is not ensure that a particular usage model or cable clear what functional criteria CableLabs service offer that is purchased through a uses to evaluate a digital content protection conditional access system is honored by technology. In fact, the use of a fixed set of downstream devices. Traditional content functional criteria may be too restrictive in protection technologies have focused on allowing for innovation of new content copy protection. As customer premises protection technologies. A more effective equipment evolves into a suite of home manner of analysis and approval should be networked devices and even to devices based on marketplace criteria where content beyond the home, the content protection owners’ views and actions can lead to system must also incorporate redistribution approvals based on the marketplace control. performance of these technologies. At the very least, CableLabs should incorporate a The typical usage rights that might be more formal process that seeks and takes granted in a cable environment include the into account input and advice from content right to make copies, the right to owners as an integral part of their decision- electronically move content around one’s making process. home (e.g. to another TV set) and the right to make a physical copy that can be carried beyond one’s home. However, unrestricted redistribution of content beyond one’s home would be inconsistent with the licensing rights negotiated from the content owners and could undermine the subscriber-based business model of the cable television IMPLEMENTATION OF A DIGITAL industry. CONTENT PROTECTION DEVICE KEY REVOCATION SYSTEM OPENCABLE™ MUST BE UPDATED TO PROTECT THE SINGLE HOME CABLE The cable distribution system must ACCOUNT provide an end-to-end solution for the delivery and processing of digital content The current OpenCable™ specifications protection System Renewability Messages do not provide the ability to distinguish a (SRMs). System Renewability Messages single digital cable subscriber with a home are the common name for the messages that network from a group of separate contain digital content protection device key households “sharing” a single cable account revocation information. Device key using wide area networking. In addition, the revocation provides a content protection content protection afforded by these technology the means to selectively disable specifications should provide the ability to the protected digital output of a signal redistribution control information and compromised device (e.g., a non-compliant manage content usage in accordance with device created using a cloned device key) that signaling. without impacting the general functioning of the device. It is therefore a critical The current OpenCable™ CableCARD™ component in managing the effective Copy Protection System Interface functioning of a digital content protection Specification does not provide a means for technology. signaling Redistribution Control. In addition to signaling numeric copy control Specifically, the cable system must restrictions of Copy Never and Copy One develop a means for efficiently delivering Generation, this interface specification must SRMs from the cable head-end to the digital have a means to signal Redistribution cable receiver. In addition, both the CHILA Control when no numeric copy control and the unidirectional DFAST license must restrictions are asserted. For example, this contain explicit obligations for the digital could be the case for programming delivered cable receiver to perform digital content on the Digital Basic Tier, where a Cable protection device key revocation processing MSO optionally wants to encrypt the service when validly received SRMs are presented. to provide protection against theft of service. Since some digital content protection In this case, the controlled content would be technologies, like High-bandwidth Digital marked in a manner to signal that there are Content Protection (HDCP), do not store no numeric constraints on copying within revocation lists, the CHILA and DFAST the home or to removable media, but the license must explicitly require real-time controlled content must be protected by the processing of SRMs. In the specific case of host device to restrict redistribution beyond the 5C Digital Transmission Content the particular cable subscriber’s home, Protection (DTCP), the CHILA and DFAST including over the Internet. license must require that the device implement “Full Authentication” of the DTCP source function, in order to ensure that full SRM processing is done. These are a few of the requirements for insuring digital cable products incorporate digital content protection technologies that implement an application of analog copy control signaling, effective device key revocation processing such as analog Copy Generation mechanism. Management System (CGMS-A) signaling, has been widely implemented for many ADDRESSING THE “ANALOG HOLE” years in a number of content protection licenses and specifications. This vertical In the process of delivering protected blanking interval signaling allows the digital content, the content must be conveyance of usage rights in analog video converted into an analog video signal in content. Many digital recorders detect order to support legacy displays that have CGMS-A in order to manage unauthorized only analog video inputs. However these copying. For example, when the CGMS-A analog video signals can be easily converted state of “Copy Never” (1,1) is detected in back to digital without any obligations to the vertical blanking interval of an analog preserve and respect the content’s usage video signal to be recorded, the digital rights information. The protected digital recording is stopped. In order for this content is said to escape through the signaling to be deployed effectively, it must “Analog Hole”. The challenge for our be generated correctly in the digital cable set industries is to determine the best way to top box. support legacy analog displays without creating an unnatural impediment to the Both the CHILA and unidirectional migration to digital. DFAST license need explicit obligations for the regeneration and the insertion of vertical Several key features of digital cable blanking interval signals for copy and products are important in addressing the redistribution control. The MPAA has Analog Hole: proposed specific language for explicitly defining CGMS-A, Analog Protection • Analog copy control signaling System (APS), and Redistribution Control implementation; Information (RCI) signaling in these • Image constraint on unprotected high licenses for all analog video format outputs. definition analog video outputs; and In order to ensure full protection, analog • Selectable output control capability vertical blanking interval signaling must for new business models. also be applied both to upconverted standard definition TV programming that is output as Each of these features is an important a high definition analog video signal and, content protection function, and in likewise, to downconverted high definition combination, provides a reasonable TV programming output as a standard approach for addressing the Analog Hole. definition analog video signal. Finally, analog video outputs should not be ANALOGCOPY CONTROL SIGNALING permitted absent a standardized means for IMPLEMENTATION carrying CGMS-A, APS, or RCI vertical blanking interval signaling. This is currently One important component of the solution the case for analog RGB VGA computer to the Analog Hole begins with the use of a monitor outputs. standardized means for signaling copy control information in the analog video outputs of digital cable receivers. The IMAGE CONSTRAINT OF applications that are advantageous to UNPROTECTED HIGH DEFINITION consumers, such as new early-window ANALOG VIDEO OUTPUTS business models. For example, in order to create a more secure environment for an Content owners are very concerned about early-window high definition video the introduction of digital recorders that programming service, an MVPD may find it exploit the high definition Analog Hole. The advantageous to deliver this service with the price of high definition analog-to-digital requirement that unprotected analog high video converter devices is falling and could definition video outputs are disabled and soon lead to the introduction of consumer only digital outputs protected with HDCP devices that digitize and record unprotected and DTCP are allowed. analog high definition video content. The use of image constraint on unprotected HD Second, selectable output control could analog video outputs is an important tool in also help address unknown problems, such addressing the high definition Analog Hole. as patent claims and court orders involving a The optional use of image constraint on previously-approved content protection unprotected analog high definition video technology. outputs has not been demonstrated to have any visual impact on legacy HDTV displays In order to make these future permitted having only analog video inputs. uses possible, manufacturers should be required to incorporate selectable output The use of image constraint provides control capability in all digital cable incentives for consumers to use the higher- products. quality, protected digital interconnects that are becoming available in the marketplace. CONTENT PROTECTION Since the obligation to implement image OBLIGATIONS FOR HARD DISK DRIVE constraint is in the DFAST license, all INTEGRATED RECORDERS unidirectional CableCARD-equipped host devices being introduced today have image Integrated Personal Digital Recorders resolution constraint capability. This (PDR) in Digital Cable receivers provide capability must be implemented in future many attractive benefits to consumers, such digital cable products. as pause, time-shifting, and the movement of temporarily stored recordings of “Copy One SELECTABLE OUTPUT CONTROL Generation” programming to removable CAPABILITY FOR NEW BUSINESS media. But in order for integrated recorders MODELS to provide this functionality, the content and associated usage rights information must be Under the unidirectional regulation, the securely and persistently protected and FCC acknowledged that selectable output content usage must be effectively managed control could be appropriate for use in the in accordance with those associated usage future. Cable is afforded two key benefits rights. by deploying selectable output control capability in Plug and Play products. In the case of temporary recordings of “Copy Never” programming, the content First, as suggested by the FCC, selectable must be cryptographically bound to the output control might enable future receiving device doing the recording so that it is not removable and not itself subject to slot, they will not be able to access further copying before it is rendered interactive programming services, such as unusable. The temporary copy should be interactive Video-On-Demand (VOD) and encrypted in a manner that provides no less impulse Pay-Per-View (PPV) offerings. If a security than that of the Advanced successful conclusion is reached in the Encryption Standard (AES) using 128-bit cross-industry bidirectional digital cable keys. Since rights associated with “Copy negotiations, a new bidirectional framework Never” content preclude making a will be created producing a new generation permanent copy, the default expiration time of bidirectional digital cable ready products of temporary recordings of “Copy Never” that incorporate advanced content content should be 90 minutes. This also protection, copy management, and device requires that the cable system provide a programmability. These features will better secure source of time to the digital cable enable cable operators to provide a wide receiver/recorder in order for it to securely range of new interactive programming manage time expiration of bound copies. services, including early-window content, to cable subscribers purchasing these new In the case of recordings of “Copy Once” bidirectional devices. programming by integrated PDRs, many of the same requirements for “Copy Never” However, content owners are concerned content are also needed. In addition, these that consumers must be properly educated recordings must be remarked to “Copy No about the more limited set of programming More” to prevent further copies from being services available to a unidirectional digital made by downstream recording devices. cable receiver as compared to the wider range of new, interactive services that will Finally, one of the most important be available to subscribers purchasing missing features of current Digital Cable bidirectional digital cable products. Ready products is the provision of a secure Although the current market availability of time source and a standardized means for unidirectional devices is helping to facilitate signaling time expiration of bound copies. the Digital Television transition, content Incorporating this functionality into next- owners believe that consumer electronics generation digital cable receivers with manufactures and consumer electronics integrated recording capabilities is critical in retailers must accept the responsibility for supporting a wider range of time-shift, clearly labeling digital cable ready products rental, and sell-through programming and for educating consumers about the options for consumers. programming and interactive service availability differences. This is critical to LABELING STANDARDS FOR help the customer make an informed UNIDIRECTIONAL AND purchase decision when considering whether BIDIRECTIONAL DIGITAL CABLE to buy a unidirectional or an advanced PRODUCTS bidirectional digital cable ready product.

Based on the bilateral-negotiated DFAST SUMMARY license, a broad array of unidirectional Digital Cable Ready products are beginning This technical paper has highlighted to be sold in the marketplace. Even though several important content protection these devices incorporate a CableCARD considerations related to Digital Cable Ready products. Addressing these issues is subscriber base. Content owners look a critical step in maintaining the viability of forward to continued collaboration with the cable television in the competitive and cable and the consumer electronics expanding market of digital content industries in addressing these issues that will distribution. It will also better position the lead to the introduction of exciting new cable industry to launch new innovative digital cable products and program service programming services that can increase offerings for consumers. revenue, control churn, and expand the

CONVERGED DATA NETWORK ARCHITECTURE (CDNA)

Michael J. Emmendorfer Charter Communications

Abstract CONVERGED DATA NETWORK ARCHITECTURE (CDNA) What is CDNA? First, CDNA is the migration of all services (Voice, Video, Data) Executive Summary both serving residential and commercial customers over a Common Network Charter Communications – like most Architecture, including optical transport, Multiple System Operators (MSOs), Internet Protocol (IP) core, IP distribution, Regional Bell Operating Companies various IP/media access layer technologies (RBOCs), and Service Providers – is (coax, fiber, or wireless). Second, CDNA is a preparing for an “All IP World,” and is at methodology with the objective to reduce various stages of implementing this IP layers of the network by converging strategy throughout our networks. This white component functionality within the network paper is an overview of Charter elements, as well as increasing the remaining Communications’ strategy and approach for a network functionality to support any service Converged Data Network Architecture over any access layer technology. Third, (CDNA). The network will need to support CDNA is the convergence to a Virtual all voice, video, data products serving National Backbone using MPLS (Multi- residential and commercial customers over Protocol Label Switching) technology. one common infrastructure. Some of these service and lines of business include Internet In summary, CDNA takes a holistic view Data Services, Telephony, Interactive TV, of the network and has three fundamental Video On Demand (VOD), All Digital principles: 1.) Convergence of Services over Video, Multi-Media Services, and a host of Common Network Architecture, 2.) local and national based Commercial Convergence of Network Layers (Reducing Products and Services. Network Elements and Increasing Service Functionality), 3.) Convergence to a Virtual The CDNA target architecture will Core Backbone using MPLS Technology continue to evolve over time. It is important to note that CDNA is not limited to coaxial- This white paper will illustrate an overview based technologies and architectures as stated to our Converged Data Network in the abstract, but rather a holistic view of Architecture, including our evolutionary the network end-to-end with target plans, and benefits as well as challenges to architecture to support any service through this target network architecture. any network technology (coax, fiber, or wireless).

The Convergence of Services over common network architecture, such as commercial and data services across an MPLS enabled CMTS, as well as switch/router platforms, will allow for the The benefits to a CDNA strategy will replacement of historic services such as create the vehicle for integration of services, Frame Relay, and has been deployed since improve the customer experience, increase 2002. Charter views MPLS first as a service loyalty, and defend our customer base. The or revenue enabling technology, second as a CDNA architecture is a migration away from network label switching technology, and proprietary systems to standards-based finally the ability to logically partition products that will enable us to accelerate the services across the network. We have advanced service deployment while driving continued to converge some services over our down operational efficiencies in terms of IP distribution and optical transport network leveraging a common work force and fewer infrastructure, which includes Residential network systems. Voice and Data Services, VOD, Simultrans (All Digital Video), and Commercial There are challenges that we have Services to various enterprise customers encountered and options that need to be considered as we make our way to a truly The Convergence of Network Layers in converged network. A few of the challenges the metro area will reduce the cost to the include driving full service features on the 1st MSO, and our target is to enable access layer and 2nd Generation CMTS, defining the 3rd elements as well as distribution layers with generation CMTS/Edge QAM to support the DWDM optical interfaces, by-passing the full breath of commercial services to address traditional transport layer as well as competitive threats (a possible PON aggregation routers. The third generation replacement technology), integration of long cable platform providing video and data reach DWDM optics on distribution as well service is being formulated. There are new as the access layer elements with fault revenue generating services and the ability to management capabilities. create CAPEX and OPEX expenditures by enabling the legacy access layer platforms Introduction (CMTS) to support a full range of commercial and residential products lines. This white paper is an overview of a target network architecture based on business, Charter’s vision is to create convergence operations, and technical requirements to to a Virtual Core Backbone implementing support Residential and Commercial Data MPLS Technology, and will allow us to take Services, Telephone Services (VoIP), VOD, the lead in various data, telephony and iTV, Simultrans, All Digital, and other commercial services. IP-based Products & Services, internally called Converged Data Network Architecture The opportunities and benefits of CDNA (CDNA). have just begun to emerge, and others will be realized over time. We are able to leverage Charter Communications has convergence multiple Lines of Business (voice, video, and at various portions of the network today, and data) to support the network infrastructure, others will emerge over time. This paper has and improve the economics for all lines of three fundamental principles for a CDNA business, both from a capital and operation strategy. perspective. First, we will begin with the Convergence of Services over a Common Network Architecture supporting Commercial and BUSINESS REQUIREMENTS Residential customers with several product AND SERVICE DRIVERS offerings including Data, Voice, and Video. An advantage of convergence of services Charter Communications offers four core across a common network is economies of services that are starting to cross-pollinate scale and the ability to offer more features with each other as outlined below. This and integrated services for the consumer. hybrid approach will require more feature sets as the services evolve, but it can be Second, the methodology of Convergence classified into the following service of Network Layers with the convergence to categories: Video, Voice, Data, and Network. IP will inherently force the migration of services from legacy platforms (often Video Based Transaction Services proprietary) to IP network elements (which • VOD, SVOD, HDTV, etc. will be standards based). In addition, this • Migration to a All Digital Network second principal of the convergence of the • Interactive Set-Top and DVR network is increasing functionality and • Migration to End to End IP Based allowing the service provider to reduce layers VOD and Content Services of the network and also enhance the remaining layers functionality to support the Voice Based Transaction Services convergence of services across a common • Primary Line Voice Services network. This can fundamentally change the • Business Class Telephone Services network architecture. • Multimedia Voice Service (SIP)

• Integration with Video Products Finally, the third principle is the

Convergence to a Virtual Core Backbone Data Based Transaction Services: using MPLS VPN Technology. MPLS technologies have been in operation and • Unified Mail Services offered as a service by backbone service • Web Hosting and DNS providers for years. However, the adoption • Centralized Storage and Data Back- of this technology in the cable industry has, Up until recently, been only in the hands of a few • PC-based Virus/Spam Protection across the world; and of those few most are only using MPLS as a network label Network Based Services switching technology and for traffic • Bandwidth Speeds (Tiers) engineering across a backbone or MSO core • Quality of Service network. In 2002, Charter embraced MPLS • Network Based Virus Protection as a Service / Revenue enabling technology • Security and VPN Services to position services against the RBOCs and • Bandwidth Management Service Providers utilizing MPLS VPN (RFC • Single or Multi-Site Connectivity 2547 bis) as well as other MPLS based Products technologies. • LAN Extension Services • Transparent LAN Services • Frame Relay Replacement • Routing Services • TDM and SONET Services • Network Storage Transport Services ENGINEERING AND FUNCTIONAL TRACKS OPERATION REQUIREMENTS Charter has taken an approach to partition The CDNA architecture requires the end-to-end service delivery network into centralized management (NMS & EMS) and Functional Tracks that represent common full Fault Management, Configuration, types of services and/or technological areas Accounting, Performance, and Security of concentrations. This paper will concentrate (FCAPS). A key driver for CDNA is on Track 2 – Converged Data Network reducing the complexity of the network, Architecture (CDNA) with an example reducing the layers of the network, and illustrated below in figure 1. differentiated products for service delivery. Also, accelerating the convergence of services and network architectures with the requirement of high reliability and availability across standards-based platforms will advance services deployment, and drive operating efficiency.

Figure 1: Functional Technical Tracks THE CDNA TECHNICAL AND The figure below is the CDNA access MIGRATION STRATEGY layer component support residential and commercial voice, video, and data service, Overview both end-to-end IP and MPEG.

Reaching our target architecture of full convergence of all services across one network will certainly “not happen over night” since this is an evolutionary path. The following section is a high-level strategy for CDNA, and a migration plan to the target Figure 3: Next Generation Access Layer architecture. The target architecture has four Elements (Headend and Primary Hub) layers, including optical transport, Internet Protocol (IP) core, IP distribution, and Stage 1 Service Enabling Legacy Access various IP/media access layer technologies (coax, fiber, or wireless). This next section Charter’s objective is to increase will examine the Access Layer to increase functionality of the current CMTS functionality to offer new services as well as infrastructure with software enhancements, technology improvement to allow seamless including Sub-interfaces, VRFs, Layer 3 integrations. The Figure below is the end-to- MPLS, MB-BGP, IS-IS Routing Protocol, end a high-level strategy for CDNA: and others. This will enable the MSOs to leverage an existing CMTS infrastructure for new revenue streams with higher margins. This asset continues with its financial depreciation schedule, but with software we enable a new revenue stream like those to support commercial services; specifically services to replace the incumbent provider. Figure 4 below represents MPLS enabled CMTS to provide frame relay replacement services to support our Commercial Business Unit.

Figure 4: Delivering Commercial Services Figure 2: CDNA End-to-End with MPLS VPN enabled CMTS Delivering These technologies are traditionally found on switching platforms. There are critical service and management features for the next generations / 3rd CMTS/Edge QAM.

We launched Passive Optical Network (PON) technology in 2002 and SONET services in 2000 to support commercial product offerings such as Ethernet and TDM services, our service enabling strategy is to place CWDM optics in the access network between the Charter facility and the customer, to conserve fiber assets. Figure 5: Commercial Services using MPLS VPN enabled CMTS and Switch/Router Stage 2 Converged IP Distribution for All Services The illustration in Figure 5 enables the customer to use our IP VPN to connect Charter has placed over its Distribution remote site(s) to a central office. This layer network Residential Data Services, example has the customer provisioning their Commercial Services, Voice Service, VOD PCs and SIP phones with their own IP Services, All Digital video product offering. address space, their SIP phones use our provided private network to use 4 digit Stage 3 Transport Network Migrations dialing as well as off-net dialing. Charter CDNA Architecture will greatly change the Optical Transport network in the metro market; to that end Charter has begun a migration of IP traffic off of SONET as well as RPR over SONET, to transponder and muxponder.

Stage 4 Bandwidth and Security Management System (BSMS)

What is BSMS? BSMS is unified solution for bandwidth management (high bit rate Figure 6: Customer Traffic Transparency application management) as well as Security across Charter’s Network Management (Intrusion Prevention Systems, Protection from DDOS, and targeted services security protection). Charter implemented We are exploring additional service BSMS in our large markets; as a result of this enhancement to the CMTS, which may deployment we captured the following include Layer 2 Services for Transparent statistics: LAN Services, using EoMPLS, QinQ (*802.1q Tunneling), as well as VPLS. • Avg. of 40,000 malicious packets for Stage 5 Convergence to a Virtual Core every 100k subscribers per day. Backbone using MPLS • In one of Charter’s larger markets there were 14 million malicious The creation of a Virtual National packets events in 30 hours Backbone using MPLS creates cost savings • Attacker Ratio: For every 1 attack and revenue generation opportunities thought that comes into a Charter Network not possible because of geographic separation there are 11 attacks from our from our facilities (Headend, Hubs, and subscribers going out Offices). Charter plans to create a Virtual Backbone using MPLS, using two technical A case study: BSMS Value Assessment and approaches representing two business needs: Network Worms 1. MPLS - Charter becomes a customer of a Impact of Network Worms and Malicious MPLS-enabled service provider Attacks Measurement from July 2003 – establishing connectivity to remote offices November 2003 and facilities around the country to support • August 11: The Blaster Network the internal enterprise network. This effort Worm was introduced replaces the frame relay technology that is • August 22: The SoBig Virus Hit was currently implemented. introduced 2. Hierarchical VPNs – For Charter to Table 1: Call Center and Operation Impact become an MPLS VPN Service Provider from Network Worm (Source: Jon Mandani, would require connectivity across an Charter Communications) MPLS VPN enabled service provider nationwide. This enables some key features:

a. Commercial Services to provide connectivity and QoS to our customers between our current network footprint as well as outside service areas. Global customers are now possible across this virtual network.

b. MPLS enabled service (VoIP and We deploy BSMS at the border of our others) to provide connectivity and network at the Internet drain; this does not QoS across carrier networks to other protect or manage traffic to and from our Charter sites or partners. subscribers, which would be on net. As we approach convergence of Voice and Video on one IP network this feature has a considerable increase in importance. Our BSMS strategy is a requirement for the next generations / 3rd CMTS/Edge QAM, Passive Optical Network, and Switch/Router. This opportunity is realized by the ability to offer services where an MSO’s backbone or metro network cannot reach without a significant capital investment.

We are considering placing other services across our MPLS Virtual Backbone as well as the Hierarchical VPNs.The diagram on this page illustrates the Internal Enterprise Network (MPLS), Revenue Network MPLS VPN - Hierarchical VPNs, and Internet. Figure 7: Connectivity Across MPLS Carrier

Figure 8: Illustration of Interconnect of two (2) MSO markets connected over MPLS enabled Carrier(s) utilizing MPLS, Hierarchical VPNs, and Internet Access. There are a multitude of technical considerations for Hierarchical VPNs and MPLS: • Using the RFC 3107 - Carrying Label Information in BGP-4 to interface with the carrier, aka BGP send label • DSCP for IPv4 to Carrier MPLS EXP • EXP MPLS to Carrier MPLS EXP • Depth of label stacking for your MPLS VPN Carrier • Multicast over MPLS VPN • MPLS Monitoring Visibility

Stage 6 ITU DWDM Enable Distribution and Access Layer with ROADM Technology

The integration of long reach ITU Dense Wavelength-Division Multiplexing (DWDM) Gigabit Interface Converter (GBIC) (or SFP)

“natively” on access layer elements, like that Figure 9: ITU DWDM GBIC (or SFP) of CMTS, L2/L3 Switches, PON, Edge “natively” on Access Layer & Distribution QAM, as well as the switch/router Layer distribution layer elements could bypass the transport layer in the metro markets. In fact, There are a host of challenges with native aggregation L2/L3 switch/router(s) located at ITU DWDM Ethernet Interfaces (1) Gigabit primary hubs would not be required as well, and/or (10) Gigabit that will need to be with native DWDM optics on the access address and solutions developed, however layer elements. once realized the capital and operational savings are significant. Assets already deployed in primary hub locations could be re-allocated from Switch Though work to improve router switch Aggregation and Optical Metro Transport over and convergence time, like those found Distribution to Access Layer devices used for in RFC 3623 - Graceful OSPF Restart and revenue generation. Figure 8 illustrates this other similar standards to improving the proposed design; this architecture reflects the protection capabilities of the router to be at optical transport layer as well as L2/L3 parity with SONET scheme. However the Aggregation Layer Switch at the Primary photonic level impairment measurements Hub (s) as these components are not required. protection schemes inherent with SONET/SDH and ITU G.709 are not available, such as: 1. Performance monitoring Fault management, errors, alarms, and performance monitoring like which exist in SONET/SDH and ITU G.709, for optical layer problems. 2. SONET/SDH and ITU G.709 defined Reconfigurable Optical Add/Drop threshold are met for link Multiplexer (ROADM) is being considered impairments a message for automatic as part of the CDNA strategy for per lambda protection switching (APS) occurs. optical wavelength managements. 3. POS interface contains this protection. Stage 7 The 3rd Generation CMTS Platform

Below, is a set of questions to the industry Charter has defined an Edge Layer that surrounding protection schemes for Ethernet can support services and technology, without wrappers (G.709) so that protection independent of media (coax, fiber, and scheme like those defined by SONET/SDH wireless), this integration of the services will and ITU G.709 could be available for native provide greater serviceability of commercial Ethernet? services, arguably one of our industry’s fast growing segments. With regards to the next 1. How can ITU DWDM Ethernet generation coaxial platform, known as the Interfaces (1) Gigabit and (10) third generation CMTS, also known as the Gigabit obtain the optical modular CMTS is certainly the “buzz” these measurement for Errors, Alarms, and days. Performance Monitoring similar to that of SONET/SDH and ITU G.709, As discussed earlier in the paper we use WITHOUT placing a wrapper around the 1st and 2nd generation CMTS platforms to the Ethernet frames? deliver data and voice service to residential customers, in addition we also offer advanced 2. How can Ethernet get the equivalent services to commercial customers. Charter of signal degrade bit error rate partitions the CMTS logically to create measurements (link up but degraded) management sub-interfaces for provisioning and issue the equivalent of SONET and management, as well as for revenue APS to a Router’s IGP to force a re- services. This partitioning enables Charter to convergence of a link that is not down apply routing rules, security rules, or service and the IGP hellos and dead timers for revenue generation per logical interface. (even if default setting are reduced) does not declare the link down? Charter is interested in this next generation platform because we see this not 3. Keeping Ethernet enacted could an just as a platform that can provide integrated equivalent signal degrade BER Data and Video services, but this has the measurement over IP link monitor potential to provide high bit rate services to and measure a consistent stream for commercial subscribers and to augment a degraded link and issue an alarm Passive Optical Network. and/or issue a signal to the IGP to converge. BENEFITS

These performance measurements and We hope CDNA will provide capital protections scheme are important with UDP expense savings through economies of scale stream (VoIP or Video), TCP data traffic and standards platform architecture. We may not detect the impairment, voice and believe that operational expense savings and video could be affected. increase network availability, manageability (remote), and a shared work force supporting 3rd generation CMTS/Edge QAM, especially many services, but one network should in the areas of logical interfaces for improve the economics of all line of management and security as well as business. commercial services capability (bandwidth and services). Convergence will enable fewer facilities for complex and expensive transaction CONCLUSIONS processes (Encoding, Storage, Telephony Convergence to IP end to end will emerge Switching, Email, Hosting and eventually over time for the cable operator, and legacy few Headends. technology are increasing using IP for

transport services, as this evolves over time CHALLENGES we are well positioned to support this

migration. This architecture can have There are two core challenges as part of significant cost savings. the CDNA strategy, Ethernet Optical management and monitoring, like those Michael J. Emmendorfer found in SONET and ITU G.709. In Charter Communications addition, the feature and functionality of the [email protected]

DOCSIS PERFORMANCE ISSUES Jim Martin Department of Computer Science, Clemson University

Abstract cates upstream bandwidth allocation over the next ‘MAP time’. The MAP provides slot as- We have developed a model of DOCSIS signments for particular CMs in the form of using the ‘ns’ simulation package. We iden- data grants, provides opportunities for CMs tify a set of possible DOCSIS performance to request upstream bandwidth using a con- issues which includes complex interactions tention-based request process and identifies between downstream TCP connections and which slots are to be used for system over- upstream MAC operation, vulnerabilities head. caused by MAC level denial-of-service attacks and fairness issues. We summarize our ideas A critical component of the DOCSIS MAC involving bandwidth management to address layer is the upstream bandwidth allocation the issues. algorithm. The DOCSIS specification pur- posely does not specify these algorithms so that vendors can develop their own solutions. INTRODUCTION However, all upstream bandwidth manage- ment algorithm will share a set of basic sys- The Data over Cable (DOCSIS) Service In- tem parameters such as the amount of time in terface Specification defines the Media Ac- the future that the scheduler considers when cess Control (MAC) layer as well as the making allocation decisions (we refer to this physical communications layer that is used in parameter as the MAP_TIME), the amount of the majority of hybrid fiber coaxial cable net- upstream bandwidth allocated for contention- works that offer data services [1]. A Cable based bandwidth requests and the range of Modem Termination System (CMTS) inter- collision backoff times. These parameters are faces with hundreds or possibly thousands of crucial for ensuring good performance at high Cable Modem’s (CMs). The original DOCSIS load levels. MAC interface (version 1.0) provides a best effort service with simple prioritization capa- We have developed a model of the DOCSIS bilities. DOCSIS 1.1, which is currently be- MAC and physical layer using the ‘ns’ simu- ing deployed, adds a set of ATM-like services lation package [2]. In previous work we re- along with the necessary QoS mechanisms. ported on the impact of several DOCSIS op- The follow on standard, version 2.0, en- erating parameters on TCP/IP performance hances the physical layer communication [3]. In this paper we extend those results by methods with higher upstream data rates and looking in greater detail at the impact that the improved tolerance to bursts of noise. MAC layer has on TCP performance when using the DOCSIS best effort service. We The CMTS makes upstream CM bandwidth show that the interaction between DOCSIS allocations based on CM requests and QoS and TCP exposes a possible denial-of-service policy requirements. The upstream channel is vulnerability. By exploiting the inefficient, divided into ‘minislots’ (referred to as slots) contention-based bandwidth request mecha- which, depending on system configuration, nism, a hacker can severely impact network contain between 8 to 32 bytes of data. The performance. We demonstrate fairness issues CMTS periodically sends a ‘MAP’ message to involving TCP and video streaming protocols all CMs on a downstream channel that indi- that are ‘TCP-friendly’. Most streaming video applications do not respond to network All CMs receive periodic MAP messages congestion. The Internet community has ad- from the CMTS that identify future upstream dressed this by developing the Datagram Con- scheduling opportunities over the next MAP gestion Control Protocol (DCCP) which pro- time. If provisioned with a periodic grant, a vides an unreliable datagram transport service CM can send at its next data grant opportu- that includes TCP-compatible congestion con- nity. For best effort traffic, a CM must request trol algorithm referred to as the TCP Friendly upstream bandwidth from the CMTS using a Rate Control (TFRC) protocol. While contention-based mechanism. To improve DOCSIS impacts downstream TCP perform- efficiency, a CM can request bandwidth to ance, it does not impact the performance of transport multiple IP packets in a single TFRC (at least to the same degree). This DOCSIS frame by issuing a concatenated causes TFRC flows to steal bandwidth from request. Further, a CM can piggyback a re- similarly configured TCP connections. We quest for bandwidth on an upstream data summarize our ideas on how bandwidth man- frame. If a CM receives a grant for a smaller agement can address these issues. We propose number of minislots than were requested, the a bandwidth management algorithm that ad- CM must fragment the data to fit into the as- dresses fairness issues that include controlling signed slots. Our model supports concatena- TCP unfriendly flows and also subscribers tion, piggybacking and fragmentation. that consume a disproportionate amount of bandwidth. Figure 1 illustrates the MAP layout used in This paper is organized as follows. The next our model. The first slot at the left of the section presents the operation and features of MAP represents time 0 in the MAP time. our DOCSIS model. We present experimental Data slots are placed at the beginning of the results illustrating the performance issues. MAP and contention slots are placed at the We then present our bandwidth management end. Figure 2 illustrates the upstream trans- algorithm. We end the paper with a discus- mission of a 1500 byte IP datagram from a sion of related work, present conclusions and TCP source directly connected to a CM to a identify future work. sink connected to the CMTS. In Figure 2, time progresses in the downwards direction. SUMMARY OF THE MODEL We assume collisions do not occur. Assuming a MAP size of 80 slots, an upstream channel The model implements the DOCSIS architec- capacity is 5.12Mbps and there are 4 ticks per ture defined in [1]. Packets sent over the slot, 96 slots are required to transport the downstream channel are broken into 188 byte entire packet. The small dark square box po- MPEG frames each with 4 bytes of header sitioned at the beginning each MAP time in and trailer. The model accounts for physical the figure represents the transmission of the layer overhead including framing bits and MAP message in the downstream direction. forward error correction data. The down- Our model sends the MAP at the beginning of stream channel supports an optional token each MAP time. Each MAP describes the slot bucket-based service rate. Each SID service assignments for the next MAP time. The IP queue is treated in a first come first serve packet arrives at the CM during the j’th MAP manner. Depending on traffic dynamics, time at time T-0. The CM sends the band- queueing can occur at either the SID queue or width request message at time T-1 and re- the downstream transmission queue. The ceives the data grant at time T-2. The grant is maximum size of either queue is a simulation allocated in the j+2 MAP time. The CM sends parameter. the frame at Time T-3 and is received by the CMTS at time T-4. The time between T-3 quest slots, all unused slots can be designated and T-0 is the access delay which represents for contention requests. the total time a packet is delayed over the

DOCSIS network not including transmission or propagation time. The model can be con- Data Slots Maintenance slots Contention Slots figured to allocate a specific number of con- tention request slots each MAP. Or, in addi- Figure 1. MAP layout tion to a minimum number of contention re-

TCP Source CM CMTS TCP/Sink 1500 byte IP datagram

MAPj+1 T-0 Map time j T-1

MAPj+2 T-2

access Map time j+1 delay T-3 MAPj+3

Map time j+2 Transmission time for 1530 bytes (96 slots) T-4

Time: At TCP sender At the CM At the CMTS At the TCP receiver

Figure 2. Upstream operation

IMPACT OF DOCSIS ON TCP audio stream), 2% of the CMs to generate downstream high speed UDP streaming traffic The results we report were based on simula- (i.e., a 300Kbps video stream) and 5% of the tion experiments using the network shown in CMs to generate downstream P2P traffic. The Figure 3. The DOCSIS parameters were based P2P model (based on [5]) incorporates an on optimal configuration parameters that we exponential on/off TCP traffic generator that found in a previous study [3]. A set of user periodically downloads on average 4Mbytes nodes were attached to the CMs and a set of of data with an average idle time of 5 seconds server nodes were located in the wired net- between each download. The downstream work. The traffic generators utilized realis- transmission queue at the CMTS was config- tic traffic models consisting of a combination ured to hold a maximum of 50 packets. We of web, P2P and streaming traffic. The net- limited the number of packets that can be con- work and web traffic models were based on catenated in a single frame to two. The the “flexbell” model defined in [4]. In addi- DOCSIS and Web traffic simulation parame- tion to downstream web traffic, we configure ters are shown in Figure 4. 5% of the CMs to generate downstream low speed UDP streaming traffic (i.e., a 56Kbps We varied two parameters in the experiments, bothered by lengthy download times when the the MAP_TIME and the number of CMs. mean WRT metric value exceeds 1 second. For a given MAP_TIME setting, we varied We do not claim this to be an accurate meas- the number of CMs from 100 to 500. We do ure of end user quality of experience. Instead, this for six MAP_TIME settings ranging from it is a convenient, reproducible performance .001 to .01 seconds. reference.

Test server 1 S-1 Model Parameters

WRT probe S-2 CM-1 Test client 1 Upstream bandwidth 5.12Mbps . CM-2 . Preamble 80 bits 45Mbps, 18ms 45Mbps, .5ms . prop delay prop delay . 4.71mbps upstream, 28.9Mbps CMTS Node Node . Downstream bandwidth 30.34Mbps downstream . . . 4 ticks per minislot Test client 2 . TCP Analysis Connection . Default map time: 2 milliseconds (80 minislots per map) 100Mbps links . CM-n …………….. S-n (1-3ms delay) . Fragmentation Off, MAP_LOOKAHEAD = 255 slots Test server 2 Concatonation ON 10Mbps links (1-5ms delay) Backoff Start: 8 slots, Backoff stop: 128 slots 12 contention slots, 3 management slots Figure 3. Simulated network Simulation time: 1000 seconds Web Traffic Model Parameters Inter-page: pareto model, mean 10 and shape 2 We obtained the following statistics for each Objects/page: pareto model, mean 3 and shape 1.5 run: Inter-object: pareto model, mean .5 and shape 1.5 Object size: pareto model, mean 12 (segments) shape 1.2

Collision rate: Each time a CM detects a Figure 4. Simulation parameters collision it increments a counter. The colli- sion rate is the ratio of the number of colli- Web Congestion Experiment Results sions to the total number of upstream packets Figures 5a and 5b plot the channel utilization transmissions attempted. as the load increases. The downstream utili- Downstream and upstream channel utili- zation reaches a maximum of about 64% with zation: At the end of a run, the CMTS com- a MAP_TIME setting of .001 second. In this putes the ratio of the total bandwidth con- case, 12 contention slots per MAP is suffi- sumed to the configured raw channel band- cient. For larger MAP_TIME values, the width. The utilization value reflects the downstream utilization ramps up to its maxi- MAC and physical layer overhead including mum value and then decreases at varying rates FEC bits. as the load increases. As the collision rate Average upstream access delay: All CMs grows, downstream TCP connection through- keep track of the delay from when an IP put decreases. Limiting each MAP to 12 con- packet arrives at the CM in the upstream di- tention slots results in fewer total contention rection until when it actually gets transmitted. request opportunities as the MAP_TIME This statistic is the mean of all of the samples. grows. This explains the high collision rates Web response time: a simple TCP client and reduced downstream utilization for the server application runs between test client 1 runs with large MAP_TIME settings. and the test server 1. Test server 1 periodi- cally sends 20Kbytes of data to test client 1. With each iteration, the client obtains a re- sponse time sample. The iteration delay is set at 2 seconds. At the end of the test, the mean of the response times is computed. The mean web response time (WRT) can be correlated to end user perceived quality by using a very coarse rule of thumb that says end users are

DS utilization versus Load US utilization versus Load 65 50

60 45

55 40 50 35 45 30 40 25 35 Utilization (%) Utilization (%) 20 30 .001 second .001 second 002 second 002 second 25 15 003 second 003 second .005 second .005 second 20 .007 second 10 .007 second .01 second .01 second 15 5 100 150 200 250 300 350 400 450 500 100 150 200 250 300 350 400 450 500 Number CMs Number CMs

Figure 5a. Downstream channel utilizations Figure 5b. Upstream channel utilizations

Avg total upstream access delay # CMs WRT avg versus Load 10 6 .001 second .001 second 002 second 9 002 second 003 second 003 second 5 .005 second 8 .005 second .007 second .007 second .01 second 7 .01 second 4 6

3 5

4

2 (seconds)WRT

access delay(seconds) 3 1 2

1 0

0 150 200 250 300 350 400 450 500 100 150 200 250 300 350 400 450 500 Number CMs Number CMs

Figure 6a. Upstream access delay Figure 6b. Web response time metric results

WRT avg versus # of CMs under attack WRT avg versus # of CMs under attack 1 1.3

1.2 0.9 1.1

0.8 1

0.9 0.7 0.8 WRT (seconds) WRT (seconds) 0.6 0.7

0.6 0.5 0.5

0.4 0.4 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 Number CMs under Attack Number CMs under Attack

Figure 7a. WRT results without rate control gure 7b. WRT with 2Mbps DS rate control Fi

Figure 6a shows that the average upstream service rate of 2Mbps, if we turn on ACK access delay becomes very large at high loads filtering or if we allocate all unused slots for when configured with large MAP_TIME contention requests. However, if we increase settings. Even for lower MAP_TIME values, the downstream transmission queue size at the the access delay was significant. For a CMTS (the point where loss occurs) from 50 MAP_TIME of .002 seconds, the access delay to 300 packets, loss no longer occurs and exceeded .5 seconds at the highest load level. downstream utilization approaches 75%. To assess the impact of the cable network on While a larger buffer improves performance, end-to-end performance we monitored web the important result is that the downstream response times. Using the rule of thumb de- TCP traffic is subject to extreme levels of scribed earlier, Figure 6b suggests that for ACK compression caused by DOCSIS. MAP_TIME settings less than .005, up to 300 users can be active before performance be- DoS VULNERABILTIES comes bothersome to end users. In this section we show that it is possible for a When 100 users are active, the collision rate hacker to take advantage of the inefficient is about 50%. What makes this result alarm- contention-based upstream bandwidth alloca- ing is that the web traffic model accounts for tion process by initiating a denial-of-service the heavy tailed distribution associated with (DoS) attack. To accomplish the DoS attack, web user idle times. Consequently, the num- a host located outside the DOCSIS network ber of users actually competing for bandwidth must learn the IP address of a number of CMs at any given time is much less than 100. As that share the same upstream channel. The the load increased, the collision rate ap- attacker transmits either a ping or a TCP SYN proached 90% depending on the MAP_TIME packet to the targeted CMs at a frequency setting. that depends on how many CMs are under attack. The objective of the attack is to cause When the dominant application is web brows- a large number of contention-based requests ing, the majority of data travels in the down- resulting in high collision rates and subse- stream direction. At high loads, the network quently poor network performance. This type can become packet rate bound causing ACK of attack has been identified in 802.11 net- packets accumulate in the CM upstream works where an attacker stimulates stations to queues waiting for transmission opportunities. initiate RTS/CTS exchanges leading to dra- Piggybacking is of limited benefit since matically reduce network efficiency [15]. ACKs that arrive back-to-back are sent in a concatenated frame. Concatenation can be We simulate an attack using the network helpful although it drastically increases the model illustrated in Figure 3. The configura- level of ‘ACK compression’ experienced by tion was identical to that described in Figure downstream TCP data transfers [3]. ACK 4. We set the MAP_TIME to .002 second. compression occurs when a network causes There were 100 CMs but the number of CMs TCP acknowledgement packets to ‘bunch’ at under attack was varied. The collision rate some point leading to bursty TCP send behav- increased from 48% to 68% as the number of ior which in turn contributes to higher loss CMs under attack increased from 0 to 100. rates and poor network utilization. The downstream utilization dropped from 45% to 10%. Figure 7a shows that the web We repeated the study using different parame- response times increased by a factor of 3. The ters and features of the model. The results are web response time monitor was located at a virtually identical if we turn on a downstream CM that was not under attack (i.e., test client 1 in Figure 3). In a separate experiment, we the CMs and a server and a similar FTP-like included the test client 1 CM in the attack and TFRC flow between another CM and server found that the CM was not able to complete a pair. The MAP_TIME was .002 seconds and single web response time sample. Not surpris- the number of contention request slots per ingly, a CM subject to a flooding attack ef- map was set to 12. We performed six runs, fecttively makes the access network unavail- increasing the number of CMs from 0 to 500. able to the subscriber. Figure 8 plots the TCP and TFRC connection throughputs for each run. The TFRC flow We ran the denial-of-service experiment a obtains roughly 3-7 times the bandwidth of second time with downstream service rates set the TCP flow depending on the number of to 2Mbps. The results were virtually identical CMs. When there are just the two CMs com- to the previous results. Figure 7b shows the peting (this is the 0 point on the x-axis of Fig- average web response times from test client 1 ure 8), the TFRC and TCP flows achieve a when this node was not under attack also in- throughput of 18Mbps and 6 Mbps respec- creased by almost a factor of 3. The result tively. The TFRC flow by itself (i.e., if we do suggests that a 2Mbps downstream service not run the competing TCP connection) ob- rate will not protect the network from the at- tains about 22Mbps while the TCP flow by tack. itself obtains about 12.5 Mbps. If the channel bandwidths increase, the maximum TCP FAIRNESS ISSUES throughput does not change (because TCP throughput is packet rate limited in the up- There are several fairness issues that can arise stream direction rather than limited by down- in a DOCSIS network primarily caused by stream bandwidth). The TFRC flow does not upstream packet rate limitations. The first have this limitation and can consume higher issue is that DOCSIS exhibits bias against downstream bandwidths. TCP connections running over paths with 6 x 10 tcp/tfrc throughput versus load small MTU sizes. If two CMs are each trans- 18 porting data from separate but identically con- 16 figured TCP connections with the exception 14 that one connection has a negotiated MSS of 12 512 bytes and the other connection uses an 10 MSS of 1492 bytes, the connection that gen- 8 Tfrc Goodput (bps) erates the larger packets will consume more 6 bandwidth than the other connection. 4 Tcp 2

0 A second issue, applicable to the downstream 0 100 200 300 400 500 600 direction, involves streaming video protocols, Number CMs such as TFRC, that claim to be TCP-friendly. Figure 8. TCP and TFRC throughput Because DOCSIS systems can be packet rate bound, rate-based protocols such as TFRC BANDWIDTH MANAGEMENT that do not require an ACK stream to clock new data can consume larger amounts of In our current research, we are exploring bandwidth than comparable TCP connections. bandwidth management to address these is- sues. One component of our work is to de- To demonstrate this second issue, we modi- velop a protocol aware scheduling algorithm fied the previous web scenario experiment by that predicts future CM upstream bandwidth adding an FTP-like TCP flow between one of needs and provides unsolicited grants. A sec- ond component is to develop bandwidth man- by a subscriber. A node attached to a CM agement algorithms that manage bandwidth node periodically sends a request to an HTTP based on a particular policy or service. For server for a 20Kbyte object. The time taken example, as an alternative to a pay-per-use for this download is monitored periodically policy, a provider might desire a policy and is averaged over a time-scale defined by where subscribers that consume large amounts the parameter WRTM (WRT monitor inter- of bandwidth in either the upstream or down- val). We assume that when the average of the stream directions are ‘punished’ by being WRT samples over a WRTM time period ap- placed in a state of reduced service rates for a proaches 1 second users will perceive poor given time period. quality. Once this situation is detected, the algorithm identifies the heavy-hitters that are We have prototyped such an algorithm in our contributing to congestion in the network. simulation model. The motivation for the al- gorithm is that future cable services will offer Identifying heavy-hitters in the cable network much higher service rates, possibly on the The algorithm maintains the average band- order of tens of megabits per second. To width rate (ABR) of all active users. The time manage fairness issues or to facilitate new interval over which the rate is averaged is service options, dynamic bandwidth manage- defined by the parameter TAVG. An active ment is required. The objective of our algo- user is a user whose ABR is not zero over a rithm is to prevent the large number of well TAVG amount of time. Based on the maxi- behaved subscribers from adverse affects mum channel capacity and the number of ac- caused by a few high bandwidth users (re- tive users present, a fair bandwidth rate (FBR) ferred to as ‘heavy-hitters’). The algorithm of each user is calculated using the following has three components: detecting poor quality equation: of service observed by normal user, identify- ing the heavy-hitters and regulating the heavy FBR = (Maximum channel capacity)/(Number of users to solve the problem. Active users).

The algorithm runs at the CMTS and does not The number of active users present in the require any changes at the CM nodes. The network is based on samples averaged over a algorithm can be used on the downstream period of time defined by the parameter TNUS channel or the upstream channel (or both). (time for number users sample). Any user We have applied the algorithm to manage whose ABR is above a threshold based on the downstream bandwidth. The majority of sub- FBR is considered to be a heavy-hitter. The scribers are well behaved in the sense that threshold value is represented by the parame- they consume a reasonable amount of band- ter THUSR . width over large time periods. A few sub- scribers are not well behaved (i.e., the heavy- The timescale parameters, TAVG and TNUS, hitters) and consume a disproportionate allows a cable service provider to implement amount of bandwidth over large time scales. different policies. For example, a small To simplify the discussion, we assume that TAVG on the order of minutes, can be used to there is one user per CM. ensure that TFRC flows consume a fair share of bandwidth. A cable service provider might Detecting poor quality of service observed by want to detect users who operate servers (e.g., normal user peer-to-peer or web servers). This can be han- We use the WRT metric described earlier to dled by setting the TAVG to days. What characterize the quality of service perceived makes the algorithm unique is the fact that a heavy-hitter is not punished unless it is im- algorithm detects available bandwidth, it allo- pacting other users. The extent of the pun- cates more bandwidth to the heavy-hitters. If ishment is determined by algorithm parame- there are no other users, the fair share allo- ters. cated to the heavy-hitters will consume all available bandwidth. More likely there will Regulating the heavy users be other users in which case the FBR will Once a heavy-hitter has been identified, the limit the heavy-hitters but not affect well be- next step is to regulate it to improve network haved users. performance. We implemented the policy that heavy-hitters never get more than the fair share of the bandwidth. The rate regulation continues for a configurable period of time (TREG). The rationale for ‘punishing’ the heavy-hitters by limiting their bandwidth to the fair share is to ensure that they no longer impact well-behaved users. Since we only consider the number of active users in calcu- lating the fair share, it is possible that the channel might be under-utilized. For instance, assume there are 100 users using 30Kbps Figure 9. Bandwidth management algorithm bandwidth. The fair share will be 300Kbps for a 30 Mbps channel. While being punished, a RELATED WORK heavy-hitter can consume a maximum of 300Kbps even though additional bandwidth While the intent of the IEEE’s 802.14 effort might be available. was to provide ATM services over a hybrid fiber coaxial (HFC) medium, the operation Simulation verification of the MAC layer is similar to that supported We demonstrate the algorithm using the simu- by DOCSIS. Therefore, prior 802.14 research lation network in Figure 3. We configured is relevant. The work in [6] found that TCP 154 CMs to generate the traffic mix described throughput over an 802.14 network is low in Figure 4. We configured 6 additional CMs primarily due to ACK compression. The au- to maliciously consume large amounts of thors propose two solutions: one involving downstream bandwidth using UDP traffic piggybacking and a second involving TCP sources. All the heavy-hitters were started rate smoothing by controlling the ACK spac- and stopped at the same time. When the simu- ing. The authors found that piggybacking can lation starts, the 100 web users are started. help reduce the burstiness associated with the The heavy-hitters start at around 3000 sec- ACK stream in certain situations. However it onds collectively generating around 35Mbps is limited in its abilities to effectively match of traffic. Figure 9 plots the aggregate down- offered load over a range of operating condi- stream bandwidth over the experiment. At tions. The author’s second solution is to con- time 3000 seconds we see the aggregate trol the TCP sending rate by measuring the bandwidth increase as the heavy-hitters start. available bandwidth and calculating an ap- The algorithm smoothly adapts subscriber propriate ACK rate and allowing the CM to rates to the penalized value. The subscriber request a periodic grant that provides suffi- will be in the penalty state for about 4 hours. cient upstream bandwidth to meet the required All web traffic stops time 13500 except for ACK rate. We distinguish our work by focus- traffic generated by the 6 heavy-hitters. As the ing on the latest DOCSIS standards (1.1 and have implemented this and found that while it 2.0) and using more realistic traffic loads. does increase the acknowledgement rate, it also increases the level of ACK compression. The observation in [7] is that an HFC network ACK reconstruction could be implemented in presents difficulties for TCP due to the asym- the CMTS to prevent the increased level of metry and due to high loss rates (possibly as ACK compression from affecting perform- high as 10-50%). Due to the problems of ance. We plan on addressing this in the future. TCP/Reno in these environments[8,9,10], the authors propose a faster than fast retransmit CONCLUSIONS operation where a TCP sender assumes that a packet is dropped when the first duplicate Using simulation we have identified several ACK is received (rather than the usual triple issues. First we saw that DOCSIS can affect duplicate ACK indication). The motivations the ACK stream in the upstream direction behind [7] are not relevant with the latest resulting in bursty downstream dynamics. DOCSIS standards as DOCSIS 2.0 provides Second, we have identified a possible DoS nearly symmetric access links with low packet vulnerability in DOCSIS. Taking advantage loss rates as long as the plant is well engi- of the inefficiency associated with upstream neered. packet transmissions, a hacker can negatively impact network performance by periodically The performance of TCP over asymmetric stimulating (e.g., by ping or TCP SYN pack- paths has been thoroughly studied [11,12,13]. ets) a number of CMs at a frequency that de- A network exhibits asymmetry with respect to pends on the number of CMs under attack. TCP performance if achieved throughput is The signature for this attack would be differ- not solely a function of the link and traffic ent than that of traditional flooding attacks as characteristics of the forward direction but in the amount of bandwidth consumed in the fact depends on the impact of the reverse di- downstream direction is low. Finally, we rection. Most of the prior work was focused illustrated that a TCP-friendly protocol turns on highly asymmetric paths with respect to out to be TCP-unfriendly in a DOCSIS envi- bandwidth where the normalized asymmetry ronment because the model of TCP behavior level (i.e., the ratio of raw bandwidths to the incorporated by TFRC fails to accurately cap- ratio of packet sizes in both directions) typi- ture how TCP performs in a DOCSIS envi- cally would be on the order of 2-4 [11]. In ronment. DOCSIS, depending on the service rate con- figuration, the level of bandwidth asymmetry We presented an algorithm that is designed to is small (or nonexistent). Instead, DOCSIS help manage user traffic in DOCSIS net- exhibits packet rate asymmetry due to low works. While the algorithm might not be ap- upstream packet rates with respect to down- propriate for todays networks that rely on low stream capacity. However the problem symp- service rates or that involves penalties for toms are similar. Various methods have been bandwidth misuse, our work is intended for proposed to alleviate the TCP over asymmet- future higher speed cable access networks that ric path problems including header compres- are likely to offer service rates on the order of sion and modified upstream queue poli- tens of Mbps. In these environments intelli- cies(drop-from-front, ACK prioritization, gent bandwidth management will be required. ACK filtering) [11,12,13,14]. Some of these ideas can be applied to DOCSIS. For exam- REFERENCES ple, a CM that supports ACK filtering could drop ‘redundant’ ACKs that are queued. We 1. Cable Television Labs Inc. , CableLabs, “Data- Over Cable Service Interface Specifications- Radio Frequency Interface Specification”, SP-RFIv2.0, avail- 9. K. Fall, S. Floyd, “Simulation-based Comparisons able at http://www.cablemodem.comspecificions/spec- of Tahoe, Reno and SACK TCP”, CCR, Vol 26, No. 3, ifications20.html. July 1996. 2. The Network Simulator. Available at : 10. J. Hoe, “Improving the Startup Behavior of a http://www-mash.cs.Berkeley.EDU/ns/. Congestion Control Scheme for TCP”, SIGCOMM 96, 3. J. Martin, N. Shrivastav, “Modeling the DOCSIS August 1996. 1.1/2.0 MAC Protocol”, Proceedings of the 2003 Inter- 11. H. Balakrishnan, et. Al., “The Effects of Asymme- national Conference on Computer Communications try on TCP Performance”, ACM/IEEE International and Networks”, Dallas TX, October 2003. Conference on Mobile Computing and Networking, 4. A. Feldmann, et. Al., “Dynamics of IP Traffic: A Sept. 1997. study of the role of variability and the impact of con- 12. T. Lakshman, U. Madhow, B. Suter, “Window- trol”, SIGCOM99. based error recovery and flow control with a slow 5. S. Saroiu, P. Gummadi, S. Gribble, “A Measure- acknowledgement channel: a study of TCP/IP perform- ment Study of Peer-to-Peer File Sharing Systems”, ance”, INFOCOM97, April 1997. Multimedia Computing and Networking (MMCN), Jan 13. V Jacobson, “Compressing TCP/IP Headers for 2002. Low-Speed Serial Links”, Feb 1990, RFC 1144. 6. R. Cohen, S. Ramanathan, “TCP for High Per- 14. L. kalampoukas, A Varma, K. Ramakrishnan, formance in Hybrid Fiber Coaxial Broad-band Access “Improving TCP Throughput over Two-Way Asym- Networks”, IEEE/ACM Transactions on Networking, metric Links: Analysis and Solutions”, SIGMETRICS Vol. 6, No. 1, February 1998. 98, June 1998. 7. O. Elloumi, et. Al., “A Simulation-based Study of 15. Saikat Ray, Jeffrey B. Carruthers, and David Star- TCP Dynamics over HFC Networks”, Computer Net- obinski, "RTS/CTS-induced congestion in ad-hoc wire- works, Vol. 32, No. 3, pp 301-317, 2000. less LANs," in IEEE Wireless Communication and 8. O. Elloumi, et. Al., “Improving Congestion Avoid- Networking Conference (WCNC), March 2003. ance Algorithms in Asymmetric Networks”, IEEE ICC 97, June 1997.

DOWNLOADABLE SECURITY

James William Fahrny Comcast Cable Communications

Abstract BACKGROUND

This paper will define a common security Conventional implementations of media architecture that overcomes some of these (e.g., video, audio, video plus audio, and the obstacles and issues described in the like) program stream delivery systems (e.g., background section above. This common cable, satellite, etc.) include a head-end hardware security platform can be used to where the media programming originates secure Broadcast Conditional Access (i.e., is encoded and compressed, groomed, systems, Video On Demand services, Digital statmuxed, and otherwise appropriately Rights Management, and the Authorized processed), a network (e.g., cable or Service Domain services in extending CA satellite) for delivery of the media into the home network. programming to the client (i.e., customer, user, buyer, etc.) location, at least one set This proposed paper would cover the top box (STB) at the client location for following topics: conversion (e.g., decryption and decompression) of the media programming • Architecture Block Diagram of the stream, and at least one respective viewing Downloadable Security System device such as a television (TV) or monitor • Description of the secure download that is connected to the STB. Alternatively, mechanism and how it can be the STB may be eliminated, and decryption secured. and decompression may be implemented in • Analysis of how this advances the receiving device. security for video and audio content • Definition on how this can be used to Conventional head-ends and STBs perform “Hardware Renewability” employ particular matching with a software download using encryption/decryption and FPGA technology. compression/decompression technologies. • Describes how the paradigm of However, there is little standardization of revocation and renewabilty are particular matching encryption/decryption modified for a better customer across media program stream delivery experience. system vendors. The encryption/decryption and compression/decompression Analysis of how the downloadable technologies in the particular conventional features can be applied to various system are fixed and often proprietary to the applications of the hardware platform vendor. Furthermore, conventional media including CAS, VOD, DRM, Trusted service processing and delivery systems Domain, Streaming and Personal typically implement security processes in Computing connection with individual implementations of point of deployment, CableCard, Smartcard, etc. systems.

Transitions to upgrades in This paper proposes a method of multi- encryption/decryption and stream security processing and distributing compression/decompression technologies digital media streams. The technology are, therefore, expensive and difficult for the comprises generating encrypted digital media program stream delivery system media streams. The method further vendors to implement. As such, customers comprises coupling a network to the head- can be left with substandard service due to end and receiving the encrypted digital the lack of standardization and the reduced media streams at the network. The competition that the lack of standardization technology yet further comprises coupling at has on innovation in media service delivery. least one receiver to the network and The lack of standardization also restricts the receiving the encrypted digital media ability of media service providers to streams at the receiver, and presenting a compete. For example, customers may have decrypted version of the encrypted digital viewing devices that could take advantage media streams using the receiver. At least of the improved technologies; however, one of the head-end and the at least one media stream delivery system upgrades may receiver comprises a security processor that be impossible, impracticable, or not may be configured to provide at least one of economically feasible for vendors using simultaneous multiple encryption and conventional approaches. A significant simultaneous multiple decryption processing level of customer dissatisfaction or vendor of the digital media streams. cost may result and the ability of media service providers to improve service and/or add new services is greatly restricted. This paper describes a system for multi- stream security processing, key management, and distributing digital media Thus, it would be desirable to have a streams, a security processor configured to system and a method for CA download and provide at least one of simultaneous reconfiguration that overcomes the multiple media transport stream decryption deficiencies of conventional approaches. and encryption processing is provided. The single chip solution described in this paper SUMMARY includes a security processor, a controller and a plurality of digital decryption engines. This paper generally describes an The digital decryption engines may be improved system and method for security selectively parallel coupled by the controller processing digital media streams. The for simultaneous operation in response to a improved system and method for security predetermined security configuration. processing media streams of the present invention may be compatible with Though this paper describes a future previously used (i.e., legacy) systems and vision Security System On a Chip (SSOC), methods using all levels of media stream the current technology widely deployed is processing and delivery service (i.e., basic done with separate physical devices. The to high-end) as well as adaptable to future Large Scale Integration (back-end) device implementations, and that is flexible, contains all codecs, transport functions, de- renewable, re-configurable, and could compression, general purpose processor, support simultaneous multiple security memory, and other subsystems. The systems and processes. Security Processor is a separate chip since it system and secure down load system to typically has on-board flash memory and install a new key management system. In numerous layers of tamper resistance and addition, the re-configurable security countermeasures to prevent hacker attacks. transport system is defined as part of the Based on current device fabrication overall technology. technologies, the SSOC is not the most cost effective solution though it can be made The following diagram shows a single more secure. In the future, there may be block diagram, which is a logical technology that enables the SSOC in a cost representation. The transport stream effective manner by including or replacing decryptors can be packaged in a separate all of the tamper systems and integrated circuit (IC) with the other set-tops countermeasures on the larger device. functions like decoders, demux, graphics engines, etc. When the system is packaged ARCHITECTURE separately, there must be methods implemented to secure communications The following diagram defines the between the transport stream decryptors and elements of a downloadable and re- the security processor in the client device. configurable security processing system. This security is not in the scope of this paper. This diagram includes the key management

Downloadable Security Processor

Video Input (Clear or Encrypted) CSA Video Output (Encrypted or Clear) AES Copy Protection-DES Transport Stream DES ECB (S-A) Random Number Generator Encryption Engine DES CBC (Moto) Configuration Logic Hardware Multiplier

Secure RAM DFAST Algorithm

ROM Firmware SHA-1/MD5 Hash

Security Processor

RSA Key Transport Stream Secure Download Digital Signatures Management Encryption Logic

Authenticated Authenticated Configuration & Firmware Key Loading Downloads Secure Flash Secure RAM

Authentication processor will request the head-end server to download a security client for the new network personality. Finally, the secure When devices are installed on a cable download process can be used to simply network for the first time, a discovery upgrade the key management methods while process must occur. The Head-end server preserving entitlements, purchases, credits would broadcast an announcement message and other important data stored in the much like a DHCP server. The client will security processor. respond to the “announce” message with

credentials to authenticate the client security SECURING THE DOWNLOAD processor. The head-end server will then

present its credentials to the client security processor so that the client device can trust The anchor of security and trust within the server and the server can trust the client this technology is in the security of the security processor. Credentials are presented download. The client should ONLY be able with digital signatures for authenticity and to download new firmware when the head- the public key is used to verify the end server commands the client to receive a credentials on the receiving device. new download. The client should not be able to force a download outside of the proper Download: Obtaining a client’s network network personality changes. If any of this is personality incorrectly design, the overall security is subject to severe compromise. To accomplish this strong security, the head-end must Once the server trusts the client security “unlock” the download ability of the client processor and the client security processor security processor. If the client is locked, the trusts the server, the security processor security processor cannot be loaded with a determines whether it needs a personality in new client. the form of a security client depending on the network information that it receives in the authentication process. If a download is In the same way, the image being sent to required, a secure key exchange occurs to the client security processor must not be setup for the transfer of an encrypted and tampered and likely has elements of data that digitally signed security image to the security should not be viewed. This leads to the use of processor in the client device. Once the client digital signatures and symmetric encryption is validated through signature verification of the image. The signature protects the and decryption, it is loaded into the security image from being modified and the processor and executed. encryption protects data elements from being viewed. This process permits a security processor to obtain its security personality (Scientific- ADVANCING CONTENT SECURITY Atlanta, Motorola, NDS, Nagra, etc) when it checks into securely the first time. This In the current systems, there are typically technology also allows a device to move no methods to upgrade the system for fear from one network to another whereby the that tampering or countermeasures would be client will determine that it has incorrect more easily installed. Therefore, most security client software loaded, and deletes it components in the security architectures are from memory. At this time, the client security constructed so that they cannot be modified. This can be great from a security view but CHANGING THE PARADIGM: leaves no ability to adapt to the changing RENEWABILITY INSTEAD OF world of content and content delivery REVOCATION systems. One of the largest problems in security The system proposed in this paper is not systems is that of revocation. Revocation is as static with the secure download very operationally unfriendly to manage and mechanism and therefore creates new is really built in a manner to not have a large abilities in support of potential business scale system revocation of keys. The single models as they are developed. Similarly, the biggest issue with revocation is that a longevity of a renewable security system that revocation event typically disables legitimate can adapt over time but remain secure customers experience. appears to be greater than conventional security methods. In this Downloadable Security system, we propose that the renewability be used in place

of revocation in all cases possible. Since this USING SECURE DOWNLOAD FOR system can securely transport data from the HARDWARE RENEWABILITY head-end server to the client security processor, keys of many types (authentication There is another unique development that and encryption) can be renewed when recently presented itself in the development compromised or periodically if desired. of this technology. Field Programmable Gate Clearly, if keys are renewed in a “live” Arrays (FPGA) have been used LSI hardware system, synchronization of the transition components where one can develop hardware must be managed using a solid time base so logic and load the logic language into an that the customer experience is completely FPGA to achieve dynamic hardware. This uninterrupted. technology has been very expensive in the past. In any case, this system provides a much better possibility with renewal since the However, recently IBM and other compromised systems are not actively research facilities have developed FPGA shutdown but is passively allowed to expire technology in 90 nanometer geometries of when the crypto period of their entitlements chips that is extremely cost effective. One end. Paying customers are then renewed in could effectively include 4,000 to 10,000 this process to a new key set so that their gates of FPGA into a security processor and experience is uninterrupted. leave it blank for future use. If a new algorithm is required because something is FUTURE APPLICATIONS compromised, the FPGA could be used to create a hardware accelerator of a new This technology was developed to focus algorithm. In this case, all of the client on a certain set of problems in the Broadcast security processors would be downloaded Conditional Access domain. However, after with hardware logic that would be loaded further review, this technology will be very into the FPGA section of the security effective if applied to On Demand security, processor to enable the new algorithm. Home Network Content distribution, Trusted Computing Platforms, Digital Rights Management, and Interactive Gaming.

In fact, this technology is so flexible that will provide the next generation of security one could deploy a product with the security for Broadcast, Video On Demand, and processor hardware and a specific security Streaming media systems. If the Large Scale application or profile. Later the security Integrated (LSI) devices are design with processor could be upgraded to add security some flexibility for the future, this system management for one of the other will have a tremendous longevity and a technologies as it is added as a service to the strong ability to counter any hacker attacks to network. steal services or clone devices in the field for signal theft. For example, a system could be deployed with broad CAS and later be upgrade to In the same way, we believe that the usage support VOD, or Home Network Content of this technology will grow with time since security with a secure upgrade in the field. we are only viewing the initial stage of this new paradigm at the present time. Applying CONCLUSIONS this technology to Home Networking, IP video delivery systems or even Digital Rights To summarize, we believe that the Secure Management will increase greatly over the Download technology described in this paper next couple of years. HIGH SPEED MULTIMEDIA HOME NETWORKING OVER POWERLINE

Haniph A. Latchman1, K. Afkhamie3, S. Katar3, R. E. Newman2, B. Mashburn3, L. Yonge3 1ECE Department, University of Florida, Gainesville FL 32611 2CISE Department, University of Florida, Gainesville, FL 32611 3Intellon Corporation, 5100 W. Silver Spring Blvd., Ocala, FL, 34482

Abstract Subscriber Lines (DSL) and Cable TV Modems. BPL has the advantage of ease of This white paper describes the unique installation with literal ‘plug and play’ and challenges associated with high speed digital greater penetration inside the home. Thus the communication over existing in-building powerline, historically used for the delivery powerlines. The solutions provided by the 14 of electrical power, now also provides a high Mbps HomePlug 1.0 protocol are described and an overview of the 200 Mbps HomePlug speed digital digital pipe to the home and a AV protocol is given. The latter protocol is ‘no new wires’ communication network optimized for multimedia voice and video inside. services, while also providing high speed data communication. Multimedia In-home Networking

INTRODUCTION While HomePlug 1.0 provides acceptable data rates and performance for data Interest in Powerline Communications communication needs in connecting multiple computers and peripherals in a LAN setting, There has been a great deal of recent higher data rates and more stringent QoS interest in leveraging the existing electrical controls are needed to support digital wiring within and connected to buildings for mulitmedia communication within the high speed digital communications [1]. In- home[8]. The HomePlug AV standard home LANs using powerline communication expected to be available in the last half of (PLC) are now a reality with products based 2005, is optimized for precisely this scenario. on the HomePlug 1.0 standard in use worldwide since 2000. [2][3]. PLC LANs A single stream of High Definition using the 14 Mbps HomePlug 1.0 chipsets, Televison (HDTV) may require about 25 provide full house coverage at typical TCP Mbps and a typical scenario may require data rates of 5-7 Mbps, and exhibit greater support for a number of simultaneous stability than competing wireless LAN multimedia streams of voice, audio and solutions [4,10]. video. Moreover multimedia applications also have latency, jitter and packet loss In addition there is curerent activity in the probability (PLP) requirements that must be deployment of Broadband Powerline (BPL) met for optimal performance (see Table 1) for Internet access [5, 6, 7]. BPL and WiFi (IEEE 802.11x) are seriously considered as two other possible offering to complement such broadband services as Digital Application Bandwidth Latency Jitter PLP specifically designed and optimized for (Mbps) (msec) (nsec) (log) Audio Visual (AV) applications and will HDTV 25 300 500 -10 provide adaptivity to satisfy relevant QoS SDTV 4 300 500 -10 requirements. It should also be noted that, DVD 6 300 500 -10 compared to the wireless in-home channel, VOIP 64 10 10000 -2 Gaming 0.1 10 N/A -6 the powerline commun-ication (PLC) Video conf. 1 75 10000 -6 channel is relatively static and thus the QoS requirements are much more easily met in the Table 1 – Typical Multimedia QoS Requirements more robust PLC environment. Although there are several existing in- The rest of the paper is structured as home communication tech-nologies that follows. Section II reviews characteristics of appear to be capable of providing the basis the powerline channel, while Section III for such multimedia communication, a gives an overview of the HomePlug 1.0 careful examination reveals several possible standard from a system perspective. Section deficiencies. For example, the popular IEEE IV provides brief descriptions of both the 802.x suite of protocols (including the PHY (Physical Layer) and the MAC emerging IEEE 802.n standard) does not (Medium Access Control) protocols of the provide complete house coverage (with a HomePlug AV specification. It describes the single access point) at adequate data rates HomePlug AV framing structures and unique and reliability to provide a robust multimedia channel adaptation mechanisms. It also solution. Although the new Ultra-Wide band presents the associated network architecture (UWB) standard will certainly have adequate that supports both Time Division Multiple bandwith, its reach will likely be confined to Access (TDMA) as well as Carrier Sense a single room rather than the entire home. Multiple Access (CSMA), with multiple The recently announced standard from the independent overlapping networks. The Multimedia over Coax Alliance (MoCA), paper concludes in Section V with some while possibly offering a solution for video observations on the efficiency and distribution between video sources and performance of HomePlug AV and players already connected to the exiting comments on further work in this area. coaxial video cabling, fails to offer whole house coverage for other applications such as II. PLC CHANNEL CHARACTERISTICS audio and VOIP, since video cabling is typically limited. Phoneline networks also Multipath Channel Effects have limited phone connections inside the home. In-building electrical wiring, designed for carrying electrical power at 50 or 60 Hz, The new 200 Mbps HomePlug AV consists of a variety of conductor types and standard from the HomePlug Alliance, on the sizes connected almost at random. The other hand, offers whole-house coverage, resulting terminal impedances vary both with with an average of 44 outlets per home (in communication signal frequenciy and with the USA). HomePlug AV will provide time as the load patterns at the consumer roughly a ten fold improvement over premises change. The net result is a multi- HomePlug 1.0, with typical TCP data rates of path effect that causes delay spread 50-70 Mbps, and thus it is able to support (averaging a few microseconds) and deep multiple simultaneous multimedia streams. notches (from 20 to 70 dB) at certain Futhermore the HomePlug AV standard is frequencies within the band used by PLC communications [9]. In North America, appropriate adaptive multi-level modulation HomePlug 1.0 uses a frequency band 4.5- schemes can be selected. 20.7 MHz, while HomePlug AV uses the band from 1.8 to 30 MHz band. Regulatory Programmable Spectral Masking: In order constraints make frequencies above 30 MHz to meet regulatory constraints and to unattractive for PLC applications. minimize mutual interference, a fixed spectral mask can be programmed such that PLC Channel Noise Issues the PLC devices do not use or cause intereference in certain specified bands. In addition to the inherent fading attenuation and phase characteristics of the Orthogonal Channel Adaptation, Modu- PLC channel, high speed communications in lation and Coding: A robust, relatively low this channel must also mitigate a plethora of data rate scheme (ROBO) featuring high time impairments and noise events which have and frequency redundancy, low order been historically a major impediment to high modulation and very powerful error coding is speed PLC. Typical noise sources are are designed to reach almost all nodes in the PLC certain types of halogen and fluorescent network. In addition, each PLC packet has a lamps, switching power supplies, brush Frame Control (FC) segment that uses a motors, and dimmer switches. Futhermore, highly reliable scheme to ensure that key the PLC channel is subject to interference parameters critical to the functioning of the from, and without spectral masking would PLC system, are reliably received by all itself adversely impact, other users of the nodes in the network. Tone Maps are used for specified spectrum, such as citizen band and high speed communication between a amateur radio. specific pair of nodes, to communicate the particular OFDM carriers and modulation Another characteristic of the PLC channel and error coding schemes to be used. Tone that has an impact on achievable data rates is maps are adjusted periodically based on on- the cyclic variation of noise with the going channel monitoring. powerline cycle. In particular, it has been found that the signal to noise ratios are much Efficient Medium Access Control Framing better in the vicinity of the zero crossings of and ARQ: PLC communication uses a highly the 50/60 Hz powerline cycle. efficient MAC/PHY framing strategy to ensure low overheads. Futhermore channel Taming the Shrew-like PLC Channel contention, reservation and backoff mechanism are optimized to maximize Several specific techniqiues are used in throughput. Also, high speed PLC features a HomePlug 1.0 and HomePlug AV to conquer carefully crafted error detection and the many hurdles posed by the PLC channel; retransmission (ARQ) strategy to ensure these are described below. reliable communication even in the unfavorable channel conditions. Orthogonal Frequency Division Multiplexing (OFDM): OFDM is ideal for the frequncey selective PLC channel since it III. HIGHLIGHTS OF HOMEPLUG 1.0 allows the division of the available spectrum into a large number of smaller, independent The 14 Mbps HomePlug 1.0 standard was flat fading channels for each of which released in 2000 by the HomePlug Powerline Alliance to provide a PLC-based in-home

LAN solution. HomePlug 1.0 stations use the SYNCP SYNCP SYNCP SYNCP SYNCP SYNCP SYNCM GI FC1 GI FC2 GI FC3 GI FC4 GI D1 SYNCM ... well known carrier sense multiple access 256 256 256 256 256 256 256 124 420 with collision avoidance (CSMA/CA) PREAMBLE DATA technique for medium sharing. This FRAME CONTROL mechanism is augmented with an enhanced back-off algorithm along with priority Figure 2 – PHY Frame OFDM Symbols resolution slots. The back-off algorithm enables the HomePlug 1.0 network to operate Preamble: The preamble is constructed at high efficicency under varying network from 7.5 special OFDM symbols without loads. The priority resolution slots enable cyclic prefixes, and is 38.4 µs in duration. four levels of strictly differentiatited QoS to This frame segment is used for traffic based on priority level. synchronization, automatic gain control and optionally for phase reference. The time HomePlug 1.0 PHY needed to detect the preamble is the Physical Carrier Sense (PCS) interval and dictates Orthogonal Frequency Division contention slot size. Multiplexing (OFDM): In HomePlug 1.0, 128 evenly spaced carriers are specified in the Frame Control: The Frame Control range 0-25 MHz. A programmable tone mask consists of 4 OFDM symbols in which all is used (in default configuration) to identify unmasked carriers are used in conjunction the 84 carriers that fall inside the 4.5-20.7 with a Turbo Product Code (TPC) for error MHz range, among which eight are correction. The four FC symbols contain 25 permanently masked to avoid conflict with information bits received with high Amateur radio bands. reliability. These bits, structured as shown in Figure 3(b), provide information for the Guard Space & OFDM Symbol Cyclic (from IFFT) Prefix Data subcarriers copy correct operation of the HomePlug 1.0

. . . protocol. (See [3] for further details). FFT/IFFT

Frequency Domain Time Domain Payload: The Payload consists of a Figure 1 – OFDM Symbol from IFFT variable number of 20- and 40-OFDM symbol blocks, protected by Reed-Solomon/ Figure 1 illustrates the generation of an Convolutional concatenated encoding. OFDM symbol from the unmasked carriers HomePlug 1.0 features fairly smooth via an IFFT process. Each OFDM symbol is adaptation from 1 to 14 Mbps; some 140 8.4 µs long, with 5.12 µs (256 samples) intermediate rates are supported by corresponding to the new OFDM symbol and combining available modulation schemes 3.28 µs being a cycling prefix obtained from and FEC coding rates. The effect of this the last 172 samples. Figure 2 shows the key smooth adaptation is seen in the stability of elements of the PHY Frame, namely the HomePlug 1.0 in good and bad channels Preamble, the Frame Control (FC) and a when compared with the larger variations variable number of Data (Payload) OFDM from IEEE 802.11 a and b. (See Figure 4) symbols. Figure 3 gives further details.. Priority Resolution (PR): The two Priority Resolution symbols (see Figure 3(a)) each consist of six OFDM symbols, lasting 30.72 µs and are used to establish four levels of priority. The PR slots are 35.84 µs long, preamble and cyclic prefix are inserted (if which takes into account the time to process needed) followed by a shaping filter to effect the 30.72 µs symbols. Like the Frame sharp band edges. At the receiver a Control, both Preamble and PR symbols synchronization block detects the presence of must be received reliably by all nodes in the a preamble signal, and the subsequent frame network, so all unmasked OFDM carriers are control and data symbols undergo receive modulated and encoded in a standard way. side processing to de-modulate data, and to de-code the bitstreams of interest. Figures 5 and 6 show the HomePlug 1.0 Transmitter and Receiver block diagrams. For ease of implementation, cost and other The Transmitter block shows the separate reasons, HomePlug 1.0 uses differential processing of Frame Control (FC) and Data modulation with only DBPSK and DQPSK bits. The data is first scrambled, then schemdes. In addition all carriers used in encoded, punctured, and interleaved before each OFDM symbol have the same being mapped according to selected tone modulation scheme. maps onto the OFDM carriers. After IFFT, a

Figure 3 - HomePlug 1.0 PHY Frame: (a) Frame Format and (b) Frame Control

802.11a Stability – High Speed Link 802.11b Stability – High Speed Link PLC Stability – High Speed Link

Powerline TCP Stability - High Speed Links 802.11b TCP Stability - High Speed Link 802.11a AP->Mobile 8 7 AP->Mobile Mobile->AP 12.00 Mobile->AP 30 6 10.00 5

25 8.00 4 3 6.00

20 Throughput(Mbps) 2

4.00 1

15 Throughput(Mbps) 0 2.00 48:59.9 49:19.9 49:40.0 50:00.0 50:20.0 50:40.1 Throughput 10 0.00 Time(M:S)

.0 .0 .4 .5 .6 .6 .6 .7 .7 5 8.4 8.6 :18 5 28 3 :48 58 08 :18 :28 8 09:08 09 09:28.1 09:38.4 09:48 09: 10:08.5 10:18.5 10: 10: 10 10: 11: 11 11 0 Time(M:S) 00:35.5 00:45.5 00:55.5 01:05.5 01:15.5 01:25.6 01:35.6 01:45.6 01:55.6 02:05.6 6 Time(M:S) 4

802.11a Stability – Low Speed Link 802.11b Stability – Low Speed Link PLC Stability – High Speed Link

802.11b TCP Stability - Low Speed Linkl HomePlug 1.0 TCP Stability - Low Speed Link 802.11a TCP Stability - Low Speed Link Mobile To AP 5 AP To Mobile 5 25.00 4.5 4 4 20.00 3.5 3 3 15.00 2.5 2 10.00 2 1

Throughput(Mbps) 1.5

Throughput(Mbps) 5.00 1 Throughput(Mbps) 0 0.00 0.5 28:17.8 28:27.8 28:37.8 28:47.8 28:57.8 29:07.8 29:17.8 29:27.8 29:37.9 0 .8 .8 .9 .9 .9 .0 .0 .1 :09 :29.9 :20.0 39:25.2 39:45.2 40:05.3 40:25.3 40:45.4 41:05.4 41:25.5 41:46.5 42:06.5 Time(M:S) 2:40.0 10:49 10:59.8 11 11:19 11 11:39 11:49.9 11:59 12:09.9 12 12:30 1 12:50 13:00.1 13:10 Time(M:S) Time(M:S) 7 9 5

Figure 4 – Adaptation and Stability Comparison or HomePlug 1.0 vs IEEE 802.11a and b

PHY Transmitter - Block Diagram

Frame Product Insert Cyclic RC Control Mapping IFFT Encoder Preamble Prefix Shaping Interleaver

Frame Control FEC

Bit Reed- Convo- Interleaver AFE Scrambler Solomon lutional Puncturing Encoder Encoder ROBO Interleaver Data FEC Power Line

Figure 5 - HomePlug 1.0 OFDM Transmitter

PHY Receiver - Block Diagram Power Line

OFDM Demodulator

Convert AFE FFT Demodulator to Polar

Synchronization Channel Detection Estimation

Data FEC Decoder

Reed- De- De- Viterbi De- Solomon Interleaver Puncture Decoder Scrambler Decoder

Frame Control Product De- Decoder Interleaver Frame Control FEC Decoder

Figure 6 - HomePlug 1.0 OFDM Receiver

HomePlug 1.0 Medium Access Control response to be sent in the form of a response delimiter consisting of the preamble HomePlug 1.0 uses Carrier Sense Multiple sequence and the Frame Control symbols. Access with Collision Avoidance (CSMA/ The Frame Control fields shown in Figure CA). Physical Carrier Sense (PCS) is 3(b) contain VCS, priority, and complemented by Virtual Carrier Sense acknowledgement information, needed for (VCS) information contained in the Frame the proper operation of the protocol. Control Field indicating whether other stations can contend for the medium or not. CIFS Figure 3(a) shows the basic frame structures RIFS Contention Window and timing involved in Medium Access FRAME TRANSMISSION SOF FRAME BODY (20-160 OFDM SYMBOLS) EOF RESPONSE P P Control. The payload is prepended with a 0 1 SUCCESSFUL CONTENTION delimiter contructed from a Preamble and Frame Transmission Frame Control as described earlier. After a Maximum Time = 1.68 milliseconds period denoted End of Frame Gap (EFG) of x Slots 1.5 µs, an End-of-Frame (EOF) delimiter is added by repeating the Preamble and Frame Backoff Procedure

Control, thus increasing the likelihood that Priority Contention Window Size 3 [8, 16, 16 32] all nodes will be synchronized and correctly 2 [8, 16, 16 32] 1 [8, 16, 32, 64] receive the important information contained 0 [8, 16, 32, 64] in the Frame Control fields. All nodes wait for a period of Response InterFrame Spacing (RIFS) of 26 µs for the Figure 7 – Basic Channel Access Mechanism

Figure 7 shows how channel contention Security and Key Management proceeds at the end of the response. If the Contention Control (CC) bit is set in the HomePlug 1.0 uses a password-based Frame Control of the response, then the node cryptography standard (PBCS) for key presently sending data at a certain priority management to effect cryptographic isolation will continue to attempt to send data of logical networks. All stations in a logical ("bursting" - only Priority CA2 and CA3 network share the same Data Encryption nodes are allowed to do this), but could be Standard (DES) key, called a Network preempted by higher priority nodes asserting Encryption Key (NEK). Encyption is enabled their priority in PR0 and PR1. Priority slot by default and cannot be disabled, but for PR0 begins after an interval of Contention proper protection, the user must select a Interframe Spacing (CIFS) from the end of unique network password. the response delimiter. Nodes seeking access to the channel must first assert priority CA0- HomePlug 1.0 Performance CA3 in PR0 and PR1. Simulations and measurement show that Nodes of the highest winning priorities HomePlug 1.0 provides typical throughputs will then contend for the medium in the of 5-7 Mbps (TCP), Full house coverage in contention slots and the node that wins the 99% of the homes tested was observed with a contention begins to transmit data. Colliding data rate of at least 1.5 Mbps. Figure 8 and losing nodes will chose new backoff shows how the theoretical MAC throughput values from the backoff window for that for HomePlug 1.0 varies with the number of priority according to the backoff schedule nodes. shown in Figure 7. Note that unlike the 802.11x backoff procedure, both colliding Futher details of the operation of the and losing nodes in the contention procedure HomePlug 1.0 protocol and the functions of may choose new backoff values, depending the various fields, such as Frame Control, on a new variable called the deferment are contained in [3]. counter, which checks how many times a particular node has lost contention. IV. OVERVIEW OF HOMEPLUG AV

HomePlug AV Bandwidth Performance vs. Number of Active Nodes

Throughput Peak data rate HomePlug AV provides an order of 9

8 magnitude throughput improvement over

7 HomePlug 1.0, while also addressing key 6 QoS issues. The bandwidth used has been 5

4 extended and subcarrier spacing reduced in 3 AV. Whereas HomePlug 1.0 uses 4.5 to 2 20.7 MHz quantized into 84 subcarriers, AV 1 operates with 1155 carriers over 1.8 to 30 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Number of Active Nodes MHz. While Homplug 1.0 in its default configuration uses 76 active carriers in its Figure 8 – MAC Throughput Vs Nodes bandwidth of operation , Homeplug AV uses

917 in its default mode.

HomePlug AV OFDM Symbol The OFDM symbol’s IFFT interval time in HomePlug AV is approximately eight Similar to the HomePlug 1.0 standard, times that of HomePlug 1.0. One advantage Orthogonal Frequency Division Multiplexing of this is that, in the basic configuration, (OFDM) is used for HomePlug AV. (5.56µs or 7.56µs guard interval) the However, various OFDM system parameters overhead due to the guard interval, used to have been updated in order to maximize mitigate intersymbol interference (ISI), is spectral mask flexibility and increase system much less in HomePlug AV. Likewise, throughput. Figure 8 shows the structure of when the system encounters a channel the HomePlug AV symbol and Table 3 gives where the delay spread is larger than the guard the values of the key PHY parameters. interval, subcarrier SNRs are not impacted as greatly due to the fact that the percentage T s of the IFFT interval affected is less. OFDM Symbol Another advantage of the longer symbol

RI GI RI time is that the OFDM symbols can be (and t T prefix are) shaped and overlapped in such a way T E that deep frequency notches can be created t=0 t=T t=T t=RI s E simply by turning carriers off, whereas HomePlug1.0 required, either turning off a Figure 8 – HomePlug AV OFDM Symbol

Table 3 – HomePlug AV OFDM Symbol Characteristics Symbol Description Time Samples Time (µs)

T IFFT Interval 3072 40.96 tprefix Cyclic Prefix Interval RI+GI 4.96+GI TE Extended Symbol Interval T+tprefix 45.92+GI (T + tprefix) RI Rolloff Interval 372 4.96

TS Symbol Period 3072+GI 40.96+GI GIFC Frame Control Guard Interval 1374 18.32

GI Data Symbol Guard Interval, generically 417, 567, 3534 5.56, 7.56, 47.19

GI417 Guard Interval, length=417 samples 417 5.56

GI567 Guard Interval, length=567 samples 567 7.56

GI3534 Guard Interval, length=3534 samples 3534 47.19 large number of carriers both in and around individual HomePlug AV carriers can be the desired notched band, or additional modulated with BPSK, QPSK, 8-QAM, 16- filtering. Figure 9 details the carrier power QAM, 64-QAM, 256-QAM, or 1024-QAM. rollof for the three guard intervals. Though This allows the system to take full it varies with guard interval, it can be seen advantage of all possible ranges of SNRs that if all carriers within approximately that a particular subcarrier could encounter. 115kHz of a desired notched band are turned Finally, in contrast to HomePlug 1.0 that off, the energy will be at least 30dB down in does not mix modulation types across the notched band. Figure 10 shows the deep carriers, HomePlug AV fully supports bit- notching achieved in HomePlug AV. loading. A mix of modulations is tailored for each channel such that each carrier communicates with the fastest modulation that the carrier's SNR can support. HomePlug AV FEC

Forward error correction (FEC) has also been improved in HomePlug AV. Whereas HomePlug 1.0 uses a concatenated code, HomePlug AV uses a state-of-the-art turbo convolutional code, allowing greater throughput for a given channel SNR, a gain Figure 9 – Notching by turning Off Carriers equivalent to about 2.5 dB. While

HomePlug 1.0 had a single ROBust mOdulation (ROBO) scheme, HomePlug -50 AV features several additional robust modes -55 of operation in which a repetition code is -60 applied as an outer code to the turbo code -65 for broadcast or for use in harsh channel -70

-75 conditions. Normalized Power -80 HomePlug AV and 1.0 Coexistence -85 -90 The HomePlug AV technology was -95 designed to be able to coexist with

0.5 1 1.5 2 2.5 3 Frequency [MHz] 7 x 10 HomePlug 1.0 nodes in a given network. HomePlug AV has the ability to send Figure 10 – Deep Notching in AV delimiters recognizable by HomePlug 1.0 HomePlug AV Carrier Modulation nodes in order to communicate protocol information regarding channel access and Carrier modulation has been improved in contention. HomePlug AV to maximize channel The major elements of the HomePlug AV throughput. HomePlug 1.0’s differential transmitter and receiver are shown in Figure modulation has been replaced in HomePlug 11. Note that in the transmitter, HomePlug AV with coherent modulation – yielding 1.0 Frame Control, HomePlug AV Frame higher carrier SNRs for a given signal Control and HomePlug AV packet body are power. Second, whereas HomePlug 1.0 generated separately, and are similarly used only DBPSK or DQPSK modulations, decoded independently at the receiver.

TX

Frame Frame Control Control Encoder Interleaver 1.0.1 Frame Control FEC

Frame Frame Control IFFT Insert Cyclic Prefix, Control Diversity Mapper (384, Preamble Window & Encoder Copier 3072) Overlap AV Frame Control FEC

AFE Turbo Scrambler Convolutional Interleaver Encoder AV Packet Body FEC

Power line RX

AFE

1.0.1 Frame Control Decoder Frame 384 Point Frame 1.0.1 Frame Control AGC Control Product FFT Control Decoder Data Out Demodulator De-interleaver

3072 Time De- Turbo FEC De- Point Demodulator AV PB Sync interleaver Decoder Scrambler FFT Data Out AV Packet Body Decoder

Frame Control Turbo FEC AV Frame Combine Decoder Control Data Copies Out AV Frame Control Decoder

Figure 10 – HomePlug AV Transmitter and Receiver Block Diagrams

HomePlug AV Medium Access incoming host packet in one or more MPDUs (MAC Protocol Data Units). The In HomePlug AV, medium access is order of magnitude increase in PHY rates primarily through Time Division Multiple achieved by HomePlug AV make this Access (TDMA), with CSMA/CA available approach very inefficient, so incoming host for bursty applications. In each network, a packets are aggregated into a stream of Central Coordinator (CCo) transmits a MAC frames, with a total of six bytes of beacon frame that contains schedule header and ICV (integrity check value) per information for the other stations. Stations MAC frame. The MAC frame stream is that source steady streams request time then segmented into fixed length blocks allocations from the CCo, and transmit in called PHY Blocks (PBs) that are the assigned regions. This avoids the independently encrypted and corrected. One overhead of contention and collision present or more PBs are sent in each MPDU, with in CSMA/CA. each PB carrying a four-byte header to allow

Framing and Segmentation correct reassembly.

In HomePlug 1.0, relatively low PHY rates made it reasonable to transmit a single SR-ARQ Error Control transmissions. The Initialization Vector (IV) is transmitted explicitly in HomePlug HomePlug AV employs Selective Repeat 1.0, whereas in HomePlug AV, it is derived Automatic Retransmission Request (SR- from frame information. ARQ). Each PB has its own 32-bit Cyclic Redundancy Check (CRC) to detect errors. QoS in HomePlug AV The receiver responds with a Selective Acknowledgement (SACK) that pinpoints To support desired delay, packet loss the PBs requiring retransmission. Only the tolerance, and jitter, HomePlug AV takes damaged PBs are retransmitted, and these several measures. As explained above, may be combined in a new MPDU with access for steady streams (such as newer PBs that are being sent for the first multimedia applications generate), is time. This approach allows full MPDUs to carefully scheduled using TDMA. be sent almost all the time, so that the fixed Allocated times reflect the latency delimiter overhead remains small relative to requirements, and provide sufficient time for the total transmission time. retransmissions as needed to meet the PLT Security and Key Management requirements of the stream. Jitter is managed by timestamping incoming data While HomePlug 1.0 uses 56-bit DES units with their target delivery time.. encryption, HomePlug AV uses 128-bit Stations execute a time synchroni-zation AES. Both use Cipher Block Chaining method to remain in tight synchronism so (CBC) to increase randomness in similar that the jitter remains below 500 ns.

Figure 12 Simulated and Calculated PHY and MAC Data Rates Vs SNR

V. COMMENTS AND CONCLUSIONS efficient and provides stringent QoS gurantees that are impossible to meet in HomePlug AV has made many significant HomePlug 1.0. improvements over the already successful HomePlug 1.0 protocol. It is much more HomePlug AV Performance VI. REFERENCES

The improved design of both PHY and 1. Yu-ju Lin , Haniph A. Latchman, MAC in HomePlug AV render it Minkyu Lee and Srinivas Katar, “Power tremendously efficient. At the PHY level, line Communication Network the data rates achieved are very near the Infrastructure For Smart Homes”, information theoretic limits. IEEE Wireless Commu-nications, Volume 9, Issue 6, Pages: 104-111, MAC framing overhead is minimized and December, 2002. the error correction and retransmission 2. HomePlug Powerline Alliance, scheme provides an excellent combination of http://www.homeplug.org [February 10, reliability and efficiency. Typical MAC 2005] efficiencies are projected to be in the 80% 3. M.K. Lee, , R. Newman, H.A. range, depending on the nature of the Latchman, S. Katar, and L. Yonge, application and the PHY rate. “HomePlug 1.0 Powerline HomePlug AV is capable of complete Communication LANs –Protocol Description and Comparative house coverage and will support multiple Performance Results”, accepted for streams of high and standard definition television, and stereophonic hi-fi music, publication in the Special Issue of the International Journal on while still supporting high speed data Communication Systems on Powerline applications. Figure 11 shows the HomePlug Communications, pages 447-473, May, AV PHY Rate and MAC throughput, as a 2003 function of signal to noise ratio. No other 4. Y. Lin, H. Latchman, S. Katar, and technology can provide such data rates with M.K. Lee*, “Theoretical and Field whole-house coverage and thus HomePlug Performance Comparison between AV is expected to provide synergistic 802.11 Wireless and Powerline HomePlug 1.0 Protocols” accepted for solutions for home entertainment equipment publication in the IEEE manufacturers and content providers. Communications Magazine with Focus Theme on Powerline Local Broadband Powerline access (BBL) is Area Networks, pages 54-63, May, also certain to benefit from the emerging set 2003. Current Technologies [February, 2005] of high speed PLC chips, with symmetric http://www.currenttechnologies.com access speeds of 20-40 Mbps or more 5. Amperion Corporation, February 2005, expected to the the home. Due regard is http://www.amperion.com/ being given to designing emerging standards 6. Ambient Corporation February 2005, so BPL and PLC LANs can co-exist. In this http://www.ambientcorp.com regard it is worthwile to note the recent 7. Baowei Ji, Archana Rao, Minkyu Lee, formation of the IEEE Technical Committee Haniph A. Latchman, Srinivas Katar, on PLC which is actively promoting PLC “Multimedia in Home Networking" - and BPL reearch, and forming relationships CITSA 2004 / ISAS 2004 with traditional academic and professional Proceedings Volume 1, Network communications organizations. Technologies, Page Nos: 397-404

8. Barnes, J.S., “A Physical Multipath 9. Minkyu Lee ,Haniph A. Latchman, Model For Powerline Channels at High Richard E. Newman, Srinivas Katar, and Frequencies”, Procceedings of the Larry Yonge, “Field performance International Symposium on comparison of IEEE 802.11b and Powerline Communication and Its HomePlug 1.0”, 27th Annual IEEE Applications, 1998, 76-89. Conference on Local Computer Networks, Pages: 598-599, 2002. HOME NETWORKING ON COAX FOR VIDEO AND MULTIMEDIA

Ladd Wardani Entropic Communications

Abstract HDTV, SDTV and DVD without degradation, and is consistent with current Home networking of multimedia, and operators’ broadcast reliability. particularly of video, is covered, including usage models, system requirements, USAGE MODEL installation and maintenance, security, and comparison of the various in-home mediums. The following figure shows the home Field characterization of the in-home coax usage model with triple play services (voice, plant is described, followed by home field video, data). This home is using coax home testing data of the technology developed by networking, however the desired usage Entropic and field tested by MoCATM. model is independent of home networking medium.

INTRODUCTION Multi-tap Splitter Splitter The ability to home network digital Splitter WiFi entertainment, including multiple video, Repeater Splitter audio and data streams has received STB-PVR STB Client significant attention and effort in recent STB Client years. This effort is driven by the desire to TV TV TV have content from DVRs, audio devices, PCs, and broadband (including FTTH) available

anytime and anywhere throughout the home. Splitter Splitter

With the major new application being sharing STB PVR Det Storage Game video to all the home’s displays, many Box Cable Modem product developers and service providers TV PC have realized that the “missing-link” to this capability is the availability of a ubiquitous, highly reliable and high throughput home Usage Model Content network. The home network must support multiple While consumers will accept lower quality and simultaneous HDTV, SDTV, audio, data, or degraded video viewing on mobile and VoIP, gaming, and other multimedia usages hand held devices, even occasional glitching, both from the broadcast network and from an blocking or “buffering please wait” will not in-home DVR or storage device. Video be tolerated for entertainment with standard content will continue to be over MPEG2 as home video devices such as TVs, flat panel well as advanced coding (MPEG4, H.264, displays, VCRs, DVRs, STBs, and media WM9, etc.). centers. This paper focuses on home networking of high quality video, defined as

Usage Model Data Flow This is the key criterion by which the home network must be judged . Each room and device may be either, or both, a source or sink of content both to and With multiple HDTV and SDTV streams from multiple simultaneous devices. requiring the vast majority of data rate and Consumers may add or move devices from the best packet error rate (PER), plus room to room, changing where sources and security, a home network that can support sinks connect. Flow within a home may multiple HDTV streams meets the main traverse the home network twice (double requirements. The addition of low jitter and hop), such as when video is sent from a point low delay for voice and gaming then covers of ingress to a DVR, and then from the DVR it. Here, then, are the essential requirements: to a display. • Coexist with existing services Usage Model Installation • Medium collocated with target devices Devices may be provided from either • No changes to wiring or splitters or retail or the service provider. While service other medium specific devices providers may choose to professionally • Full mesh, peer-to-peer network install these systems initially or perpetually, • Data rate net (MAC) > 60 Mbps they do not want to choose a home o 100 Mbps preferred networking technology that precludes a retail • PER < 1e-6, BER 1e-9 self-installation model. • Delay < 20 msec

o < 10 msec preferred REQUIREMENTS • Jitter < 1 msec

• Privacy from neighboring homes The requirements on the home network • No degradation due to neighbor’s are numerous. However, the key home networking or general requirements are reliability and ubiquity, appliances without which deployment is not feasible. There does not exist any reasonable home • No degradation due to other in-home networking solution that provides 100% networking products or general immediate penetration and ubiquity. Even appliances 100 Mbps Ethernet over cat-5 has been • No retransmissions above MAC layer shown not to be a 100% solution with issues • Plug n’ play on its medium including that rate negotiation problems can • Carry key protocols (Ethernet, 1394) drop or bounce networks to 10 Mbps, many • Support various DRM of the devices being deployed in homes can • Meet consumer price points not in reality support more than 40 to 60 • Support multiple independent Mbps of real video, consumers can add hubs networks and collision domain devices to the network, • Service provider may open the home and most devices lack QoS. The home network to CE devices or keep it networking solution that can and will be closed deployed is one that is: • Futureproof to higher data rates

• A >95% solution, with reasonable remediation for the remaining <5%. Key Requirement Drivers data rate for ATSC streams at 19 Mbps regardless of their own HDTV data rates. Data Rate Requirements Table 1. Data Rates in Mbps An often overlooked but crucial point is Ave Peak Trick that the home network must support the peak Modes data rate per stream, and that streams can SDTV 1 – 3 3 – 9 3 – 20+ peak simultaneously. The vast majority of HDTV 6 – 18 9 – 19 9 – 40+ broadcast content in the USA today is ATSC 19 19 19 – 40+ MPEG2. Measurements of streams from the Double Hop x2 x2 x2 top 4 USA service providers showed that SDTV MPEG2 broadcasts are variable bit Then, for example, a usage scenario with rate with average data rates between 2 and 3 1 HDTV stream, 1 ATSC stream, 2 SDTV Mbps, and peak rates around 9 Mbps. VOD double hops, and broadband at 10 Mbps SDTV streams tend to be constant bit rate requires allocation of 50 to 84 Mbps for play, between 4 and 5 Mbps. and 50 to 148 Mbps if all streams did trick mode at the same time. A usage scenario HDTV MPEG2 streams can carry with 3 ATSC streams and broadband at 3 requirements from the content providers that Mbps requires allocation of 60 Mbps for a minimum of around 12 Mbps is allocated. play, and 60 to 123 Mbps for trick modes. Measurements of programs in San Diego showed that non-ATSC HDTV MPEG2 None of these numbers include another streams carry average data rates between 10 typically > 10% protocol and management and 18 Mbps and peak rates between 13 and overhead from the application. Additional 19 Mbps. data rate of 30% or more of a stream can also be needed to fill IPSTB buffers at channel Advanced coding schemes may reduce changes in order to reduce channel change these average bit rates by around a factor of times. 2; however the peak rate does not reduce by as much as the average rate. Reliability Requirements

Fast forward and reverse, and other trick Digital cable programming is delivered modes, can increase the peak data rate by a with threshold PER of below 1e-6. The factor of 3 or more if continuous looking home network should have similar or better video is desired during the trick mode. If a performance so as not to degrade viewing. decimated, fast slide show looking trick When supporting UDP and 1394 protocols mode is acceptable, then data rates stay there is no request and retransmission of comparable. packets above the Link/MAC layer, so that this PER must be constantly maintained at Service providers will still need to make the MAC layer by the home network. A terrestrial or network ATSC content available home network with potential collision access to their subscribers. So even if an advanced will have a very difficult time achieving 1e-6 codec brings the service provider’s data rate PER with any significant loading of its data down for HDTV, the subscriber may rate. A fully coordinated MAC, collision sometimes home network ATSC. Service free, is practically a necessity for currently providers must allocate the home networking

reasonable home networks to achieve this home networking data rate, then something PER for the desired data rates. must be delayed and/or dropped. Prioritization, for example using 802.1p, Reliability from Appliance and Neighbor must be used to delay or drop lower priority Interference Requirements traffic (typically data traffic). Video packets can not generally be delayed or dropped and The home network should not degrade due must either have their data rate fully to other networking devices or typical supported at PER 1e-6 or better, or else the electronic devices and appliances such as video will be degraded... If there is not cordless phones, wireless laptops, wireless sufficient home networking aggregate data hot spots, vacuum cleaners, power drills, hair rate remaining to support a video stream or dryers and microwave ovens. This must be other traffic that can not accept dropped true for both the subscriber’s other devices as packets beyond a 1e-6 rate, then a bandwidth well as the neighbor’s devices. If reliability manager must inform the home owner that were subject to other devices and/or the the traffic can not be supported and provide neighbor, then not only would there be options for stopping other streams or service interruptions but they would not be services. predictable or easily diagnosable. In practical usage then, where data rate is Imagine a service provider’s installer taken up by high speed videos, the important completing a successful installation of elements to ensure a quality experience are: several home networked devices, rolling away in his truck, only to soon return when • very high aggregate data rate the subscriber experienced interruptions due • guaranteed packet delivery at PER < to an appliance that when the installer 1e-6 returned, was no longer on, and there was no • a bandwidth manager and user longer any interruption. Such a situation is interface not reliable, maintainable, or viable. Installation Requirements Quality of Service Requirements No new wires or changes of any sort to the In addition to a PER < 1e-6, the home in-home infrastructure should be the > 95% network must deliver delay and jitter rule. consistent with the services it carries. Gamers claim to be able to sense delays A PC must not be required for self- between 10 and 20 msec. Voice applications installation. The target subscribers are TV would like to see the delay budget allocated viewers and not necessarily PC users. Even to home networking to be 10 msec or less. PC households do not expect to need a PC to Jitter must be smoothed out in buffers at the install or operate their video systems. transmitter and receiver. Such buffers must be a fraction of the total allowed delay. MEDIUMS AND TECHNOLOGIES

Bandwidth Management Requirements The mediums in the home are wireless, phone line, powerline and coax. Each When the aggregate content’s data rate medium has innate characteristics, exceeds, on a peak or average basis, the advantages and disadvantages. Each of these mediums has had one or more home Real life tests of wireless “whole house” networking technologies developed for it. home networking shows that even in benign environment and optimistic scenarios, Selecting the Right Medium existing wireless solutions cannot come close to meeting the minimum data rates described For triple play, the medium selection is above. Practical rates in homes are below dominated by video, since all of the mediums 20% of the minimum required rates. can reasonably distribute the 1 Mbps received through broadband service today. The wireless medium enables very easy The broadband service home networking will pirating of a single subscription since it is usually be wireless for mobility and laptop very difficult to control wireless network use. Video home networking, however, has reach. A neighbor’s living room is often very different requirements outlined above. much closer to an access point than the Using these video requirements, the home furthest bedroom. Installers have a much mediums are evaluated below. clearer understanding of existing wired networks and will require a learning curve to Wireless for Video Home Networking? service wireless networks. The remedies for a whole wireless network can be fairly

complex. Numerous wireless solutions have been

proposed and implemented. Despite its Thus, whole home wireless home inherent attractiveness, and significant networking of video does not provide close technology advances, no reasonable wireless to a 95% solution and does not provide a solution has yet achieved the requirements reasonable remediation solution. At best, for “whole-house” home networked video. wireless home networking of video will some day be an in-room solution, with a wired Furthermore, service providers can not backbone for whole house coverage. subject their revenues to the potential of propagation and interference problems from Phone Line For Video Home Networking? existing and future other ever-expanding ISM unlicensed band wireless products and Real life experience and measurements of services. There can be no guarantee that current phone line home networking devices services will be maintainable in an is and has been available for years and shows unlicensed band due to non-controllable and that ubiquitous whole-house coverage is a non-predictable interference. As a result, fraction of the required data rates described there exists no premium video service in an above. unlicensed wireless band. Furthermore, in many homes, there is no Unfortunately, no technology can remove phone connection in the proximity of the the “un” in unlicensed. Only the FCC can. television sets, especially in the primary With requirements of 100 Mbps, and a family room location. frequency band that should support going through walls, the economics do not make Powerline for Video Home Networking? any sense for a licensed wireless home networking band to be created. Powerline connectivity seems quite attractive since it provides a wide home

coverage and is available at any location the 4 to 30 Mhz band, and block or impair where a powered CE device is present. powerline communications. Powerline home networking has been used traditionally for various lighting and other Thus, powerline home networking of low data rate control applications. For the video does not provide close to a 95% past several years higher data rate solutions solution and does not provide a reasonable have been available. They have shown the remediation solution for the problematic limitations of the medium, since home testing percentage. showed that the 95% outlet can only be relied on to provide less than 2 Mbps, even though Structured Wiring For Video Home the particular technology was capable of Networking? much higher data rates. An attractive medium for multimedia Powerline has a similar situation as the home networking is the Cat-5 structured unlicensed wireless band, but instead with wiring widely used for business local area effectively “unlicensed jammers”. networks. Cat-5 (or Cat-6) supports a Appliances such as vacuum cleaners, drills, reliable, repeatable, viable 100 Mbps or even hair dryers, light switch dimmers, and power 1000 Mbps. Many of the new homes built supplies in many devices output noise onto today, provide Cat-5 wiring with connectors the powerline that overwhelms powerline which are mostly co-located with the phone communications. These appliances can be connectors. turned on and off at any time and in different locations in the home or in the neighbors’ The main drawback of the structured homes. wiring is its relatively low residential penetration. Critical mass penetration into the As with wireless, there can be no tens of percentage of US homes is not guarantee that services will be maintainable expected for many years to come due to the since powerline is subject to largely varying, costs of outfitting existing residences. non-controllable and non-predictable interference. This variation prevents a Other drawbacks are that the Cat-5 viable installation and service. Remediation locations are not always near the home’s of powerline would require a certified television locations, and that it is easy for the electrician, which cable operator installers consumer to add data products, hubs and are not. other devices that can degrade performance of the network. Typically 6 to 8 homes in the USA share a transformer and therefore degrade each Coaxial Wiring For Video Home others performance. Powerline only has one Networking? frequency channel that all homes use. Isolating homes requires a certified Coax and Cat-5 wiring stand alone as the electrician and expensive filter isolators. reliable, repeatable, high bandwidth mediums in the home. However, coax is in around 100 Most all STB, PC and other device times as many whole homes as Cat-5, and is manuals recommend using a filtered power located at over 250 million video devices in surge protector. A significant percentage of the USA, including, TVs, flat panel displays, surge protectors greatly attenuate signals in DVRs, STBs, Media Centers, DVDs, VCRs and Cable Modems. Almost anywhere video • No degradation due to neighbor’s is being watched today in USA homes it is home networking or general via coax. Coax is already validated and used appliances today to carry many gigabits per second of • No degradation due to other in-home video, audio and data in the 5 to 860 Mhz networking products or general band, as well as the 950-2150 Mhz band for appliances satellite operators. • Supports multiple independent, non- overlapping networks Coax is a shielded medium and is not • Easily physically isolated from all subject to interference or changes due to other homes if desired appliances in the home or neighbors’ homes, • Does not have a slew of legacy and and does not have consumer adoption issues data home networking products that since all consumers expect to get their video can create reliability issues for service over coax. Coax is a contained medium that provider’s video home networking can not be easily shared with the neighbor for piracy, and can be physically isolated from Shared Mediums all neighbors with reasonable effort. Usually homes can not hear each others signals, and When a medium is a “shared medium,” total physical isolation between homes can be there are two concerns that must be dealt accomplished simply by putting a small $1 with: filter at the POE or multitap. Coax has the bandwidth to support greater than 8 non- • Preventing the neighbor from overlapping frequency channels so that snooping content or pirating service homes do not degrade each other even • Preventing neighbors from interfering without physical isolation. with each other and degrading each others service Coax supports sufficient bandwidth above 860 Mhz to enable multiple 100 Mbps If neighbors’ mediums are physically frequency channels to coexist, while phone isolated from each other this solves all issues. line, powerline and the 2.4 Ghz ISM wireless Otherwise, encryption is relied upon to band can only hope to support one such prevent snooping or pirating of service. channel best case. However, interference is only reasonably solved by creating orthogonality between Coax is the only medium that neighbors on different non-overlapping simultaneously satisfies all of the following frequency channels. requirements: Wireless, powerline and coax can all be • Exists at > 250M home video shared mediums in that the neighbor can locations potentially demodulate another neighbor’s • Ubiquitously supports whole home signal. data rates > 100 Mbps • Capacity of many gigabits per second Powerline in the USA is connected via the for future higher bandwidth transformer to typically around 6 to 8 homes, implementations whereas in Europe it can be to more than 100 • Stable. Not a time-varying quality homes. Powerline communication is link

restricted to low frequency operation multitaps. Coax can support reliable 100 (roughly 4 to 40 Mhz), due to attenuation, Mbps in a 50 Mhz bandwidth or less, and can only support one high speed channel enabling more than 8 channels to exist above at best. Physically isolating homes on 860 Mhz. powerline requires a certified electrician to Physically isolating homes on coax install expensive filters at the transformer. requires simple 860 Mhz low pass filters be Wireless can leak to many neighbors, or installed between the multitap and the POE. be purposely pointed to a neighbor with a However, since there are more channels directional antenna. The 2.4 Ghz ISM band supported than shared homes on a multitap, only supports one whole home, high speed each home can be operated on its own channel at best. Physically isolating home frequency and eliminate the need for physical for wireless is not possible. isolation filters until the neighborhood is heavily penetrated with multiple home Coax between homes is isolated by drop networking channels per home and uses more cables and multitaps. Multitaps give varying channels than are available within the total amounts of isolation between tap ports that bandwidth. Thus, in combination with varies with frequency, and can be insufficient encryption, coax supports private to fully isolate homes in the 860 – 2000 Mhz independent networks that are not degraded band. Multitaps come in 2, 4 and 8 tap by neighboring homes. versions. The majority are 2 and 4 tap

Medium Shared Channels Physical Isolation Without Method Physical Isolation Power Typically 6 to 8 1 Certified electrician Without filters, homes Line homes at a installs expensive will interfere with each transformer filters other Wireless Practically unlimited 1 Not practical Interference is practically 2.4 ISM unlimited Coax 2, 4 or 8 homes > 12 Simple $1 passive Homes can coexist on share a multitap filters between separate frequency multitap and POE channels, without filters, preventing interference.

COAXIAL HOME PLANT to room/outlet. This isolation is for two CHARACTERIZATION reasons:

All communications on in-home coax • Reduce interference from other today traverses between the POE and outlet, devices’ local oscillator (LO) leakage or from the headend to the room outlet • Maximize power transfer from POE to (downstream) and from the room outlet back outlets to the headend (upstream). There is no communication on in-home coax from With tuner LO leakages around -35 dBmV, room/outlet to room/outlet. In fact, the coax and isolation output to output in splitters at splitters used in homes are really directional around 20 dB, then the LO leakage will arrive couplers designed to isolate splitter outputs at other devices around -55 dBmV, which is and prevent signals from flowing room/outlet sufficient to have minimal or no degradation.

If true splitters were used instead of Multitap directional couplers, then the power loss through a 2-way splitter would be around 3dB NORMAL higher in the normal direction. 2-WAY CATV PATH 2:1 Splitter Direct Path Coax Echo Characterization Echo Path The following diagram shows the direction of signal flow in the normal and device to 2:1 Splitter 2:1 Splitter device directions on in-home coax. Note that SPLITTER device to device flow requires “splitter JUMPING jumping” output to output.

Node Node Node Node Measurements of homes and splitters have Device Device Device Device shown that splitter output to output isolation is frequency selective and varies between 8 and 35 dB. Echoes in the home vary between around 12 dB to > 35 dB and so cover the Multitap same attenuation range, and often have close Echo to the same attenuation, as the splitter output Path 2 to output signal path.

Echo 2:1 Splitter The normal direction signal flow in a home Path 1 creates echoes that are always delayed and attenuated. The following diagram shows that device to device signal flow creates echoes that can be more, less or equal in attenuation 2:1 Splitter 2:1 Splitter to the shortest path. Systems designed for Shortest POE to outlet echoes, such as DOCSIS and path other single carrier modulations with QAM and a decision feedback equalizer, can not operate reliably in an environment from outlet Node Node Node Node Device Device Device Device to outlet where there are regularly zero dB echoes and seemingly non-causal echoes. Two examples of characterized homes are shown in the following 2 figures. The first figure shows 3 scenarios: (i) POE connected to drop cable, (ii) POE terminated with 75 ohms, (iii) POE terminated with a reflector. Note that in the first figure the reflector case has less attenuation since the signal can go the normal path through the POE splitter and then perfectly reflect back instead of having to go output to output through the POE splitter and take the isolation attenuation. The 75 ohm

termination is very similar to when connected echoes. However, the second figure shows to the drop cable, but does not have the heavy that with or without a reflector at the POE, the notches from an echo off of the multitap. home can have echoes generated from nested splitters, not the multitap, and thus still cause From the first figure it might seem that extreme echoes, which are manifested by the installing a reflector at the POE removes the deep notches in the frequency response. need to operate in zero dB and non-causal

Attenuation VS Frequency 4283AB 4284AB 4285AB

0 -10 -20 POE Reflector P a -30 t ( -40 h d -50 B L ) o -60 s -70 s -80 POE Connected to Drop Cable -90 POE 75 ohm Terminator -100 750 780 810 840 870 900 930 960 990 1020 1050 1080 1110 1140 1170 1200 1230 1260 1290 1320 1350 1380 1410 1440 1470 1500

Fre que ncy (M Hz)

Attenuation VS Frequency 4885AC 4890AC

0

-10 POE Reflector P a -20 t ( h -30 d

B L

) -40 o s -50 s POE Connected to -60 Drop Cable -70 750 780 810 840 870 900 930 960 990 1020 1050 1080 1110 1140 1170 1200 1230 1260 1290 1320 1350 1380 1410 1440 1470

Frequency (MHz) Coax Link Budget Characterization plant must have a minimum quality in order for existing services to work. In addition to echoes that cause frequency selective attenuation, the overall average Existing digital and analog services attenuation of an outlet to outlet link is tolerate a worst case attenuation of around 25 substantial. Fortunately, the in-home cable dB POE to outlet, at the highest frequency in the plant (750 Mhz, or 860 Mhz). Attenuation beyond this results in non-operation of the Home Testing: Outlet-to-Outlet Attenuation at 1050 Mhz target devices, such as STBs, and therefore Reflector at POE Histogram and Cumulative Distribution Function 100% the home network is not required to work of X Axsis Attenuation Number of Occurences Histogram: is Greater than X Axis Probability that Attenuation CDF: either. This was corroborated with home 90% testing and is shown in the following figure. 80% 70% Max Min Worst Case 60%

Power at Power at POE to 50%

POE Device Outlet 40% dBmV Attenuation 30% Analog +15dBm -10dBmV 25 dB Video V 20% Digital +10dBm -15dBmV 25 dB 10% Video V 0% Å dBs Attenuation or DOCSI Home Testing: Outlet-to-Outlet Attenuation at 1050 Mhz S Connected to Drop Cable Histogram and Cumulative Distribution Function 100% Attenuation Axsis of X Number of Occurences Histogram: Greateris than X Axis Attenuation that Probability CDF: 90% 80% Home Testing: POE to Outlet Attenuation at 750 Mhz Histogram and Cumulative Distribution Function 70% 100%

of X Axsis Attenuation Number of Occurences Histogram: Axis than X Greater is Probability that Attenuation CDF: 60%

90% 50% 80% 40% 70% 30% 60%

50% 20%

40% 10%

27 dB 30% 0% Å dBs Attenuation 20%

10%

0% Home measurements showed that Å dBs Attenuation attenuation typically grace fully increases

frequency out to around 1500 with high which it can become very With a POE reflector the signal traverses Mhz, after attenuative and frequency selective. outlet to POE, gets reflected and traverses POE to another outlet. The attenuation is thus Home Coax Signal Environment around 50 dB at 750 Mhz or 860 Mhz. Without POE reflector the signal must go The cable operator uses 5-42 Mhz for through the splitter isolation instead of twice upstream and approximately 54-864 Mhz for through the normal attenuator direction, downstream. The 42-54 Mhz band is for increasing the attenuation 1 to 28 dB. This diplexer roll off in TVs, cable modems, and was corroborated with home testing and the STBs and can not be effectively used by the following two figures show outlet to outlet home network without an unacceptable attenuation with and without a POE reflector probability of interfering with such devices. at 1050 Mhz. Thus 5-864 Mhz can not be used and the

home networking must operate above 864 Remediation consists of either replacing the Mhz. inline amp with one that supports signaling above 864 Mhz, or else putting a diplexer Coax Device Environment bypass around the existing inline amp. Such a bypass diplexer is low cost, and allows signals The home coax plant has many existing below 864 Mhz to operate as before, while devices in it including TVs, VCRs and STBs signals above 864 Mhz are routed around the that are susceptible to interference from other inline amp resulting in a passive network signals on the coax. Even signals above 864 above 864 Mhz without high attenuation. Mhz can degrade or overload existing devices so that there is a maximum power level that ENTROPIC’S c.LINKTM SOLUTION the home network signaling above 864 Mhz can be transmitted at. Characterization of Entropic has a production chipset and over 130 existing devices showed that this system solution for home networking triple non-interfering maximum power level is play on coax. It includes a baseband specifically frequency selective and varies in controller doing MAC and Physical layers level from device to device. Double (BBIC), and an RF chip for conversion above conversion tuners are in general less 864 Mhz (RFIC). susceptible than single conversion tuners. A specific method of power control is required in order to meet the outlet to outlet link The BBIC includes adaptive and modified budget, yet not overload or degrade existing multi-tone modulation over a 50 Mhz devices. bandwidth, forward error correction, TDD burst generation and detection, mixed signal Home coax amplifiers are classified in two conversion, and an embedded processor that types: amplifiers in the drop side of the first executes the TDMA MAC in software. splitter are termed “drop amps” and amplifiers Packet delivery across the link is guaranteed in the home side of the first splitter are termed at 1e-9 BER without ever having a collision “inline amps.” Drop amps are outside of the and thus not requiring retransmissions. home network signaling path and do not directly affect the home networking signal. The RFIC includes LNA, PA, PLL for LO Inline amps typically amplify in the direction generation, quadrature up/down converter, of the devices from 50 to 864 Mhz and then and TDD controller to enable operation above roll off and can have significant attenuation 864 Mhz with up to 16 non-overlapping above 1000 Mhz. Field tests and research frequency channels. indicate that around 2% of homes have inline amplifiers. Testing of inline amplifiers and Entropic’s overall system solution provides: homes indicated that in about half of the homes with inline amps the attenuation is not • Full mesh, peer to peer networking prohibitive and stays within the bounds of between all end points of a passive home attenuation described previously. This home coaxial cable plant leaves half of the inline amp homes, or about • Multiple, independent networks 1% of homes, requiring remediation at the support via non-overlapping RF inline amplifier. channels • Physical layer data rate of 270 Mbps • MAC layer (net) data rate of 135 rate, and 2% of outlets would not receive 80 Mbps Mbps or more MAC rate. • 95% USA home MAC (net) data rate of >100 Mbps The field testing experience indicated that this puts the home networking reliability • Co-existence with MSO cable services comparable to the probability that an outlet by operating above 864 Mhz will even support existing digital and analog • Guaranteed/reserved bandwidth services. In other words, the home communications at 1e-9 BER networking solution is consistent with MSO Including isochronous traffic with • operational parameters. jitter < 200 nsec

• Asynchronous communications at 1e-9 CONCLUSIONS BER • Including 802.1p 8 level prioritization A home usage model, system requirements • Latency < 3 msec and comparison of home mediums led to the • Link layer privacy encryption similar conclusion that MSOs must run their whole to DOCSIS privacy home networked premium services video over either Cat-5 or coax. The unique coax home Home Field Testing Entropic’s Solution environment when communicating from outlet/room to outlet/room was then described Entropic has characterized and tested and field characterization was presented. This around 100 homes in southern California. environment was shown to include zero dBC and pre-causal echoes that make single carrier Entropic’s testing showed that, for coax solutions such as DOCSIS and other QAM outlets that support existing digital and analog with equalizer solutions non-reliable. services, the 95% MAC (net) data rate is more Entropic’s solution and production chipset than 100 Mbps from a physical layer rate of were described. Field testing and around 180 Mbps, and the 98 % MAC rate characterization in around 100 homes showed (excluding inline amplifiers) is more than 80 the Entropic solution meets a 98% reliability Mbps. This means that only 5% of outlets for the MSO, with reasonable remediation for would not receive 100 Mbps or more MAC the remaining few percent of outlets.

IP MULTICAST IN CABLE NETWORKS

Joe Godas1, Brian Field, PhD2, Alon Bernstein3, Sanjeev Desai3, Toerless Eckert3, and Harsh Parandekar3 1Cablevision Systems Corporation, 2Comcast Corporation, 3Cisco Systems, Inc.

Abstract 1. INTRODUCTION

IP multicast is as an integral technology in IP multicast is a bandwidth-conserving networked applications throughout the world. technology that reduces traffic by Any network application involving the simultaneously delivering a single stream of transmission of the same information to information to potentially thousands of multiple recipients can benefit from the subscribers. Multicast routing establishes a bandwidth efficiency of multicast technology. tree that connects a source with receivers. Multicast represents a key inflection point for Multicast delivery sends data across this tree the cable industry. While multicast is being towards receivers. Data is not copied at the used in cable networks today, two key new source, but rather, inside the network at technologies—the wideband protocol for a distribution branch points. Only a single copy Data Over Cable Service Interface of data is sent over links that lead to multiple Specification (DOCSIS) network and Single receivers, resulting in bandwidth gains. Source Multicast (SSM)—are expected to Multicast packets are replicated in the dramatically accelerate multicast deployment. network at the point where paths diverge by routers enabled with Protocol Independent These technologies will help operators Multicast (PIM), and other supporting dramatically incease the operational multicast protocols. Unlike broadcast, the efficiency of the Hybrid Fiber Coax (HFC) traffic is only received and processed by network, create a mechanism to accelerate the devices that are listening for it. delivery of advanced services, leapfrog recent announcements of fiber-to-the-x (FTTx) IP multicast was developed in the early deployments and service, and drive industry 1990s and was first deployed in education and agendas for years to come. research networks. About 1997, multicast was deployed on a large commercial scale when This paper is jointly authored by stock exchanges required a fast, efficient Cablevision Systems Corp., Comcast Corp., method to send market data to many and Cisco Systems, Inc. The paper describes subscribers simultaneously. For the past few multicast deployments at Cablevision and years, multicast has gained wider acceptance Comcast, highlights other multicast as enterprises and service providers have applications, and discusses key challenges. realized the benefits of the technology. The paper proposes enhancements to DOCSIS specifications that should significantly Two multicast service models are deployed increase multicast deployments in cable today: networks. • Any Source Multicast (ASM) is the original model introduced in 1990

(RFC1112) where an interested Figure 1 on the next page depicts a sample receiver of a multicast session notifies multicast-enabled network. the network via Internet Group Management Protocol (IGMP) that it This section also discusses multicast is interested in joining a specific group virtual private network (VPN) services. These associated with that multicast session. services are offered primarily by telcos, but The receiver then receives content sent are of high value and interest to cable by any source sending to this group. operators as well. The challenge for cable This model is targeted to support operators is to allocate enough spectrum and dynamic multi-source sessions like bandwidth to support these services. The conferencing and financial trading. wideband protocol for a DOCSIS network is The standard protocol set in support of the leading contender for providing this ASM is IGMPv2 or IGMPv3 for hosts capacity. This section briefly describes the to join a group and Protocol wideband technology and its relevance to Independent Multicast-Sparse Mode cable operators as they converge IP services. (PIM-SM), together with Multicast Source Discovery Protocol (MSDP), for interdomain operations and 2.1. Digital Simulcast at Comcast rendezvous point (RP) redundancy. Support for IGMPv2 and ASM is 2.1.1 In Deployment covered in DOCSIS 1.1.

• Source Specific Multicast (SSM) is a In today's broadcast video networks, more recent model in which an proprietary transport systems are used to interested receiver of a multicast deliver entire channel line-ups to each hub session specifies both the group and site. These transport systems are often the source (or sources) from which it dedicated to broadcast, both digital and would like to receive content. The analog, video delivery and are not easily or SSM model is superior for services economically extendable to other services. where sources can be well-known in advance of the multicast sessions. The By its very nature, broadcast video is a SSM model is achieved through the service well-suited to using IP multicast as a use of IGMPv3 which allows the host more efficient delivery mechanism. Comcast to specify both the group and the is in the process of moving its broadcast video sources of interest, as well as the PIM- service from a proprietary Baseband SSM which generates S, G joins in Video/Audio, IF, and DVB-ASI-based direct response to the IGMPv3 reports. delivery system onto an IP network that is architected to support and deliver all 2. SAMPLE APPLICATIONS UTILIZING Comcast-based services. IP MULTICAST

This section highlights solutions in which IP multicast is an important element of the network. The section details deployments at Cablevision and Comcast.

The IP multicast delivery of broadcast 2.1.2 Futures video works as follows. Encoding devices in digital master headends, encode one or more 2.1.2.1 National Backbone video channels into a Moving Pictures Expert Group (MPEG) stream which is carried in the Comcast is in the process of deploying a network via IP multicast. Devices at each hub backbone designed to support Comcast's site are configured by the operator to request specific service needs. This backbone will be the desired multicast content via IGMP joins. multicast-enabled and be able to deliver The network, using PIM-SM as its multicast broadcast video content to Comcast regional routing protocol, routes the multicast stream networks. Having a multicast-enabled IP from the digital master headend to edge backbone that is able to deliver broadcast device receivers located in the hub sites. video has a number of economic benefits, These edge devices could be edge QAM including the ability for the backbone to act as devices which modulate the MPEG stream for the backup origination location to the regional an RF frequency or ad insertion devices which networks for core video channels. The cost of splice ads into the MPEG stream and then re- deploying high-quality video-encoding originate the ad zone-specific content to a equipment in a backbone backup facility can new multicast group. Edge devices within the be more easily justified as its expense is ad zone would use IGMP joins to request this offset by reduced redundant encoding ad zone-specific multicast content. equipment needs in Comcast regional networks.

2.1.2.2 SSM support the real-time broadcast video service requirements. While multicast, as available today, is a useful technological solution for a number of 2.1.3.2 Redundant Sources cable service applications, there are areas in which further enhancements to multicast, multicast's interaction with the rest of network For critical channels or content, a cable routing protocols, and with devices which operator may choose to have dual origination participate in multicast, may be useful. One points in the network. If the primary facility enhancement to multicast relates to using becomes unavailable due to a catastrophic SSM instead of ASM. With today’s failure or unplanned maintenance, the content ASM /IGMPv2-based service with PIM- can be multicast from the backup location. SM and IGMPv2, the complexity of IP The operator can opt to have this backup multicast in the network is larger than stream always "on" and immediately available necessary for applications with one or few to devices in the network. This mode results (redundant) sources like DOCSIS Set-top in fast service recovery, but at the expense of Gateway (DSG) and Digital Simulcast. using more bandwidth in the network. Migrating to PIM-SSM and IGMPv3 reduces this complexity, and thus, lowers the cost of On the other hand, since losing the primary operations. The challenge to adopting this facility should be an uncommon event, upon technology lays primarily in edge device losing the primary facility, the operator may support like quadrature amplitude modulation opt to manually enable the backup feed. This (QAM) devices. reduces the amount of bandwidth needed in the network. The service recovery time, 2.1.3 Challenges however, will be greater.

Main challenges of this application are 2.1.3.3 Fast IP Unicast Convergence quality and availability. This translates into the applications and network redundancy Since multicast relies on the underlying IP design and failover times. routing infrastructure to build the multicast

distribution trees, the time to rebuild the 2.1.3.1 Better Overall Service Quality multicast tree when a failure occurs in the

network, is in part dependent on how quickly As broadcast video is the core service the unicast routing protocols re-converge. offered by Comcast, the video service Only when the unicast routing protocols have delivered via the IP transport network must be converged can PIM begin to rebuild the as good, if not better, than what is provided multicast trees. Thus, for real-time multicast via existing legacy systems. Thus, it is critical applications, it is important that the operator’s that network and edge devices are highly network design enables—and corresponding available; the encoded video content is of unicast and routing architectures—support very high quality; sufficient redundancy exists fast network re-convergence. when a hardware failure occurs to enable the video service to recover quickly. Cable operators, therefore, have a number of design decisions to make regarding the level of device and network redundancy needed to 2.2 Multicast at Cablevision (described as follows) are allowed for subscriber cable modems of that service. 2.2.1 In Deployment 2.2.1.2 VoD Server Resource Management 2.2.1.1 System and Conditional Access Telemetry Information Distribution to STBs Cablevision's video on demand (VoD) Cablevision currently uses IP multicast to resource management telemetry currently drive the conditional access (CA) and system occurs over IP multicast, with a future eye on information (SI) carousels to their set top utilizing reliable multicast for content boxes (STBs). Their first advanced STB had a distribution to disk farms. The telemetry single out-of-band tuner which acted as both messaging is constant and averages close to as an interactive and out-of-band (OOB) 184 pps @ 458kbits/sec. management interface. Cablevision originally supported DOCSIS and Digital Audio Visual The command and control servers are a Council (DAVIC) delivery mechanisms, but suite of servers that talk to clients and servers quickly adopted a DOCSIS-only approach in the VoD cluster. The asset management once multicast robustness was demonstrated. component of the system must know what Cablevision's newer STBs use DOCSIS to resources are available on each server in order carry vital SI streams. Both CA and SI will be to know if it has additional disk space and carried over into any DSG deployments that streaming capability. These status messages Cablevision evolves to in the future. are carried over a proprietary multicast messaging system (developed by Seachange).

IP multicast is delivered via PIM-SM for The basic network design consists of all applications. The system and conditional centrally located asset management and access distribution itself does not rely on command and control servers that speak to IGMP signaling from the STBs, but instead remotely located VoD disk farms that statically joins and forwards the traffic from communicate over the multicast network. the cable modem termination system (CMTS). Newer network design models and data Depending on the STB system, the packet transmission capabilities have enabled flow ratios will be approximately 50 pps @ Cablevision to centralize the VoD disk farms. 300 kbits/sec or 110 pps @ 430kbits/sec. This in turn lends itself to a reduced need to route the VoD multicasts. In terms of separation, Cablevision's high speed data (HSD) customers are shielded 2.2.1.3 Streaming Audio/Video to Cable from seeing these STB multicasts via standard Internet Customers DOCSIS cable modem filters which are established by the configuration from the This year, Cablevision is beginning a trial cable modem’s Trivial File Transfer Protocol of real-time multicast video streaming to HSD (TFTP) boot file. These filters also prohibit users. This is meant to differentiate and add unintended sources from hijacking or value to Cablevision's data service, promote disrupting multicast flows. In addition, loyalty, and reduce churn. Cablevision will Cablevision prohibits IGMP from cable start by porting selected Interactive Optimum modems in the upstream direction through (iO)—Cablevision’s Video Service— configuration on the CMTS. Only the groups content/functionality to Optimize Optimum of the streaming audio/video service Online (OOL)—Cablevision’s cable modem service—for use exclusively by subscribers to • More multicast-aware customer both iO and OOL services. Video and audio premises equipment (CPE) gear: home content will be offered. Rates for video will routers and home wireless gear must approach 500 kbps and 55 pps per stream. The continue to evolve to better support same sparse-dense mode network used for IGMP snooping, IGMP relay and STB’s SI will be used. The encoding is firewall configurations that allow Windows Media version 9 (WM9), but multicast streams to make it to alternative encodings are also being intended destinations without allowing investigated—specifically for future content users to cannibalize their own over the wideband protocol for a DOCSIS experience. For example, a wired network. client on a home router should not be able to cause a multicast flood of his 2.2.2 Futures own wireless spectrum, if no wireless clients are requesting the flow. Looking to the future, Cablevision sees the use of multicast as a delivery mechanism for • IGMPv3 and source-specific multicast push-VoD models where content is streamed (SSM) support to a group of PCs for viewing at a later time. Cablevision’s VoD libraries can be leveraged, • More bandwidth: Cablevision must in addition to third-party content providers. select their content carefully since there is a tight bandwidth budget with Beyond that, Cablevision sees their respect to the quality of the streams switched broadcast architecture incorporating they want to offer. Cablevision is IP multicast to help drive the efficient encouraged by the progress of the delivery of popular content across the video wideband protocol for a DOCSIS backbone and down the respective QAM network and feels they will be able to devices—be they traditional MPEG or IP over exploit these opportunities further DOCSIS. once larger backbones and modem contracts can be configured to handle 2.2.3 Challenges cost-effective high-bandwidth services. While Cablevision has been successful with the systems and services it has deployed 2.3 Multicast VPN Services to date, the company needs to continue to refine its network strategy and fine tune its Commercial services over DOCSIS are architectures to address ongoing changes and steadily gaining traction in the cable challenges. Some of these challenges include: environment—both in the U.S and abroad— because of strong revenue potential. One such • DOCSIS 1.1 support is a necessity so service is VPN which allows businesses to that multicast flooding does not occur connect multiple remote sites or devices over in a customer's home network. either a Layer 3 or Layer 2 VPN. Figure 2 depicts a multicast VPN service architecture. • Enhancements to DOCSIS must be made such that multicast can be reliably scheduled and assigned a priority on DOCSIS segments. For a Layer 3 VPN, the provider network MDT) for each VPN between all the is involved in the routing of traffic inside the associated mVRF-enabled PE routers. This VPN. A Layer 2 VPN provides a bridging tree is used to distribute multicast traffic to all transport mechanism for traffic between the PEs. For high-bandwidth multicast traffic remote sites belonging to a customer. While that has sparsely distributed receivers in the these services are just gaining momentum in VPN, a special MDT group called a Data- the cable world, they are quite pervasive in MDT can be formed to avoid unnecessary the telco world. In the telco VPN flooding to dormant PE routers. environment, enterprises have shown significant interest for native multicast IP multicast can also be supported in Layer support in the service provider's network. 2 VPNs via IGMP and/or PIM snooping in the Current estimates are that ten to forty percent provider's network. L2VPN services can be of VPN customers want IP multicast support provided by configuring the CMTSs for point- in their VPN service to transport traffic for to-point tunneling or for multipoint bridging. one or the other enterprise multicast Depending on the configuration, snooping application. takes place on the external Layer 2 aggregation device or on the CMTS. Based on When VPN services are offered, the snooped messages, the multicast traffic multiple system operators (MSOs) will see the can be forwarded only to those customer edge need to support IP multicast on these services. (CE) devices that are interested in that traffic, Typical enterprise multicast applications versus flooding it to all the CE devices. include NetMeeting, video conferencing, corporate communications, and finance- Security and data privacy are of primary specific applications. concern in a VPN environment. The service provider network, including the CMTS, must To support multicast over Layer 3 VPNs, be able to distinguish between multicast each VPN receives a separate multicast sessions that belong to different VPNs. On the domain with an associated multicast VPN shared cable downstream, packets belonging routing and forwarding (mVRF) table to separate VPNs must be encrypted using maintained by the provider edge (PE) router. separate BPI keys. Since group addresses used In the cable environment, the PE router can be within different VPNs can overlap, multicast a routing CMTS. The provider network builds support in VPNs can be complex without the a default multicast distribution tree (Default- right support in DOCSIS.

2.4 High-Speed IP over Cable with the 3. TECHNICAL CHALLENGES Wideband Protocol for a DOCSIS Network While there are numerous challenges Cable operators are now entering the third overall, this section concentrates on what we phase of service convergence as they consider to be the top two challenges: increasingly add IP video services to existing data and voice IP offerings. Service delivery • Issues with ASM requirements are rapidly evolving. A • Limitations in DOCSIS 1.1 significant percentage of traffic will shift from broadcast video to per-user streams as 3.1 Issues with ASM deployment of network VoD services enables consumers to move to a user-controlled Three basic issues with ASM and its “watch whenever” viewing paradigm. protocols exist:

In the short term, the transition to per-user 3.1.1 Address Assignment video streams is largely taking place in the

MPEG domain. VoD and personal video recorder (PVR) services are being delivered to In ASM, only one application can use a conventional STBs via MPEG transport group address G at a time. As long as streams. Over the longer-term, more of the multicast applications only need to run video content will be delivered via an end-to- between participants within a single end IP infrastructure—directly to televisions administrative entity, this is manageable. A and PCs in the home. Therefore, the single administrative entity can construct an infrastructure deployed must be capable of address plan of RFC1918 type IP multicast evolving into an all IP network. The wideband group addresses (from 239.0.0.0/8). But this protocol for a DOCSIS network, along with requires operational coordination which adds multicast and IP Version 6 (IPv6), are key cost. If on the other hand, multicast traffic is ingredients of this evolution. to be transported across domains—for example from a content provider onto one or The wideband protocol, which is under more cable operator or telco network—then evaluation for inclusion in pending DOCSIS coordination of IP multicast group addresses 3.0 specifications, allows cable operators to becomes an almost unsolvable problem. make the leap to IP video faster and cheaper than telco companies. The technology 3.1.2 Denial of Service Attacks supports bonding multiple channels to allow cable operators to add downstream channels, The ASM service model is prone to attacks independently of upstream channels. The by unwanted sources, because receivers do technology enables operators to leverage not specify which source(s) they want to previously deployed DOCSIS CMTSs and receive traffic from. While it is possible in a take advantage of declining prices for external walled garden network to provide additional edge QAM devices. It will allow operators to network-based access control, the operational use the same edge QAM pool for both data cost of such control rises as more and more and video services. The technology provides multicast applications are deployed in the plenty of bandwidth for multiple standard- network. definition digital and HDTV channels, IP telephony and data offerings, with a capacity of up to 640 Mbps.

3.1.3 Complexity of Provisioning and happen because only the lower order Operations 23 bits of an IP multicast address are mapped to a multicast Ethernet Unlike older multicast protocols, the PIM- address. For example, a CM SM/MSDP protocol provides efficient configured to receive traffic for group delivery of traffic and high availability. This 224.1.2.3 will accept traffic for comes at the cost of adding many protocol 239.1.2.3. This is particularly an issue elements which increase the complexity of the with VPN and SSM support. network. • Limited support for multicast

protocols: DOCSIS 1.1 does not have Amongst these are: IGMP support for IGMPv3 and SSM,

• Placement of RPs, operations, and Generic Attribute Registration troubleshooting of RPs Protocol (GARP), GARP Multicast • Operations of BSR or Auto RP Registration Protocol (GMRP) or protocols for RP redundancy Multicast Listener Discovery (MLD). • Alternatively, static configuration of • IPv6: No support for IPv6 multicast RPs and set up of MSDP-mesh groups since IPv6 has a dedicated control for anycast-RP plane for multicast. • Operations of MSDP between • PacketCable Multimedia (PCMM): administrative domains PCMM does not yet define how • Troubleshooting of PIM-SM protocol multicast is to be supported. elements such as RPT/SPT switchover • Lack of explicit tracking of multicast and register tunnel encapsulation listeners: Because of IGMP v1/v2 report suppression, the CMTS cannot 3.2 Limitations in DOCSIS track which hosts are actually listening to a given session and cannot support Current DOCSIS specifications define fast-leave for multicast sessions. several hooks for enabling multicast on the • Quality of Service: QoS for Multicast RF. These include: flows is not defined • Routed networks on the CPE side: The • Baseline Privacy Interface (BPI) network cannot easily support routers extensions that allow encryption of connected in the CPE network that run multicast sessions PIM instead of IGMP. • IGMP snooping in the cable modem • JOIN acknowledgment: There is no (CM) that is used to trigger the BPI way for a CM or CPE to make sure a exchange for multicast. multicast session is successfully activated since there is no explicit The purpose of IGMP snooping is to acknowledgment in IGMP V2. restrain multicast traffic and specify how a host can register a router to receive specific multicast traffic. These specifications leave a wide range of issues unaddressed. These are discussed below:

• Aliasing of traffic: According to RFC1112, aliasing of traffic may 4. SOLUTIONS automatically backwards compatible with IGMPv2. 4.1 SSM 4.1.2 Challenges of SSM Deployment 4.1.1 Solving ASM Issues The challenges in deploying SSM are SSM solves several problems with the ASM adoption and support of IGMPv3 with (S, G) service model: receiver reports in applications and appliances; for example, STBs, PCs or QAM • Network-wide group address devices. allocation: In SSM, the multicast group G does not need to be unique SSM mapping can be used as a transition over the network because only (S,G) strategy. In SSM mapping, the router channels need to be unique. Groups connected to receivers is seeded with the can be reused. source address belonging to groups G. While • DOS attacks: Receivers will only the receivers only send IGMPv2 reports for receive traffic from the source which groups G, the router itself adds the source was explicitly indicated in their address and then continues to use PIM-SSM. IGMPv3 joins.

• Simplified operations: In PIM-SM, a 4.2 DOCSIS 3.0 Multicast Proposal receiver host joins to a group G. The network builds a delivery tree towards an RP (the RP tree) and sources Multicasting on the RF can save bandwidth register to the RP via an encapsulation on the RF interface. Currently, however, tunnel. Then, the RP joins to the DOCSIS RFI specifications do not fully source to receive traffic from the address multicast. A DOCSIS 3.0 source and sends it down the RP tree. specification proposal has been submitted to Once the router connected to the CableLabs to address current DOCSIS receiver sees packets from a new limitations on multicast. source S arriving on this tree, it joins to this source via the Shortest Path It suggests the following framework: Tree (SPT)—also called the (S,G) tree • Multicast flows are signaled in the In contrast, with PIM-SSM, the IGMPv3 same way that unicast flows are— (S,G) report from the host allows the router through registration or DSx message connected to the receiver to bypass all the exchange. They use the same TLV initial steps involving an RP and start out unicast flows use—an admission immediately by establishing the SPT for control function to keep track of the (S,G). fact that multicast flows do not consume additional bandwidth once SSM with IGMPv3 and PIM-SSM is an the first one is established. evolutionary technology because PIM-SSM is • The multicast control plane handling is a subset of PIM-SM. Routers that support moved to the CMTS. PIM-SM also support PIM-SSM. Applications supporting SSM with IGMPv3 The list below outlines current issues with will also work in an existing PIM-SM DOCSIS multicast support, and explains how IGMPv2 network, because IGMPv3 is the multicast proposal addresses these issues: • Aliasing of traffic: Current DOCSIS trigger a multicast DSx based on the specifications support only RFC1112 PIM state machine. mapping of multicast addresses. With • JOIN acknowledgment: Currently, this mapping, two separate groups can there is no explicit response to an be mapped to the same MAC address. IGMP v2 JOIN. If the JOIN triggers a The proposal recommends setting a DSx message exchange, the DSx-RSP multicast media access control (MAC) will return specific error codes if the address from a CMTS-allocated pool multicast session cannot be of multicast MAC addresses outside of established. the RFC1112 MAC address range. The CM can later replace this locally 5. SUMMARY AND CONCLUSIONS assigned address to a standard RFC1112. IP multicast has a wide range of applications for current and future cable • Limited support for multicast operations. protocols: Current DOCSIS specifications use "IGMP snooping" to Cable-specific applications (to QAM devices detect that an IGMP was sent from the or STBs) include: CPE. By moving the IGMP control

plane processing to the CMTS, the • Digital simulcast of live TV via IP system is not limited to "snooping" multicast (e.g., Comcast) multicast and inherent problems • Switched video/TV broadcast associated with snooping. Instead, the (dynamic overprovisioning to save end point that was supposed to receive bandwidth) the multicast—the CMTS—is the one • DSG/proprietary STB system responding to it. information and crypto key • PCMM: Currently there is no PCMM distribution (e.g., Cablevision) definition on how multicast can be • VoD server resource management handled. Since the multicast proposal (e.g., Cablevision) to CableLabs treats multicast as unicast in terms of flow definition and "Enterprise" applications include: setup, then PCMM will tie seamlessly into this framework. • Reliable content/software distribution • Limited monitoring on the CMTS: An / preprovisoning with Pragmatic explicit signaling for multicast flow General Multicast (PGM) or other set up will allow for deterministic multicast transport to VoD or Web tracking of multicast users, instead of servers relying on "report suppression" • VoIP: multicast music on hold, voice • Quality of service (QoS) definitions: conferencing (Hoot & Holler) Current DOCSIS specifications have a • Enterprise corporate communications, rich set of methods to define QoS. video conferencing, corporate event However, these are tied to a specific broadcasting, and training modem. The proposal allows these • Financial applications including stock definitions to be re-used for multicast trading and market data distribution as well. • Retail including warehouse-distributed • Routed networks on the CPE side: if applications (e.g., with TIBCO PIM is running between the CMTS middleware) (typical drivers for VPN and a customer's router, the CMTS can customers asking for multicast)

DOCSIS 1.1 applications include: The wideband protocol for a DOCSIS network will expand cable operator service • Live audio/video streaming to HSD profiles in the IP/data arena. Its challenge, in customers (e.g., Cablevision) conjunction with IP multicast, is to ensure • L2/L3 VPN services: delivering improved support for several elements in IP "enterprise" multicast applications multicast (like SSM), as outlined in our

DOCSIS 3.0 proposal. Wideband Protocol for DOCSIS Network applications include: 6. REFERENCES

• Higher bandwidth applications, more [ASM] "Host Extensions for IP Multicasting", customers/content, and HDTV RFC1112, S.E. Deering. Aug. 1989 • Key to migrate cable-specific [IGMPv3] "Internet Group Management applications to IP (with an IP STB that Protocol, Version 3", RFC3376, B. Cain, supports the wideband protocol) S. Deering, et. al., Oct. 2002

[PIM-SM] "Protocol Independent Multicast - While cable operators may start with as Sparse Mode (PIM-SM) Protocol little as one application, they will likely need Specification (Revised)", draft-ietf-pim- to support multiple applications over time. sm-v2-new-11.txt, PIM-WG, Oct. 2004. This leads to the conclusion that IP multicast Notes: 1. The PIM-SM RFC2362 is will be one of the core capabilities of a cable obsolete. 2. PIM-SSM is a subset covered operator’s IP network for the foreseeable by this specification, too. future and will help operators unleash the full [SSM] H.Holbrook, B.Cain, "Source-Specific power of their HFC network and architecture. Multicast for IP", draft-ietf-ssm-arch-*.txt, But before this promise can be fulfilled, Sept. 2004 there are a number of items that must be [IGMPv3SSM] "Using IGMPv3 and MLDv2 considered and decisions to be made. The two for Source-Specific Multicast", draft- most important network technologies that holbrook-idmr-igmpv3-ssm-08.txt, H. cable operators must consider in conjunction Holbrook et. al, Oct. 1, 2004 with IP multicast are SSM and the wideband [DOCSIS 3.0 Proposal] “Multicast Proposal protocol for a DOCSIS network. for DOCSIS 3.0”, Submitted to CableLabs by Cisco Systems, Inc., Dec. 10, 2004 SSM can be deployed today. The [DOCSIS 3.0 Proposal] “Wideband Proposal challenge is to ensure it is supported in for DOCSIS 3.0”, Submitted to CableLabs applications and appliances such as STBs. by Cisco Systems, Inc., Dec. 10, 2004

IP VIDEO TRANSPORT SOLUTIONS FOR CABLE OPERATORS

Mark Davis1 and David Brown2 1S. V. Vasudevan, 2BigBand Networks, Inc.

Abstract requirements explode along with operator consolidation in the 1980s and 1990s. With content and service expansion a Needs emerged to retool the metropolitan constant in the cable industry, operators network architecture from standalone have had to adapt their plants for increasing headends to headend-hub architectures. capacity demands, while also containing Most cable operators maintained wireless costs for maximum return on investment. practices by leveraging CARS band This has driven several cycles of evolving microwave for video trunking between these transport techniques to deliver content from facilities, until fiber optic transport became centralized sources to distributed subscriber the new industry standard capable of bases. transmitting analog video over relatively long distances without sacrificing picture IP technologies have extended new quality. This optical transport method promise and seen initial implementations utilized FM and AM modulation techniques. that achieve superior economics and functionality for data and voice, and The AM super-trunk eventually emerged increasingly video content. The openness as the low-cost reliable solution and is still and flexibility of IP enables rapid advances used for many hubs today. The advantage and cable operators can now contemplate has been that signals could be processed or the economic utilization of fiber distribution modulated at the central headend and then at both regional and national levels. There distributed to hubs up to 30 miles away over are several courses to consider for more a single mode fiber. 1550nm optical widespread utilization of IP technologies for transmission technology evolved with higher the distribution of core cable programming, output lasers and EDFA amplifiers which and this paper considers their relative pushed distances up to 90 miles. A simple advantages and challenges, while low-cost optical to electrical receiver kept recommending commitment to a highly space requirements to a minimum and flexible infrastructure able to adapt to allowed retransmission to optical nodes in emerging opportunities. neighborhoods, commonly known as hybrid fiber/coax today.

VIDEO TRUNKING EVOLUTION However, system consolidation and the pressure to cut more operating costs and add Analog satellite was the first “trunking” more channels continued through the 1990s, technology used to cost effectively deliver which in turn drove a new digital transport multiple video programs over long distances technology. Systems emerged using from a single point to multiple destinations proprietary encoders integrated with of nationwide headends. But evolutionary SONET-like transport, allowing delivery drivers proceeded quickly, as cable without the picture quality challenges or operators continued to see capacity distance limitations of AM super-trunks. In addition, small decoders converted digital DWDM (Dense Wavelength Division video streams back to IF frequencies and Multiplexing) and CWDM (Course allowed the use of small low-cost IF-to-RF Wavelength Division Multiplexing) upconverters. Limitations included only 16 technologies. New standards-based SFP channels per fiber capacity and proprietary (Small Form-Factor Pluggable) optical techniques that constrained interoperability transceivers have allowed transport and edge and concentrated market power with QAM vendors to drastically cut the costs particular vendors. SONET and even ATM associated with both CWDM and now vendors did make several attempts to solve DWDM transport. In addition, GigE over these issues and were able to capture a small next generation SONET as well as IP number of sites, but primarily failed due to routing solutions costs also plummeted. the costly and heavy floor space These events opened up a host of new requirements of baseband to RF conversion standards based, low-cost, and flexible video or modulation required at all hubs. transport solutions.

Thousands of SONET-oriented terminals Pushing digital programming deeper into were deployed at distant digital hubs. The the network has improved picture quality industry then launched digital programming and reliability, lowered operating costs by to the home in the late 1990s, requiring consolidating headends, and provided expensive digital video headend and video capacity to meet ever growing programming processing equipment. AM super-trunking requirements. Catalyzed by VOD remained a compelling solution through the requirements, utilizing IP technologies for transition to MPEG video because of its cost this deep digital transport has opened new effectiveness. However, limitations on opportunities for the industry. transport distances and functionality such as program grooming and hub-based local As digital broadcast channel capacity insertion still required some use of transport continues to increase, an operator has to terminals until now. consider the economics of transitioning away from traditional 1550nm super-trunks IP TRANSPORT EMERGES or continuing to expand proprietary digital transport. With the drastic decreases in costs Recent widespread deployments of VOD outlined above, transitioning all digital (Video On-Demand) has driven new programs to MPEG-2 over IP/GigE can be requirements for transporting large achieved at similar costs to proprietary quantities of video streams and high volume digital system expansion, with significant QAM modulation at the hub or edge of the benefits from these open and highly network. This has prompted reconsideration functional standards. of the industry’s video transport techniques. Early VOD deployments relied on costly A more recent driver of digital transport ASI transport and distributed VOD servers. evolution and bandwidth management Technology vendors responded with new requirements is the introduction of digital low-cost, high-density QAM modulators simulcast. This is the digital encoding of all with integrated IP-to-MPEG-2 decoders that analog channels, along with QAM used Gigabit Ethernet as the transport layer. modulation of the traditional analog tier that These edge QAMs were deployed using a is then simultaneously, along with a decoded variety of optical connections including version of the analog tier, transmitted to subscribers. Encoding of all remaining As transport trends progress, the metro satellite delivered analog programs along concept is expanding and the benefits of with hundreds of PEG (Public access, nationwide terrestrial transport based on Educational and Government) channels is GigE, IP and optical technologies is required for digital simulcast. becoming apparent. Several operators now have substantial wide-area fiber resources, Simulcast allows operators to start the or can cost-justify lambda leases from long migration to an all-digital network and haul fiber providers and consolidate all provides multiple near-term benefits such as video, voice and data traffic onto a common use of less expensive, all-digital STBs (Set- backbone. Smaller operators are also faced Top Boxes), and improved picture and audio with all the same challenges and have quality especially to high-end televisions. started forming partnerships to build a Simulcast also drives the transition of analog common channel IP video backbone fed by commercial ad insertion to an all-DPI redundant primary headends. (Digital Program Insertion) solution. DPI servers can be centralized and upgraded to FUNDAMENTALS OF MPEG-2 AND IP provide GigE outputs, allowing DPI or VIDEO NETWORKING digital splicing anywhere in the network, including hub-based ad zones. Simulcasting MPEG-2 transport as standardized by leverages the benefits of Gigabit Ethernet ISO 13818-1 has established itself as the de- and IP transport techniques, advancing their facto protocol for the carriage of broadcast predominant establishment in cable digital television services. Even the newer networks for all media and services. advanced video codecs such as H.264 can be carried over MPEG-2 transport. A coded Industry consolidation is continuing to video or audio frame is fragmented into drive the need to further consolidate and several MPEG-2 transport packets, which convert headends to hubs in larger metro are a fixed 188 bytes in length. A 13-bit PID areas and even fiber-connected rural areas. (Packet Identifier), present in every MPEG- Many smaller headends in sparsely 2 transport packet, identifies the elementary populated areas are now connected with video and audio packet streams that leased or owned fiber for high speed data comprise a broadcast television program. By and telephony applications. using different PIDs to separately distinguish different elementary streams, an As the number of subscribers served by a entire program can be multiplexed and headend increases, operators must consider carried as an SPTS (Single Program adding a second redundant headend on the Transport Stream). Expanding this backbone in order to ensure the highest methodology, several programs, each with reliability or network availability, while their own unique video and audio PIDs, can delivering the lowest possible cost service. be broadcast as a unified multiplex, known MPEG-2 over standards-based GigE and IP as a MPTS (Multi-Program Transport allows operators to utilize MPEG aware Stream). switching platforms to perform automatic failover or improve fault tolerance at the To carry MPEG-2 transport over IP program level. networks, a further encapsulation step is required to place the MPEG-2 transport packets in an IP networking envelope. UDP (User Datagram Protocol) is most seven MPEG-2 packets are placed in a UDP commonly used to carry both broadcast and message. This number is chosen because it narrowcast MPEG-2 traffic. This has represents the maximum UDP message size advantages for real-time content like video that fits into the maximum 1,500 byte over widely known TCP (Transmission payload size that is dictated by the 802.3 Control Protocol) by eliminating the need Ethernet framing format. While larger UDP for packet-receipt acknowledgements back message sizes are possible, limiting the to the transport source. Since the 24 byte overall payload to less than 1,500 bytes UDP header represents additional guarantees that the message will not be transmission overhead, it is prudent to carry subjected to any unnecessary IP as many MPEG-2 transport packets in a fragmentation procedures that could be single UDP message as possible. Typically, asserted by lower-level protocol stacks.

Bandwidth

Reliable

Scalable

WDM Simple

Multi-vendor

Flexible

GigE Standards UDP\IP MPEG2

Figure 1. Multiple digital encapsulation techniques can be openly nested within each other providing a range of attractive benefits for transport of content and services.

MPEG-2 is sensitive to jitter, which is times/second), and are in essence real-time variation in delay in the arrival of transport snapshots of the counter in the encoder. packets. This is because the transport stream Decoders receive and extract PCRs for the carries timing information that is used by the particular program they are decoding and receiver (e.g. an STB) to faithfully decode use the values and appropriate filtering to and regenerate the baseband program of drive a VCXO (Voltage-Controlled Crystal interest. In an end-to-end digital television Oscillator) on the receive side to accurately system, a 27 MHz system clock is typically regenerate the 27Mhz clock, which is then locked to the incoming baseband video used to regenerate the baseband video stream, although it is also possible to timing. This transmission model allows a originate a stream with a free-running clock. decoder in a consumer’s home to regenerate This 27 MHz system clock drives a counter a video clock that is locked in phase to the that generates a 42-bit PCR (Program Clock encoder that originally compressed the Reference). PCRs are inserted into the program, which could be in a satellite uplink transport stream at a regular rate (at least 10 facility thousands of miles away. Minimizing PCR jitter is important to between a network sender and receiver. maintaining a robust and high-quality end- However, jitter that is induced by the UDP to-end signal. ISO 13818-1 specifies a encapsulation of MPEG-2 transport packets maximum PCR jitter of 500 ns, but this cannot be recovered with RTP timestamps, value specifically does not include jitter that and RTP adds another 12 bytes of overhead can be caused by UDP/IP encapsulation and to the overall message size. network transport by IP or other protocol. It is not uncommon to observe UDP/IP- NEW VIDEO BACKBONE encapsulated network traffic with tens of REQUIREMENTS milliseconds of PCR jitter, almost two orders of magnitude above that maximum IP capabilities are becoming increasingly limit specified by MPEG. applicable for video traffic in cable networks at an opportune juncture of both service To cope with this high network jitter, expansion opportunities and competitive most video-aware networking components threats to the industry. Customer tolerance de-jitter the stream by intelligently to network outages will quickly diminish correcting the PCR value in the transport now that the competitive landscape is packet and/or shaping the flow of the packet changing with telcos and satellite video traffic as it transits through the device. providers vying for market share. Given this Without an effective de-jittering mechanism, risk, any video network architecture design excessive PCR jitter can cause packet must have very high availability, fault deliveries that violate the buffer models detection and rapid recovery features. specified by MPEG, and more importantly can prevent the decoder from accurately Increasing channel counts along with regenerating the 27MHz clock required to more HD programming will continue to reconstruct the baseband signal. Consumer drive the need for additional capacity. televisions that encounter this type of Channel expansion must be easy to impairment will generally be unable to lock implement while minimizing costs. to the corrupted “color burst” waveform that precedes the delivery of a field of NTSC Pressure to minimize headcount expenses video – the most common symptom of this while driving high network availability, will situation is that the video picture will be require networks to remain simple to unable to render any chrominance provision, operate and easy to troubleshoot. information and becomes “black-and- Minimizing the number of appliances and white”. network elements such as small-profile but limited-purpose “pizza boxes” minimizes Another approach to de-jittering is the the amount of training and device use of RTP (Real-time Transport Protocol), dependencies. Complex routed networks can as specified by IETF RFCs 1889 and 2250. also be difficult to troubleshoot when a RTP is a popular protocol for streaming number of multi-service traffic related media applications, and has mechanisms to challenges like denial of service attacks support multi-source applications such as appear on the video network. Safeguards videoconferencing. A UDP message would and sound QoS (Quality of Service) be the data payload of an RTP packet. An practices must be implemented to avoid RTP header contains a timestamp that can these challenges. be used to recover and restore packet timing Now that the billion dollar access 1270nm. The same techniques are used for network upgrade is complete, there is DWDM networks as well. However, the financial pressure on MSOs to drive free wavelength spacing is much tighter at .4nm cash flow by limiting capital spending. This (50Ghz), .8nm (100Ghz) and 1.6nm will drive efficient use of existing (200Ghz) from 1525 to 1615nm which infrastructure like fiber and existing data drives up component cost for both lasers and transport elements. WDM technologies are passives. an attractive way to extract maximum Most optical transceivers are now returns on existing specialized network designed to comply with the SFP industry- element investments while allowing growth standard MSA (Multi Source Agreement). and graceful transition to an all IP video Ethernet switches, routers and SONET transport network. There is no need to do a multiplexers have all now incorporated SFP fork-lift upgrade of all transport devices standards. Low-cost SFP repeaters can with WDM. Upgrades can happen when provide 3R regeneration (re-amplify, service specific requirements dictate as with reshaping and retiming) to overcome loss the case of video transport expansion and dispersion limitations of long links. requirements.

The renowned flexibility of IP expands cable operator choice in how transport networks are configured and operated. A combination of business and technology drivers are positioning the industry to consider much more widespread transport implementations than the classic headend- hub metro connection with nationwide digital distribution. Several implementations Figure 2. SFP modules dramatically improve of this are available for MSO consideration. Robust and flexible infrastructure proves price-performance and flexibility of optical key to maintaining best-of-breed options networking. over time while implementing the best techniques to address current needs. SFP transceivers feed optical combiners and splitters called optical mux-demux PASSIVE WDM NETWORKS passives. These low-cost devices separate wavelengths or lambdas while avoiding high Passive WDM systems are now readily insertion losses. These devices use GRIN available for a variety of network transport (GRadient INdex) lenses or thin filters that systems. Passive WDM utilizes industry are normally packaged together to filter two standard ITU (International or more wavelengths. GRIN lenses focus Telecommunication Union) grid lasers that light through a precisely controlled radial are tuned to a specific wavelength. CWDM variation of the lens material’s index of typically uses bands S, C and L normally refraction from the optical axis to the edge with 20nm channel spacing and center of the lens. By gradually varying the index wavelengths at 1491nm, 1511nm, 1531nm, of refraction within the lens material, light 1551nm, 1571nm, 1591nm, 1611nm. rays can be smoothly and continually CWDM wavelengths can start as low as redirected towards a point of focus. This allows a GRIN lens with flat or angle Most metro networks have larger fiber polished surfaces to collimate light emitted counts and need fewer channels to transport from an optical fiber or to focus an incident a full broadcast video lineup to remote hubs. beam into an optical fiber. End faces can be A four lambda system carrying four Gbps of manufactured with an anti-reflection coating bandwidth over 100 km costs under $2,500 to avoid unwanted back reflections per GigE. Two GigE links can carry over GRIN Lens 480 standard definition video programs. Low-cost regeneration at drop sites or mid span repeaters can be deployed to overcome long distances.

WDM Filter The real-world, large metro-based Passive WDM Filters CWDM network represented in figure 4 provides digital video transport between two redundant headends and 24 digital hubs

Lambda 1 serving over 1.3 million homes. Two GigE links on a single fiber provide capacity for up 48 38.8Mbps MPTS broadcast video multiplexes. Each hub has dual route diverse Lambda 3 Lambda 2 WDM Filter links fed from two separate headends which Figure 3. GRIN assures stable and reliable prevents fiber related outages. optical networking performance through precise lens characteristics.

Figure 4. A passive CWDM network utilizing primary and secondary headend feeds and a bidirectional transport architecture for service assurance.

The biggest advantage of passive CWDM long fiber spans up to 200 km. Active is simply the lower cost. Wide channel transponders can also be equipped with spacing enables less stringent laser electrical multiplexers for aggregating performance requirements and lower cost multiple GigE feeds into a single 10 Gbps optical passives. Small, integrated SFP lambda. Active DWDM networks are best optics require less rack space and power when fiber resources are extremely limited consumption. Passive DWDM systems are and the maximum amount of bandwidth per best when more that eight Gbps are required lambda is required. and fiber resources are limited. Passive WDM networks are very simple to provision The downsides of active DWDM include and expand when required. In short, there added costs and more actives or moving are less active or moving parts to break parts to fail. which results in better availability. WDM systems are able to transmit multiple bit- GIGE OVER SONET rates and protocols, which allows the use of optimized application devices. SONET for SONET networks have been the CBR (Constant Bit Rate) T1 and DS3 workhorse of local and inter exchange traffic, Ethernet switches and routers for IP carriers for over a decade. SONET was traffic and MPEG-2 routers for GigE over IP optimized for high availability or lifeline traffic allows systems to deliver more CBR transport such as DS3 and DS1 traffic. specialized features for a longer useful life. Several layers of hardware protection ensured five nines of reliability. Today’s There are a few limitations when using next-generation SONET network elements CWDM including the lack of wideband now have integrated lower cost layer 2 amplifiers. SFP CWDM repeaters work well Ethernet functionality and GigE interfaces. when there are a limited number of These interfaces can be used in a one-way wavelengths to retransmit but can be costly cross-connect mode that can efficiently when dozens of lambdas need amplification. transport MPEG-2 MPTS over IP to distant DWDM allows the amplification of multiple hubs. Figure 5 shows how one-way drop and wavelengths with a single device. continue cross-connects for unidirectional video traffic operate while still providing a ACTIVE WDM NETWORKS protection path in the case of a catastrophic fiber cut. Active DWDM systems work similar to passive networks but have finely tuned lasers and cross-connect features. These devices are known as transponders and muxponders.

One challenge when operating passive DWDM networks is crosstalk and balancing of optical power levels. Active DWDM networks use precise transponders for managing power levels and 1+1 optical redundant switching. High performance lasers also allow transmission over ultra-

Figure 5. IP/GigE and SONET techniques can be combined for reliable and economical transport while leveraging prior investments and practices.

Figure 6. real-world network leveraging multi-use SONET rings.

OC-192 SONET terminals have had provisioned for a robust MPEG-2 over significant drops in price and size over the IP/GigE broadcast video transport link. last five years. Many operators already have SONET terminals by far have the best track IP and voice traffic traversing these rings. record for network availability performance. Excess SONET capacity can easily be Today, however, these issues have been The biggest challenge associated with addressed. All major router vendors include SONET networks is scalability to meet the QoS mechanisms, IP transport is rapidly growing increases in pure IP-based traffic. becoming a standard transport option on Many new IP-based switching and routing video equipment, and 10 Gigabit Ethernet is platforms are catching up to the high available at very competitive price points. performance track record of SONET-based As a result, MSOs can now make use of the solutions at far better price points. A recent same IP routed network for all of their SONET approach called RPR (Resilient video, voice and data services. Packet Ring) is now emerging. If prices can continue on a downward path, RPR may IP networks can use unicast (one-to-one) challenge classic IP routing with several of or multicast (one-to-many) transmission. the advantages of packet-based routing. While VoIP and data services currently running on MSO IP backbones is unicast, video services are multicast in nature. As a CONVERGED REGIONAL result, video services carried over IP AREA NETWORKS networks are typically multicast. It is critical to have a high-performance multicast- One way to design the network is using a enabled IP network in order to carry video routed IP backbone. Cable modem and VoIP services. (Voice over IP) services have been running over a routed IP network for years, while A key benefit of running video services video has traditionally been sent over a over IP routed networks is the fact that parallel network. There are a number of routing protocols (e.g., OSPF and BGP-4 for reasons that MSOs have not been sending unicast; PIM-SM and MBGP for multicast) video traffic over an IP routed network until make the network highly dynamic and now: robust. Rather than having operators • Until recently, IP routers did not statically define paths from a video feed in a have sufficiently robust QoS headend to a hub site, the IP network will mechanisms to ensure lossless, low- automatically calculate and transmit along jitter video delivery. the least-cost routed path. If there is an • Video equipment vendors only equipment failure, the routers will recently added support for IP-based dynamically discover the failure and re- video transport in their products. route around it – without manual • Video uses a tremendous amount of intervention. network bandwidth, and the Gigabit Ethernet capacity has been Today’s multicast IP networks generally traditionally insufficient to benefit use ASM (Any Source Multicast). In ASM from the convergence of multiple networks, receivers (e.g., a splicer/groomer services. in a hub site) use a protocol called IGMP (Internet Group Management Protocol) version 2 to join a multicast group. It is can be a bottleneck and potentially a single called “any source” multicast because the point of failure. IGMPv2 protocol does not specify the sender, only the group. After sending the A new approach called SSM (Source IGMP request, the network re-configures Specific Multicast, RFC 3569) eliminates itself to make that video feed available to the these challenges. With SSM, receivers must requesting device. specify both the multicast group and the IP address of the multicast source. IGMPv2 When using ASM for multicast IP, a does not support this capability, so IGMPv3 multicast routing protocol is generally – a new version of the IGMP protocol – is required to ensure that the multicast traffic is required in order to support SSM routed to all the correct network functionality. A key advantage of SSM is destinations; PIM-SM (defined in RFC that because the receiver specifies a source 2362) is typically used. In PIM-SM, all in advance, a rendezvous point is no longer traffic initially is transmitted via a router needed. This eliminates a potential designated as the RP (Rendezvous Point). bottleneck and point of failure from the Because all multicast traffic is sent via the network. The short term challenge with RP, an appropriate multicast stream can be SSM is that most existing devices do not yet found for each multicast group. The support IGMPv3. disadvantage of this approach is that the RP

Figure 7. Incorporation of IP routers is a substantial modification of cable architectures, allowing high flexibility to support various functionalities and future directions.

An advantage of moving to an IP routed packet loss do not negatively impact network is that the MSO can converge video video quality. This may increase the and other services such as data and VoIP operating expenses significantly services on the same IP backbone. With only enough to negate any benefits of a one network to manage, operations costs are converged network. reduced – and bandwidth can be shared among the various applications. Several mitigating practices can address these challenges. Running a converged network requires QoS mechanisms to be in place. Video A primary network design consideration traffic is very sensitive to jitter and packet is security. This should include location of loss, while VoIP is sensitive to latency and encryption which can be central, at the packet loss. In IP networks, the DiffServ acquisition site, or at the edge, in the RAN protocol (defined in RFC 2474) is the most (Regional Area Network). From a practical commonly used QoS mechanism. A DSCP standpoint, each RAN must have its own (DiffServ CodePoint) is a six-bit field in the DNCS (Digital Network Control System), or IP header of each packet – specifying the something similar, for handling queueing behavior for that packet. Video communications with STBs in the region, and VoIP packets will be marked as high and they already have CA (Conditional priority, and will be transmitted ahead of Access) systems in place. As such, the data packets. Using DiffServ, there can be content need not be encrypted centrally. lossless, low-latency, low-jitter transmission of video and VoIP even in the face of data However, particularly in a converged congestion. Another protocol called MPLS network, content owners – fearful of internet (Multi Protocol Label Switching) can be hackers somehow getting free, unencrypted used in conjunction with DiffServ, providing programming – may require that all content additional control over which traffic flows be encrypted. As such, IPSec or a similar over which links. The disadvantage of protocol may be used to encrypt traffic MPLS is that it can add significant between the central acquisition site and complexity to network operations. regional or metropolitan headend. This remains a challenge, since IPSec devices Detractors debate the merit of converging generally do not have sufficient performance video and other services on a single IP to encrypt such a large quantity of data. backbone, citing a number of concerns: • On a converged network, video Because a large number of headends services can be brought down by (potentially every system owned by a given denial of service attacks. MSO) rely on feeds from the central • Video programming is always-on, acquisition headend, guaranteed resiliency is and of fixed bandwidth, reducing the a must. To maintain such resiliency, several statistical gain made possible by the actions should be implemented. sharing the network with other Programming identical to that at the primary services. headend must be available from a secondary • A converged network increases the headend that is in a geographically separated complexity of network configuration, location. In a converged network, video requiring careful QoS and traffic traffic must be given bandwidth guarantees engineering to ensure that jitter and to ensure the network is fully non-blocking for video traffic. Video quality monitoring To ensure high quality, the video portion capabilities must be in place at each regional of the network must also be virtually free of or metro headend. Program-level packet loss. redundancy should be used, and quality monitored continuously on a per-program CONCLUSION basis. Passive WDM, active WDM, GigE- Ensuring video quality across a national SONET and converged RAN are all network is not easy. In general, today’s worthwhile considerations for cable cable networks may have an end-to-end operators who want to leverage IP latency of 50 ms. As MSOs move to technologies for transport of core national networks, latency will increase, programming services. This has further additional jitter will be introduced, there will advantages in to varying degrees enabling be multiple paths from source to destination. convergence with IP techniques used for In addition, there are more points at which emerging services like VOD as well as voice congestion and packet loss may occur. and data offerings. As a result, management capabilities and economics are improved. For MPEG traffic, the maximum Benefits can be extended to national fiber acceptable jitter is 500 ns. Encapsulating distribution that revolutionalize program MPEG in IP alone introduces significant sourcing practices. jitter, since up to seven MPEG packets are encapsulated in one 1,500 byte IP (over However, all of these techniques do bring Ethernet) packet, producing queueing delays their relative advantages and challenges. that introduce far more jitter than the Selecting for short-term optimization can specification allows. As a result, it is position an operator less well as needs necessary to de-jitter the traffic at the change. Key to this is building current IP receiving side. Poor de-jittering may result infrastructures on robust, flexible and in stutter in the picture, and lower video scalable platforms that are programmable quality. for ongoing implementation of the best and most recent transport techniques.

Mark Davis S. V. Vasudevan David Brown Vice President, Chief Architect and Director, Network Solutions VP Systems Engineering Product Marketing mark.davis vasu david.brown @bigbandnet.com @bigbandnet.com @bigbandnet.com

BigBand Networks, Inc. 475 Broadway Redwood City, CA 94063 650 995 5000 http://www.bigbandnet.com

MAKING BUSINESS SENSE OUT OF THE WIDEBAND PROTOCOL FOR A DOCSIS NETWORK

John T. Chapman Cisco Systems, Inc.

Abstract With sixty percent of today's HFC spectrum being used to carry less than two To date, cable operators have enjoyed an percent of its data capacity, the goal is to upper hand in the competition to deliver high- take the available HFC bandwidth and data speed services because of their network capacity and use them more efficiently. By capacity and bundling strategy. In the U.S., simply modifying the connectivity of the telcos’ attempts to counter this advantage by backbone to the plant, operators will achieve reliance on digital subscriber line (DSL) ten times, 100 times, or even 1000 times of service have been somewhat successful in today's cable data capacity and at attracting new subs. But because of data significantly lower price points than existing throughput restrictions and the carrier’s per-port broadband costs. reluctance to enter the video space, these services did not threaten cable operators’ The technology that makes this possible is advantage. However, with recent the wideband protocol for a Data Over Cable ® announcements of fiber-to-the-X (FTTX) and Service Interface Specification (DOCSIS ) metro Ethernet architectures in the U.S., and network. The technology promises to leapfrog deployments of advanced DSL, fiber networks the telco fiber strategy, dramatically alter the and true triple play services elsewhere in the communications industry competitive world, the playing field looks much more level landscape, and unlock more upside revenue and the true battle is ready to begin. The potential for the cable industry than the situation is even more intense in the Asia original specifications. Pacific, Japan, and European regions where competitive broadband services are showing a TECHNOLOGY OVERVIEW growing traction with residential customers. In order to fully understand the potential of So how can cable operators respond to these existing HFC networks, let’s first look at an threats and the telco promise of 25, 50 or example of how HFC plant capacity is used even 100 Mbps and greater broadband service to the home? today. The chart below details the current utilization of a cable network supporting 100k Fortunately for the cable industry, the HHP. answer does not lie in replicating its $80+ billion investment to add new physical capacity on top of existing networks or matching the $10 to 20 billion telcos will invest in fiber-based IP services. The answer lies in unleashing the full power of the cable industry’s existing hybrid fiber coax (HFC) networks.

An analog video channel occupies (in this number of approaches have been suggested to example) 6 MHz of RF bandwidth, but it only achieve this goal. The accompanying table carries one video signal. Also, because of the summarizes the pros and cons of these broadcast nature of the network, the 6 MHz of methods. bandwidth effectively serves a single service group: the entire 100k HHP. The bottom line is that sixty percent of the HFC network’s Data Pros Cons bandwidth currently yields less than two Throughput percent of the network data capacity. The Enhancement spectrum utilization efficiency increases in Use higher Low silicon Requires very the case of digital video, where typically up to order risk clean plant 10 or 12 video channels can be transmitted in modulation Minimum Higher carrier-to- a 6 MHz RF channel, but cable operators are (e.g., change MAC noise ratio only starting to scratch the surface of their from 64- to change necessary for investment potential with adoption of video 256-QAM, or same BER 256-QAM to on demand (VoD) and DOCSIS. 1024-QAM)

Different PHY Improved Incompatible The take away from this exercise is layer per-channel with existing shocking. With 1000 Gbps available in technology data cable modems today’s cable networks, less than two percent (OFDM, capacity Requires fork-lift of its total capacity is used today! Wavelet, etc.) upgrade

Let’s look at the way current DOCSIS Not proven technology makes use of HFC capacity. In technology in North America, DOCSIS 1.0 and 1.1— cable networks collectively known as DOCSIS 1.x—support Extensive MAC 30.34 Mbps to 42.88 Mbps (approximately 27 changes Mbps to 38 Mbps usable throughput) in a Increase cable Low risk Requires major single downstream RF channel, and 320 kbps network’s technology rebuild or operating RF upgrade to 10.24 Mbps (approximately 300 kbps to Increase bandwidth available Capital expense 9 Mbps usable throughput) in a single (e.g., from 50- upstream RF channel. DOCSIS 2.0 moved the downstream to upgrade plant 860 MHz to 50- spectrum bar higher yet, allowing symmetrical data 1000 MHz) and channel transmission by increasing the upstream raw capacity data rate to as much as 30.72 Mbps in a single Change from Increase Requires RF channel. DOCSIS 2.0’s downstream subsplit to upstream changing diplex technology though has remained the same as midsplit band spectrum RF filters in all DOCSIS 1.0 and 1.1. plan (e.g., bandwidth actives reverse Low-risk Requires With the adoption of emerging Internet spectrum 5-108 technology changing actives applications such as music and video MHz rather if diplex filters download or interactive on-line gaming, per- than 5-42 are hard-wired MHz) channel data throughput is rapidly becoming a Must balance/ bottleneck. In order for cable operators to sweep align all meet subscribers’ growing bandwidth actives after requirements, the downstream and upstream diplex filter mods data rate limits will need to be increased. A Need to make Data Pros Cons One solution, originated by the author in Throughput 2001 and under development by Cisco Enhancement Systems® Inc., is known as the wideband sure reverse protocol for a DOCSIS network. It solves amplifier bandwidth and throughput problems by modules/circuits logically bonding multiple RF channels work to 108 together to form a wideband “channel,” and MHz; if not, works equally well in the downstream or replacement upstream. required

Loss of some Using the wideband protocol, data is downstream RF striped across multiple quadrature amplitude spectrum modulation (QAM) channels, yielding a Node splits Backward Capital expense single logical channel—a wideband compatible to upgrade plant (materials and channel—the aggregate capacity of the Well- individual QAM channels. The number of understood labor) QAM channels logically bonded is Being done dynamically configurable, providing the by many flexibility to increase the aggregate channel cable capacity with simple software configuration. operators now For instance, by bonding four 256-QAM channels, one would obtain a wideband Increases channel with a data rate of 171.52 Mbps effective RF bandwidth (~152 Mbps). If downstream data were striped per across 24 256-QAM channels, the result subscriber would be a wideband channel with a data rate of 1.029 Gbps This table was adapted from “Next Gen (~912 Mbps). This approach allows operators Full-Service, All-Digital HFC Network to overcome the current DOCSIS per-channel Beyond DOCSIS 2.0”, a paper presented by downstream limit, without changes at the John Eng at the Society of Cable physical (PHY) layer. The same 64-QAM or Telecommunications Engineers’ 2004 256-QAM modulation formats used today by Conference on Emerging Technologies. All of DOCSIS 1.x/2.0 can be used for each of the the alternatives in this table have a substantial channels in the wideband bundle) and without impact on CapEx and/or OpEx. Of those touching at all the network topology. listed, splitting nodes is the only one being The following figure shows a high-level done to any extent by the cable industry. view of wideband technology overlaid on a Given the spectrum inefficiencies of one DOCSIS 1.x and 2.0 network. analog TV channel per 6 MHz of RF bandwidth, along with the maximum per- channel throughput limitation in DOCSIS 1.x and 2.0 technology, it’s clear that cable network capacity has room to grow. That growth doesn’t require increasing the available RF spectrum, but rather making better use of the spectrum.

• Grow subscriber base: Attract new customers, retain existing subscribers, and fend off competitive threats with higher speed services • Increase average revenue per user (ARPU): ability to offer higher tier services and expand service portfolio to allow operators to benefit from higher average revenue per user

Optimize CapEx Investment

• Fully exploit today’s HFC plant The wideband protocol is designed to be potential: no network upgrades are backwards-compatible with existing DOCSIS required to take advantage of the 1.x and 2.0 networks. It also delivers some of throughput increase offered by the the benefits of CableLabs’ Modular CMTS wideband protocol (M-CMTS™) architecture, such as separation • Leverage existing CMTS platforms: of media access control (MAC) and PHY, or implementations will enable existing the use of edge-QAM devices using today’s CMTS platforms to support the new technology. protocol, lowering incremental capital expense As we will see in the next section, it not • Reduce downstream port cost: only addresses the increasing data capacity leverage lower prices of QAM and throughput demands, but it does so technology and enable operators to use leveraging current cable modem termination existing edge QAM devices system (CMTS) technology and existing • Eliminate port under-utilization edge-QAM devices to provide a lower cost (stranded ports): add upstream and per port than current DOCSIS QAM downstream ports independent of one technology. In addition, we will see how cost another to accommodate traffic efficiencies are possible because additional requirements downstreams may be added independent of • Improve network efficiency: increase upstreams and existing RF spectrum can be the subs/port ratio by taking advantage used without impacting cable plant of the enhanced statistical infrastructure costs. multiplexing characteristics provided by a larger pipe ECONOMIC BENEFITS OF THE • Scale to future requirements: WIDEBAND PROTOCOL FOR A DOCSIS wideband components are designed to NETWORK meet rapidly changing subscriber

demands which in turn will protect This section outlines the economic and initial investment. business benefits of the wideband protocol for a DOCSIS network. Benefits can be effectively categorized as follows: Increase Revenue

Minimize Operational Costs aggressively attacking this position by investing in a combination of copper-based, • Backwards-compatible with DOCSIS fiber-based, and wireless technologies. 1.x/2.0: fully leverage existing DOCSIS 1.x/2.0 provisioning systems These facts point to the need for cable and operational processes to operators to increase their service offerings to simultaneously serve wideband meet real world capacity requirements and customers competitive positioning. Operators must show they can continue to meet customer needs Deploying the wideband protocol for a going forward. While DOCSIS 1.x and 2.0 DOCSIS network represents an evolution of have accomplished this to a certain point, the existing cable infrastructure, while network data throughput demands will offering a revolution in service capabilities. continue to eclipse this capacity. Wideband technology and the bonding of multiple This section examines in more detail the channels give operators the ability to far previously listed benefit categories. exceed today’s data offering and put cable networks in a decidedly favorable position. Increase Revenue: This added capacity gives operators tremendous flexibility in defining new service Grow Subscriber Base models and revenue opportunities. Operators can provide increased choice and the ultimate It’s a fact that the speed of Internet on-line experience. They can expand beyond connectivity is at the top of the purchasing established subscriber demographics to attract decision criteria for broadband subscribers. a broader set of users. Today, in many metropolitan areas around the world, consumers are already being offered The wideband protocol for a DOCSIS the following broadband choices: network offers the ability to deliver the highest service throughput (100 Mbps and

higher) and unmatched service selection. This Access Provider- Offered puts cable operators in a leading position in Technology Throughput the highly competitive broadband market. Competitive Carrier - Fiber 100 Mbps + Overbuild Increase Average Revenue Per User (ARPU) Incumbent Carrier - DSL 25 Mbps Cable Operator 8 Mbps Several analyses have shown the positive Wireless/Satellite Provider 2 Mbps effect of a tiered service offering versus a flat rate model. Cable operators are taking In Japan, YahooBB! has found tremendous advantage of consumer behavior that shows market success with its 100 Mbps up to the faster premium tier service gives them the 1 Gbps residential service, demonstrating just ability to trade-up existing customers from how beneficial it is to be the highest-speed lower tiers. The availability of a “wideband” provider. tier will amplify this trend, moving the service mix towards a higher average revenue per Within the U.S., cable operators have user (ARPU). enjoyed success in the broadband battle. Thus far, roughly two thirds of broadband At the same time, as broadband service customers use cable modems. Competitors are matures, the types of services included in the broadband portfolio will grow more complex Leverage Existing CMTS Platforms and bandwidth-intensive. The wideband protocol removes the bandwidth bottlenecks The flexibility of the wideband protocol for existing today and provides the opportunity a DOCSIS network does not stop with for cable operators to re-think their business modulation choices or channel bonding models in the context of emerging techniques. The protocol can be deployed in applications. Whether capturing a portion of parallel with DOCSIS 1.x/2.0 technology, the music download business with iTunes-like leveraging the investment made in existing portals, hosting IP-based video download CMTSs. Existing edge-QAM modulators can libraries, or rolling out multi-media rich be leveraged as well and physically connected interactive gaming services, cable operators to a wideband module in an existing CMTS. can dramatically expand their service Downstream traffic is now supported from portfolio. In the process, they can either existing CMTS line cards (DOCSIS significantly increase the average revenue 1.x/2.0) or edge-QAM modulators generated per subscriber. (wideband), and upstream traffic for both variants is supported from DOCSIS 1.x/2.0 Optimize CapEx Investment line card ports.

Fully Exploit Today’s HFC Plant Potential The figure below compares the normalized costs for a DOCSIS 1.x/2.0 CMTS with the A key objective of cable operators today is incremental costs required to upgrade the to find ways to leverage the powerful data CMTS to support the wideband protocol. capacity of their HFC networks—estimated to The chart illustrates that with first generation use less than two percent of the total available wideband technology, you can more than capacity. As seen in the previous section, double the downstream throughput, at less there are a number of alternatives available, than one fourth the cost! This is the first step but unfortunately, these are either cost- or towards fulfilling the wideband objective of labor-intensive (or both). delivering ten times the throughput at one tenth the cost. By taking advantage of flexible channel management techniques and channel bonding of up to 24 channels in the first generation of products, the wideband protocol for a DOCSIS network is able to transport packets at Gigabit Ethernet speeds using existing modulation formats—whether they are 64- or 256-QAM. This gives operators the ability to quickly take advantage of the available spectrum on the existing HFC plant, optimizing CapEx investments.

Reduce Downstream Port Cost

Current CMTS designs are optimized for synchronous traffic for multiple IP-based services. As a result, CMTS line cards are more complex than single-purpose edge- QAM devices used for uni-directional, originally formulated. This resulted in the asynchronous traffic. This complexity bears simultaneous support of traffic in upstream with it incremental infrastructure costs. and downstream directions and line cards equipped with a fixed ratio of upstream and The proposed wideband technology downstream ports. Typical calculations held separates the channel bonding functionality to a calculated ratio of one downstream per (performed by a wideband module on an four or six upstream ports. existing CMTS) and the Physical Layer adaptation to the RF plant. The latter can be The ratio, understandably, was very performed by an external edge-QAM device, conservative in the assumptions it made. In offering a significant opportunity to leverage addition, it was difficult to predict or declining costs of QAM device designs, and anticipate the development of new IP provide a graceful evolution towards a applications that dramatically challenge these Modular CMTS architecture. In advance of assumptions and prevent operators from the availability of standards-based products, optimizing port utilization. The net result of operators can plan an evolutionary option that this is that some number of ports (either saves infrastructure costs. upstream or downstream) will be under- utilized. The following chart depicts the relative costs per port for CMTS and edge-QAM The following figure shows how line cards, devices. The chart shows that edge QAM deployed with a 1:4 port domain are under- devices used in video applications hold a utilized for various traffic scenarios. Since the significant price advantage over equivalent capacity of a downstream channel is roughly CMTS port costs. A CMTS supporting the four times the capacity of an upstream wideband protocol leverages this advantage channel in this example (256 QAM in the by allowing the use of existing QAM devices, downstream, 16 QAM/3.2 MHz in the thus, lowering the cost of implementation. upstream), the ports in the one downstream – four upstream domain are 100% utilized if the downstream traffic required on the network is equal to the upstream traffic (1:1 traffic ratio line). In the cases where twice as much traffic is required for the downstream than for the upstream (2:1 traffic ratio line), the upstream ports will be 50% underutilized. Finally, if the traffic in the downstream is four times the upstream traffic (4:1 traffic ratio line), the 1:4 port ratio will translate into a 75% upstream port under-utilization.

Eliminate Port Under-Utilization (Stranded Ports)

CMTS port assignments were designed around a traffic formula based on studies of anticipated preliminary broadband services and subscriber usage when DOCSIS was The wideband protocol offers a simple, yet powerful tool: a “larger pipe”. Complex mathematical models have demonstrated that in the case of bursty traffic, such as the traffic generated by Internet users, a higher capacity transmission medium will show better statistical multiplexing characteristics since the probability of transmission collisions are reduced. Wideband offers such statistical multiplexing gain by logically bonding multiple QAM channels and increasing the total channel capacity. In other terms, this means that logically bonding multiple QAM The only way to avoid the under- channels will increase the number of utilization of upstream or downstream ports is subscribers per QAM (or per port). to be able to assign upstream and downstream ports independently of one another. Looking ahead, both higher data rate tiers, CableLabs’ M-CMTS initiative aims at as well as emerging applications such as IP- providing this benefit. The wideband based video will change the traffic technology achieves the same goal with characteristic. In this scenario, the number of today’s CMTS and edge-QAM technology, subscribers that will be able to share a single yielding better utilization of infrastructure and channel will be significantly reduced, making lower cost per subscriber. the current technology less and less profitable. The wideband protocol offers the ability to Improve Network Efficiency maintain the same level of oversubscription rates, despite increasing customer demands As a shared access medium, cable for higher throughput. networks hold an economic advantage over other access technologies due to the ability to Scale to Future Requirements oversubscribe the access plant. Not all users are going to be actively transmitting data at Wideband technology offers deployment the same time. The determination of an flexibility. Because a wideband logical appropriate oversubscription rate is a complex channel can be dynamically defined to be a effort—and dependent upon a number of bundle of two, four or any number of QAM factors such as customer usage patterns, the channels up to 24, as defined in my initial size and frequency of the traffic bursts, the proposal for DOCSIS 3.0, cable operators will nature of the content carried, the priority be able to choose, over time, the required levels for the various types of traffic, and the channel capacity according to their business ratio of peak customer traffic to offered needs. capacity. Cable operators are constantly trying to balance subscriber satisfaction metrics with On the modem side, cable modem tuner the need to maintain oversubscription as high technology is available today to allow as possible in order to have a large number of multiple QAM channels to be received subscribers sharing the same port, thus simultaneously, enabling wideband service. reducing the cost per sub. Such technology represents an important evolution because it allows multiple channels to be demodulated by a single, digital multichannel receiver chip, instead of their current customer base or requiring costly multiple discrete traditional receivers, adding equipment upgrades. The wideband protocol complexity and cost to the modem. This for a DOCSIS network addresses this by means that first generation wideband modems leaving unchanged the DOCSIS 1.x/2.0 will have built-in capability to receive up to network and the subscribers it supports. Yet it 16 channels, enabling the operator to scale to provides incremental opportunities to begin future bandwidth and throughput servicing customers that demand higher data requirements, without changing the wideband rates. installed base. Looking ahead, cable operators will The figure below illustrates projected high- increasingly support service tiers and speed data services that cable operators may packages that offer varying throughput speeds see in the coming years. In each case, the and quality assurances. This will inevitably service offering is split into three or more fragment customer requirements; many will tiers. A graphic (the diagonal stripe) is focus on lower-speed, value-priced services, superimposed on the chart, indicating the while a smaller percentage will demand long-term costs of using wideband protocol higher-speed services. The challenge to cable versus DOCSIS 1.x/2.0 technology. The operators will be to cost-effectively serve all precise placement of this diagonal stripe is customers. determined by exact costs for the CMTS and cable modem expenses, as well as The wideband protocol is ideally suited for oversubscription rates. this situation. Because it builds upon existing CMTSs and the DOCSIS 1.x/2.0 protocol, it does not affect the customers that are currently served on the plant today. New customers can be served simultaneously with DOCSIS 1.x/2.0 customers from the same CMTS. Additionally, existing provisioning systems and operational processes can be leveraged. This offers an ideal environment for cable operators to employ, given the flexibility that is offered.

WIDEBAND BUSINESS CASE SUMMARY

Regardless, the conclusion is that for many The previous sections have detailed a years to come, the most cost-effective multitude of factors that each contribute to the network is one that simultaneously delivers business case for wideband technology. This services to both sets of customers. section illustrates the combined effect of each of these factors in providing a compelling Minimize Operational Cost business case for deploying the wideband protocol. The charts that follow summarize Backwards-Compatible with DOCSIS 1.x/2.0 the financial results for two deployment scenarios: one using DOCSIS 1.x/2.0 Cable operators must consider how to technology, and one using both DOCSIS continue to innovate and introduce new 1.x/2.0 and wideband. The model is based services to the market, without abandoning upon a cable footprint of 1 million HHP, with existing high-speed data take rates of 25% in year 0 (to reflect the embedded base of equipment and customers) increasing up to 40% in year 5. The chart illustrates how the wideband technology significantly lowers CapEx per subscriber.

But this direct comparison tells only half of the story. A primary reason for using wideband technology is to offer even higher The following conclusions can be made service throughputs, delivering more value, from this business analysis: and capturing more revenue. The remaining charts summarize the financial benefits when • Wideband can dramatically lower the wideband protocol is deployed, enabling network CapEx as subscriber cable operators to deliver services as high as throughput rates increase 100 Mbps data throughput. • CapEx incremental costs required to support the wideband protocol are The CapEx cost per Mb will decrease modest and significantly drive down significantly as a result of using the wideband per Mb cost protocol. The improved revenue and earning • Wideband offers the ability to retain results speak volumes to the benefits of being and continually upsell existing able to offer higher-speed services. customers to higher service tiers • Wideband offers the ability to deliver higher-value and revenue • Operating margins can be improved by offering higher priced service tiers

This analysis captures only the cost of delivering data capacity. Incremental services delivered over this transport offer further upside, particularly in the case of higher data rate services

FULFILLING CABLE’S VISION

Cable operators were the first, and arguably the most credible service providers to articulate the vision of an intelligent, REFERENCES flexible, secure, scalable network suitable for supporting multiple services simultaneously. Chapman, J., "The Wideband Protocol for a The use of IP, and the framework established DOCSIS Network," 2005 SCTE by DOCSIS have been key factors leading to Conference on Emerging Technologies the articulation and fulfillment of that vision. Eng, J., "Next Gen Full-Service, All-Digital But the vision can only be fulfilled if the full HFC Network Beyond DOCSIS 2.0", capacity of the network is leveraged and made 2004 SCTE Conference on Emerging available to the end user. Technologies

The wideband protocol for a DOCSIS ABOUT THE AUTHOR network, with its ability to unleash the full John T. Chapman is currently a potential of the HFC network, is the key to Distinguished Engineer and the Chief making this happen. No longer will cable Architect for the Cable Business Unit at Cisco operators be blocked by artificial restrictions Systems in San Jose, California. As a imposed by historical assumptions and RF founding member of the Cisco Cable BU, channel management techniques. For as much John has made significant contributions to of the spectrum an operator allocates towards Cisco and the cable industry through his er can IP-based services, a custom pioneering work in DOCSIS and development theoretically access. But while the wideband of key technologies and concepts critical to protocol for a DOCSIS network can provide the deployment of IP services over HFC access to this inherent competitive advantage, plants. the real challenge will come in how operators translate this potential into a true strategic Included in these achievements are being business initiative. the primary author of significant portions of the DOCSIS and PacketCable specifications Cable won the initial battle of broadband as well as the originator of DOCSIS Set-top and the triple play. Will the industry be able Gateway (DSG) and evolving specifications to win the next round? Wideband and for DOCSIS Wideband and Modular CMTS subsequent strategies lay the foundation for architectures for the industry’s Next this and provide the catalyst for change for Generation Network Architecture (NGNA) years to come. The outlook for the cable initiative. John has also published a number of industry's success very much depends upon ground breaking whitepapers on Multimedia how effectively cable operators leverage this Traffic Engineering (MMTE), DSG, QoS, and advantage. But there is an additional benefit high availability and is a respected and that is as important if not more so. While the frequently requested speaker at industry wideband protocol for a DOCSIS network events. makes more efficient use of the bandwidth John has 18 patents issued and 27 patents inherent in the HFC network, it also gives pending in a variety of technologies including operators the ability to unleash the full power telephony, VoIP, wide area networking, and of IP, as well as take a major step towards broadband access for HFC cable networks. In achieving true network and service his spare time, John enjoys spending time convergence. As the subject of follow-on with his wife and two daughters. John is a 6th papers, this will be the remaining piece to our Degree Black Belt Master in Tae Kwon Do challenge in achieving a successful business and enjoys white water canoeing and skiing. plan that embraces a true wideband perspective. Previous papers by John may be found at http://www.johntchapman.com

MAXIMIZING BANDWDITH UTILIZATION VIA ADVANCED SPECTRUM MANAGEMENT

Jack Moran Distinguished Member of the Technical Staff Motorola Connected Home Solutions

Abstract This paper discusses the impairments that Extensions beyond DOCSIS 2.0 can allow must be addressed to maximize bandwidth operators to monitor RF performance in a utilization over DOCSIS infrastructure. It non-intrusive manner to truly understand the presents an effective approach for impact of impairments. overcoming these impairments and delivering increased throughput for DOCSIS 1.X and Once these impairments are carefully 2.0 cable modems by taking advantage of measured, operators will then possess the DOCSIS 2.0 extensions. It also shows how information they need to operate at the extensions to DOCSIS can allow operators to highest throughput possible at all times. This implement advanced spectrum management is enabled by the ability to implement to cost-effectively buildout standards-based sophisticated ingress noise cancellation and infrastructure today while retaining the impulse noise immunity techniques. With the flexibility to support emerging or evolving majority of cable modems installed today architectures in the future. supporting only DOCSIS 1.0, there are also advanced spectrum management extensions available to allow operators to nearly double INTRODUCTION the throughput of their installed basis of DOCSIS 1.0 modems by enabling them to The twin demands of more cost- run 16 QAM modulation virtually anywhere effectively utilizing existing buildouts while that QPSK is currently running. delivering increased capacity to the customer require innovative means of maximizing the MEASURING SPECTRUM utilization of existing bandwidth. Clever new approaches are required to optimize the use The first step toward managing spectrum of existing cable modems while achieving is to measure and understand the impairments increased throughput for DOCSIS 2.0 on the DOCSIS infrastructure. modems. Advanced spectrum management can be Without a clear understanding of the deployed along with DOCSIS standards to impacts of impairments, it is difficult to allow operators to improve upstream optimize throughput and meet or exceed throughput by dynamically evaluating the RF DOCSIS performance goals. characteristics of available upstream spectrum and then selecting the spectrum What is truly required from a DOCSIS frequency, modulation mode (16 QAM, 32 Cable Modem Termination System (CMTS) QAM, 64 QAM, 128 TCM, 256 QAM) and is the ability to intelligently assess the channel width (1.6 MHz through 6.4 MHz) to impairments on a given return path in real optimize throughput. time. Unfortunately the time required to make an accurate assessment of impairments Advanced spectrum management is an using traditional measurement tools far extension to DOCSIS that essentially relies exceeds the time budget allotted for on a spare receiver to perform time- measurements. consuming measurements in the background.

Simply stated, the more time that is Operators can therefore gain access to all required to perform an accurate of the return nodes connected to one of the measurement, the more the throughput of the receiver ports and perform tests on any data traffic is impacted. available modem on any one of the receiver While Fast Fourier Transform (FFT) port’s supported nodes. Advanced spectrum measurements are a useful analysis tool for management is a necessity for any cable ingress and impulse noises, this technique operator to be able to efficiently support has little ability to accurately measure micro- VoIP or other demanding real-time services. reflections and group delay. Instead, coherent measurements are required to measure the Given the unknown limitations that exist effects of linear distortions such as micro- in the return path, the DOCSIS CMTS must reflections, amplitude, and group delay assist the cable operator in determining what distortion. any given return path is capable of yielding. Advance spectrum management must be When discussing the system non-linearity completely transparent, with absolutely no class of impairments, only a coherent impact on voice, data, or video throughput. It measurement can be be used. And, the larger must also be able to discern between a linear the QAM constellation, the greater and non-linear distortion to be truly effective. susceptibility to system non-linearity. This has been demonstrated by comparing the Advanced spectrum management is difference between 16-QAM and 64-QAM critical to the ability to optimize the billable given the same non-linear circuit and the capacity of the DOCSIS network. Operators same RMS power for both constellations. simply cannot fix impairments that they

cannot identify and measure. Hard data is The entire point of advanced spectrum necessary so that the effects of multiple management is to assess the unused return impairments can be discerned and path bandwidth for all impairments. This successfully addressed. takes time an active data channel simply cannot afford to spend. The only alternatives to advanced ADVANCED SPECTRUM spectrum management are to ignore (or guess MANAGEMENT DEFINED at) the impacts of impairments, or to deploy expensive, dedicated testing gear to measure Advanced spectrum management allows the various impairments that are present on operators to identify impairments and make any given DOCSIS network. Operators have the necessary adjustments to improve relied on advanced vector signal analyzers performance. An effective method to perform and next generation CATV Analyzers that advanced spectrum management is to utilize support spectrum and DEMOD a dedicated receiver on the CMTS to monitor measurements for several years to performance on any one of the upstream characterize the return path characteristics. paths without impacting performance.

But this approach is expensive, difficult to The receiver technology provides post- deploy and extremely time-consuming to equalization support that can double the measure. It also impacts the performance of throughput for existing customers because the DOCSIS network, requiring operators to virtually all DOCSIS 1.0 cable modems able take a segment of the network out of service to run in QPSK would be able to operate in as the reference sources tend to be constant 16 QAM mode utilizing a post equalization carrier signals that disrupt service in some technique. When there is a significant instances. number of DOCSIS 2.0 cable modems installed, the cable operator can begin the IMPLEMENTING ADVANCED ATDMA Logical Channel Operation in SPECTRUM MANAGEMENT which the Symbol Rate remains the same (2560 ksym/s) but the DOCSIS 2.0 cable By implementing advanced spectrum modems can begin to transmit in a pure management techniques in conjunction with ATDMA mode of operation, i.e. with the DOCSIS specifications, operators can extended Forward Error Correction, byte ensure bandwidth-efficient co-existence interleaving (if necessary), and higher between DOCSIS 1.0, 1.1, and 2.0 cable constellation rates such as 32 QAM or even modems. They can deliver higher upstream up to 256 QAM. bandwidth that can even exceed DOCSIS specifications while ensuring a successful The financial benefits of this migration transition to DOCSIS 2.0. Operators can approach are compelling. Operators can

transition to 2.0 without providing a accelerate revenue from 2.0 services, and performance burden to legacy subscribers they can implement gradual migration at the because the CMTS can operate in DOCSIS pace that makes the most economic sense for 1.X mode. them. They can continue to support legacy modems while introducing new services to LINEAR IMPAIRMENTS these subscribers, and they can concurrently support DOCSIS 1.X and 2.0 operation Micro-reflections are the most common across the same infrastructure. distortions that exist in every plant, but they occur differently in each plant. They are Cable operators can double the upstream caused by impedance mismatches, and the bandwidth for a large population of modems, most significant micro-reflections tend to thus creating increased billable bandwidth occur at the lower tap values of the coax without further network buildout. They can plant. create upstream bandwidth that supports higher-speed services and enable new The lower the tap value, the poorer the broadband services that command premium isolation between the other ports and the pricing. This approach is not only the most cable modem signal. Micro-reflections are practical migration path; it is also the one frequency dependent, so that not all channel with the lowest risk to the cable operator. bandwidth is affected equally. A fundamental problem in isolating a micro-reflection lies in To take advantage of these business the fact the impedance mismatch resulting in opportunities, operators first need to address a rather large micro-reflection tends to be the transient impairment issues that today dominated by a poor termination in a tap constrain upstream bandwidth in the return adjacent to the problem cable modem and not path. to the tap that the cable modem is connected to. This phenomenon is directly related to the MEASURING IMPAIRMENTS poor isolation characteristics of the low tap values. Operators need to be able to improve the Signal-to-Noise Ratio (SNR) so they can Amplitude tilt or slope distortion is also more efficiently manage spectrum and present in every cable plant, and can be improve noise cancellation simultaneously caused by coaxial cable loss or more likely over diverse populations of DOCSIS 1.0, 1.1, by the use of diplex filters. Every plant also and 2.0 modems. Advanced spectrum faces group delay—or “phase” distortion— management is important so that operators which becomes a bigger problem as filtering can understand the various impairments is introduced. The major source of group present in their infrastructure. delay distortion is diplex filters. The more amplifiers in cascade, the more dramatic the This is particularly critical because real- impact of both amplitude and group delay world environments face the following three distortion on a DOCSIS transmission. major classes of impairment, which are present at some level the majority of the Advanced spectrum management allows time: operators to address these types of linear impairments by: • Linear Impairments • Non-linear Impairments • Migrating away from the problem frequencies • Transient Impairments • Reducing the symbol rate, usually by half • Equalizing the distortion

NON-LINEAR IMPAIRMENTS With advanced spectrum management, one can easily observe that the effects of any There are no specifications that exist in non-linearity is that the outer constellation DOCSIS 2.0 that address system non- points are impacted far greater than the inner linearity. DOCSIS 2.0 Technology— constellation points. particularly any modulation type greater than 16-QAM—is inherently weaker against non- TRANSIENT IMPAIRMENTS linearity than is DOCSIS 1.X technology. Transient impairments include ingress This is due to the higher crest factor of noise and impulse noise. Narrowband AM DOCSIS 2.0 that ranges from a minimum of modulation carriers such as shortwave radio 3 dB higher to a maximum of 7 dB higher signals can suddenly appear anywhere in the peak power. There are many variables that return path spectrum. Ingress noise refers to contribute to the final crest factor, such as any interference that is coupled into the the: return path plant via an external source.

• Theoretical crest factor of the The predominant coupling mechanism for modulation constellation ingress noise is a poorly shielded drop • Number of diplex filters in cascade coaxial cable that is acting more like an • Carrier frequency antenna than a drop cable. The overwhelming majority of ingress noise is narrowband AM By definition non-linearity is a signal- modulated carriers whose bandwidth is dependent distortion, which simply means usually less than 20 kHz and ingress noise in that the effects of a non-linearity can only be general seldom has a bandwidth over 200 observed in the presence of a signal. kHz. Return path characterizations conducted by Motorola over time have found that this ingress interference ranges from around -25 Non-linear impairments include Common dBc (25 dB below the DOCSIS signal power) Path Distortion (CPD) and not-so-common to +15 dBc (15 dB above the DOCSIS signal path distortion, often referred to as return power). laser non-linearity. Occasionally, ingress noise is recorded CPD is well understood by the CATV that is as high as +25 dBc. It is therefore industry in general. It is the phenomena of a important to ensure that the CMTS receiver coaxial connector becoming or temporarily front end can take at least +31 dBc or 6 dB acting as a diode. It is easily observed by more than the +25 dBc ingress incidents. seeing analog video carriers spacing 6 MHz Frequency avoidance has been the sole apart throughout the return path. While CPD technique to deal with this parameter until is easily detectable, the return laser being the development of advanced ingress noise either clipped or just becoming marginally cancellation techniques. non-linear can only be witnessed today by advanced spectrum management on a Impulse noise is also a reality in virtually dedicated receiver on a CMTS card or by all return path cable plants. It is made up of deploying vector signal analyzer test short bursts of high-level noise such as that equipment on the network. resulting from the coupling of transients into a channel. Wideband noise events occur continuous testing that negatively impacts the typically in bands wider than 6 MHz, and bandwidth being tested. Hence the paradox; there are multiple sources of this type of many CMTS platforms are only performing impairment. the nearly transparent FFT measurement because operators cannot afford to impose However, the duration of this noise the overhead of measurement. usually lasts in the 1-100 microsecond range. There is another class of impulse noise that is Many operators therefore cannot afford powerline related, and when this type of the time for a coherent measurement transient event occurs, the duration is in the approach—even though it is universally 1-10 millisecond range. agreed that coherent measurement is the only accurate assessment of any impairment’s The one saving grace of either type of impact on a DOCSIS service. impulse noise is that neither has any significant energy beyond 15 MHz. Evidence Advanced spectrum management of this fact is that virtually all DOCSIS 1.X implemented on a CMTS with spare receiver systems operate error free even in 16 QAM ports on interface cards can monitor modulation with a moderate amount of performance on any one of the upstream Forward Error Correction enabled when ports without impacting performance. It can operating over 20 MHz. non-obtrusively gain access to all of the return nodes connected to one of the receiver UNDERSTANDING PERFORMANCE ports and perform tests on any available modem on any one of the receiver port’s Advanced spectrum management allows supported nodes. The spare receiver is tremendous flexibility for understanding the effectively connected in parallel with a complex interaction of impairments on the selected receiver port so the operator can DOCSIS network. No single impairment can measure traffic and performance in real-time be clearly singled out for testing because on any given live receiver port. most—if not all—impairments are present at some level the majority of the time. It It should have access to all of the mapping therefore becomes a matter of assessing the information as well as a full list of cable magnitude of each impairment’s impact on modems available to whichever receiver port the DOCSIS service. is currently being evaluated. Therefore, while the receiver port being monitored is For example, conducting an impulse noise performing its function at full capacity, the performance test without also measuring spare receiver has the luxury of time to ingress noise performance is not particularly perform detailed, lengthy, and coherent SNR insightful. Another fundamental challenge is measurements. It can also perform a host of that measurement time directly impacts other measurements by simply borrowing an throughput, because in the typical scenario idle cable modem for a rich set of return path you cannot send data while you are taking calculations. The borrowed cable modem is measurements. automatically released for service if demands are placed upon it and another idle modem is Operators trying to improve performance selected to ensure there is no intrusion on are hard-pressed to impose increased customer service. demands on the infrastructure by performing Advanced spectrum management also • SNR per cable modem with ingress requires support for adaptive noise noise canceller enabled. cancellation at the receiver to measure the • SNR per cable modem with ingress diverse types of noise, process this noise canceller disabled, which information and take action to cancel it out in reveals whether there is any real time. For example, an ingress canceller significant ingress noise present. on the CMTS can track and cancel rapidly • RX level per cable modem, which changing severe CPDs. The net effect is that when coupled with the cable modem the operator is able to maintain a high-order transmit level from the SNMP MIB QAM modulated digital carrier. will allow operators to calculate network loss per cable modem. If the noise cannot be cancelled out—such as a very large ingress noise or interferer— The reality is that there is a virtually the CMTS can avoid the noise by changing unlimited number of tests operators can the modulation mode or moving frequencies. perform using advanced spectrum Operators can therefore continuously management without impacting the improve performance, proactively recognize performance of active cable modems. and resolve potential bottlenecks, and create more billable bandwidth. CONCLUSION Continuous monitoring and adaptation allows cable operators to aggressively It is very difficult to improve something implement advanced noise cancellation in that you cannot measure. Operators need environments where the types and degrees of hard data on network performance, but noise change frequently. These impairments cannot afford the cost and complexity of have historically been the limiting factors in deploying test equipment throughout the achieving QAM modulation higher than 4 network. QAM (QPSK). The combination of post equalization and superior ingress noise Merely guessing at the impact of cancellation capabilities results in a DOCSIS impairments is a wasteful and frustrating 1.X system today where 16 QAM, error-free exercise. The ability to accurately measure operation is achievable virtually anywhere in and monitor impairments is essential to the return path. optimizing the productive use of bandwidth and ensuring the successful delivery of real- SAMPLE MEASUREMENTS time video and voice services.

The following is just a small sampling of By implementing advanced spectrum the types of measurements operators can management, operators can non-obtrusively implement using advanced spectrum monitor the impact of impairments on an management: ongoing basis and take corrective actions when necessary to improve performance of • SNR per cable modem with post voice, data, and video services. equalization enabled. • SNR per cable modem with post ABOUT THE AUTHOR equalization disabled, which reveals whether there is any significant micro Jack Moran is a Distinguished Member of reflection on a per-modem basis. the Technical Staff for the IP Solutions Group for Motorola Connected Home included many live plant characterizations in Solutions. He is the holder of 11 U.S. patents an effort to simulate on a repeatable basis the in data communications, with many more type of real-world impairments that DOCSIS pending. systems must overcome.

Moran is responsible for DOCSIS Physical Layer performance over an HFC RF He is also a member of the DOCSIS 2.0 network system. For over four years, he has Technical Team PHY Layer Issue as well as been modeling the return path for DOCSIS a member of the former IEEE 802.14 Cable 1.0, 1.1, and now 2.0 performance Modem Group. Moran can be reached at capabilities. This modeling effort has [email protected]. MAXIMUM CAPACITY: THE ROLE OF INTELLIGENT EDGE DEVICES IN CABLE NETWORK CONVERGENCE

Michael Adams Terayon Communication Systems

Abstract management system, which doesn’t ‘know’ about the other services. The ultimate goal of many cable systems operators is the migration to an IP end-to- Most cable operators also manage two or end delivery model. However, there are many more transport networks (regional, and constraints at the HFC level such as existing headend-to-hub interconnections). The analog channel assignments, must-carry history of technical development is often the regulation, legacy set-top box investment, driver for the right choice at the appropriate and interface standardization. time, and four different technologies are in

common use today: Nevertheless, the internal structure of cable systems is evolving rapidly towards a • AM Supertrunking converged IP backbone that carries all • IP over ATM over SONET content distribution and signaling. • IP (Packet) over SONET • IP over Gigabit Ethernet Therefore what is needed is an intelligent edge device that can perform format The complexity of protocol conversion conversion, protocol translation, and content and bandwidth adaptation at the edge is localization so that the IP backbone can be significant, but it is now possible to build implemented today, with the ultimate goal of cost-effective, intelligent edge devices that extending IP carriage all the way to the allow a single transport network and home. considerable simplification of the HFC network. These developments also support INTRODUCTION efficient bandwidth management and a graceful migration to a single, converged IP network topology, which will enable rapid Today cable operators are typically deployment of new services without having managing four, discrete network protocols to invent new operation management over their HFC plant: systems.

• Analog broadcast video This paper will describe this evolution in • Digital broadcast video detail, and show how flexibility and cost- • On-demand video effective deployment can be achieved at each • High-speed data and VoIP step of the process. The steps described will cover: The HFC bandwidth allocation for each is statically defined as 6 MHz channels. It is 1. Compression of all services to a common difficult to re-distribute network capacity coding and transport standard. between services because each has a separate 2. Transmission of all services over a single, result of this the cable operator is faced with IP-based transport network. the challenge of operating multiple different 3. Localization of video services to the zone systems, each with its own operational or set-top level using digital program quirks. insertion (DPI). Switched Broadcast services may leverage the same For example, when digital services were infrastructure that is used for localization. added, the most cost effective deployment 4. Adaptation of services to maintain was to modulate and combine the analog and backward compatibility with legacy digital channels at the headend and to use devices in customers’ homes. (This is existing AM supertrunking to deliver that especially important for regulatory combined signal to the distribution hubs. reasons and to preserve the multi-billion Figure 1 illustrates the basic transformations dollar investment the industry has made that are done at the headend to allow video in MPEG-2 set-tops.) feeds to be selectively groomed by the operator into an optimal channel lineup. HEADEND AGGREGATION AND GROOMING As shown in the diagram, all functions that transform the content are done at the Cable networks have developed over the master headend. From the master headend, years by adding new services in an the network uses Amplitude Modulation incremental fashion – each new service (AM) Supertrunking to distribute the RF bringing with it new headend equipment and signals to the hubs and ultimately to the leveraging existing network transport. As a viewer’s home.

Figure 1: Headend Aggregation and Grooming

In Figure 1, all digital feeds from satellite 1. The feeds are aggregated together, are fed into a series of groomers, which allowing two satellite feeds to be perform a number of related functions: typically combined into a single QAM channel over the cable system. 2. The feeds are groomed to remove any services over a single backbone. The single programs that are not required for biggest issue becomes how to carry feeds carriage over the cable system over the backbone that are only available in 3. The feeds are re-mapped so that any analog format. Until recently it was Packet Identifier conflicts are removed prohibitively expensive to MPEG encode and program map and program analog channels, but as encoding costs have association tables built for the come down, and as cable systems have consolidated transport stream output. grown larger (so that the cost of encoding an 4. The programs are statistically analog channel is spread over a larger multiplexed so that the instantaneous bit subscriber base) this is no longer such an rate does not exceed the output channel issue. limitation (38.8 Mbps for a 256-QAM channel). The other issue, again until recently, was the need to do local commercial insertion Although this approach has been into analog channels. This has been solved by extremely successful in allowing cable the implementation of digital-in-digital operators to design their own efficient digital insertion systems that operate at the MPEG channel line-ups, there are some significant layer, using a standardized approach called disadvantages: Digital Program Insertion.

1. The digital line-up is fixed for all parts of DIGITAL PROGRAM INSERTION the cable system – this may be a problem if different parts of the system are not The basic principle of Digital Program upgraded to the same capacity. Insertion (DPI) is that at a given digital cue 2. As regional clusters are interconnected, it signal (signaled using the SCTE 0351 is possible to push much of the headend message) an individual output program can functionality back to a regional super- be seamlessly spliced from the network feed headend. to a local feed generated by a server. The DPI 3. The AM supertrunk is still being used to system has been divided into two main sub- deliver the digital video channels. As systems; the Ad Server and the Splicer. The cost-effective digital transport two communicate using a set of standard technologies are becoming readily messages according to SCTE 0302. available and are being used to Video On Demand (see later). It is becoming After the local commercial is inserted, the practical to unify all transport over a program is spliced back into the network single, IP transport system. feed. To do this successfully, the splicer has to be aware of the MPEG decode buffer For these reasons, an all-digital model and the structure of the MPEG distribution model is rapidly gaining encoding syntax. In practice the bit rate of acceptance. As we will see, it is also a the inserted commercial has to be modified to foundation for Digital Program Insertion, ensure a smooth transition without frame Video-on-Demand, and Digital Simulcast. drop or repeat, therefore rate-shaping technology is incorporated into the DPI Of course, all digital distribution solves a splicer. number of problems. There is adequate capacity available in modern 10 Gbps multi- Early deployments of DPI were wavelength transport systems to carry all implemented at the MPEG-2 physical layer using Asynchronous Serial Interface (ASI) transport packets are encapsulated into a interconnections. However, DPI is becoming UDP flow, which can be carried over an IP the first step in the migration to an IP network. network layer. To achieve this the MPEG-2

Figure 2: Digital Program Insertion

Figure 2 shows the latest generation of introduction to the DPI standard is available DPI implementation. There are a number of in SCTE 0673. key points: VIDEO ON DEMAND 1. The Digital Ad Server does not have to be co-located with the DPI splicer Video-on-Demand is being rolled out because they are connected via an IP aggressively by all major MSOs. The connection (and not a distance-limited technology has evolved to larger capacity ASI connection). servers with Gigabit Ethernet or 10 Gigabit 2. The Digital Ad Server can be centralized Ethernet output ports connected over an at the headend for ease of operations and optical IP transport network to high-density maintenance. edge-QAM devices. 3. The DPI Splice can be located at the edge of the backbone network at the Figure 3 shows a typical current Distribution Hub. This allows the implementation of VOD. Note that the VOD operator to provide zoned ad insertion to server is centralized and that the Edge QAMs an arbitrarily small serving group area. are distributed to the distribution hub sites. In this diagram a potential future DPI is a sophisticated new technology implementation of rate shaping is shown, in with tremendous flexibility. A good which CBR streams are statistically multiplexed before delivery to the edge- is a nice property as the same equipment can QAMs. In the current CBR implementation, be easily re-configured between broadcast, 10 streams encoded at 3.75 Mbps per stream DPI, and VOD services. In addition, can be delivered over a 256-QAM channel. Switched Digital Broadcast can also be Depending on content, a significant gain of accommodated using this same basic maybe 2-4 streams per QAM can readily be architecture, the additional complexity of achieved. In some cases this technology may Switched Digital Broadcast being extensions be used to accommodate peak demand to the control plane to allow set-tops to without blocking. request channels (in the same way that VOD allows a set-top to request sessions). The important thing to note is that Figure 3 is identical architecturally to Figure 2. This

Figure 3: Video-on-Demand

DIGITAL SIMULCAST backbone network, which also supports all other connectivity required in the system for Digital Simulcast makes all programming on-demand programming, high-speed data, available as part of a digital tier. To do this and voice. any analog feeds must be encoded, even off- air signals. Figure 4 shows the processing at When the signals arrive at the distribution the headend to support digital simulcast. hub, an intelligent edge device terminates Once all of the program feeds are in MPEG-2 them and distributes them to the various format, they can be encapsulated into IP legacy channels for distribution over the HFC packets for transfer over the converged IP plant as shown in Figure 4.

Figure 4: Digital Simulcast

While the operator still has to provide SUMMARY analog channels, an MPEG decoder and analog modulator are required for each This paper has illustrated the evolutionary channel to convert it back into NTSC format. migration path from today’s hybrid transport Over a period of time, the goal of the model (using AM supertrunking) to a digital operator will be to drastically reduce the transport model that uses a converged IP number of analog channels because they backbone for all video, data, and voice consume so much HFC bandwidth compared transport. with their digital equivalent. As analog channels are removed, more digital channels REFERENCES can be added for broadcast or on-demand 1. ANSI/SCTE 030 2001 (formerly DVS viewing. Another advantage is that the DPI 380) Digital Program Insertion Splicing splicing in the digital domain is done before API conversion back to analog, and so older 2. ANSI/SCTE 035 2004 Digital Program analog insertion technology can be retired. Insertion Cueing Message for Cable

At this point in the migration path, we 3. ANSI/SCTE 67 2002 (formerly DVS already have achieved complete convergence 379) Applications Guidelines for SCTE at the IP backbone level, with all four 35 2001 services – analog video, digital video, video- on-demand, and DOCSIS – running over a (Available online from the SCTE web site at single backbone. This allows the operator to www.scte.org) reduce operation costs and to operate larger networks more efficiently. NEXT GENERATION VOD ASSET MANAGEMENT MANAGING THE VOD CHAIN END TO END

Chris Stasi - Vice President, Operations TVN Entertainment

On Demand product has become a files and data as well. Currently the total significant source of content to consumers hours of content available in the On Demand but has only just begun to tap into its market is on the order of 3,000-4,000 per potential. As more and more content is month. As the platforms become ubiquitous, accessed, stored and viewed in an on- and the technology easier for the average demand environment, the requirements of consumer to understand, the number of asset management become exponentially hours will increase quickly to 10,000 hours greater. Systems will need to coordinate and beyond. Ominously, this does not scale asset distribution from the time assets are linearly. With the introduction of the new exposed from a provider throughout their elements and features of the Cablelabs lifecycle, including distribution, ad standards and the rapidly growing incursion insertion, streaming and financial reporting. of advertisers into the On Demand market, One centralized system, managing content the type of asset management required scheduling, offer creation, updates, changes, as well as the amount. distribution, tracking, metadata management, edge playout and wholesale Where We Are Now reporting makes both MSO and provider VOD systems more efficient and scalable. Current asset management consists, predominantly, of two areas: metadata and This paper will present an asset content files. Both require a somewhat management strategy for both MSO’s and straightforward management style. providers covering the entire lifecycle of all pieces of the VOD architecture. In the current Cablelabs1 environment the • MSO asset tracking, updating and metadata file is capable of a relatively small reporting number of functions. It is the steward of the • MSO headend management content from the time the content is • Distribution management created/encoded until EPG population. In • Content Provider Asset Scheduling, the interim it guides the asset into residence tracking, updating and remote on the VOD server and populates the monitoring associated databases. Subsequently its only functions are updates to certain allowable fields (on capable systems) and deletes. On Demand content has only begun to scratch the surface of what is achievable in The content file obviously functions the consumer entertainment market. As the solely to be viewed and the corresponding nascent technology becomes more and more management of it is currently refined a new level of interoperability and straightforward. It exists only as part of its asset management is required. This includes own, single, offer/package and when the not only the content itself but the associated license window runs out it is deleted from the video server. It is also tracked solely by product and had shown a willing appetite. references to it within its single metadata Most of this programming was long-form file. All of this is about to change and those and therefore small in number and easily changes will introduce an enormous layer of trackable. Over the past 18 months the complexity into the On Demand introduction of music videos, barkers, environment. magazine shows and various other short- form assets has shown the direction content Changes on the Horizon is headed. Advertising is the obvious next step and with its plethora of short-form, The biggest issues facing On Demand unique, content appears to be the 900-pound players are Cablelabs upcoming 2.02 gorilla entering the room. standard and the changing nature of the On Demand content itself. Both cause some of Macro Problems the same issues and many of those issues can be solved in similar ways. So what asset management problems present themselves in this environment? The first asset management issue is the There are a myriad of small ones, the devil Cablelabs specification change. Whereas is always in the details, but the major ones currently there is one offer associated with can be broken down into some sizable each content file, and each file exists in only buckets: its own, single, content package, in 2.0 the content file(s) each exist at their own level. 1. The lack of a widely used, There are one or more title/offer level ubiquitous data set surrounding all metadata files tracking each asset and On Demand content potentially multiple potential files tying each 2. Vendors within the On Demand asset to other assets. Thus the single content space exhibiting different asset that currently exists on its own with its capabilities and requirements own metadata file will now be able to sit on 3. Protecting the content to the a server as part of many different offers and satisfaction of content providers be acted upon in many different ways. This 4. The inability to update content in a affects the content provider offering the near/real-time, meaningful way content, the distribution company sending and tracking the file, the server allowing Problem 1: Differing Data Sets access to the file, the EPG’s finding and displaying each offer of the content and the Potentially the biggest threat to billing systems tracing usage. Many of the widespread content usage is the inability for links in this particular value chain are not providers and MSO’s to track content in a currently capable of the required changes. meaningful way. Currently the Cablelabs specification does a comprehensive job The second issue is the changing type of giving all parties the same ability to do so On Demand content. The On Demand but not everyone is taking advantage of the platform began mostly as a sister entity – information. The most widespread current programmatically – to the Pay Per View Cablelabs spec is 1.1 and it has a fairly product. Movies were the first entry as well straightforward, well thought-out structure. as Premium channel’s SVOD offerings. Each content file, be it the main asset, a Consumers were most familiar with this corresponding piece of artwork, or a trailer, has its own asset ID. The metadata file Solution: corresponding to that asset and all the offer information it contains also has an asset ID, So what can be done to make it easier to called Title ID. The combination of all this distribute and track content in an On is the content Package and it has an ID of its Demand environment? The easiest and most own. The media files should keep their ID obvious is for providers to all work off a forever and any changes to the metadata common data system. Not that the onus should result in a change to the Title ID. should fall only on them but as they are the Any change whatsoever to any aspect of the beginning of the road they get the first package results in a change to the Package responsibility. The Cablelabs 1.1 format ID. However this is not occurring across the allows all providers as much flexibility as industry. Widespread interpretive needed to identify their content. The use of differences of the spec as well as embedded domain name as the first half of a functional roadblocks have caused many concatenated ID guarantees uniqueness providers as well as equipment and across providers. The assignment of the ID’s distribution companies to reach a different within each provider is then up to the conclusion about how the data is managed. provider itself to ensure and maintain This has caused issues along the entire uniqueness. length of the On Demand chain. A provider has problems because if he/she does not Following that, it is up to the distribution track a content file with the same Mpeg ID companies to maintain the provider’s ID there is no way to know how many times integrity throughout the industry. that same piece of content has been Maintaining this trackable ID structure is reintroduced or to get any meaningful crucial to insuring that the content itself is comparative usage data back from an MSO. traceable from a providers own management A distribution company cannot maintain any system, updatable from that same system semblance of an organized library and and won’t clash with assets being distributed distribution system if an ID structure is not to a location by another distributor. followed for the life of an asset. A server vendor cannot reasonably be expected to At that location it is then up to the cover all various types of content differing equipment vendors, mostly introduction, and make any type of gateway, server and billing systems, to be meaningful error reporting on content, if it able to ingest the ID string in the way it is cannot expect a standard set of rules to be intended. Currently there is a severe followed. If Titanic is introduced into VOD disconnect throughout the industry at this for the first time at one price, which filters point and it causes not only a loss of throughout the server and billing interfaces, traceability but literally a stoppage of and the following year returns at a lower content propagation and usage-data flow, price, but the Title ID isn’t changed, there is which results in revenue loss. Following the no standard way to report on the different protocols in the chart below shows the price points. Data fields within the metadata intended use of each ID level and what files, such as Billing ID’s, cause the same should and should not be passed. issues.

Original Offer New Offer Provider TVN TVN Provider ID TVN.com TVN.com Title Spider Man 2 Spider Man 2 License Start Date 1/1/05 3/1/05 License End Date 1/31/05 3/31/05

Package Asset ID TVNX1234500000000000 New Title Asset ID TVNX1234500000000001 New Billing ID 56785 New Movie Asset ID TVNX1234500000000002 Same Movie Content Value TVNX1234500000000002_movie.mpg Same Preview Asset ID TVNX1234500000000003 Same Preview Content Value TVNX1234500000000003_tr.mpg Same Poster Art Asset ID TVNX1234500000000004 Same Poster Art Content Value TVNX1234500000000004_photo.bmp Same

Upon the introduction of Cablelabs 2.0, reintroduction to the server each time, the data set, if possible, becomes even more saving money, time and processing power. important. In 2.0 the elements exist on their own, apart from a singular, ever-present The widespread adaptation of the same package ID. Collections come into play and data structure everywhere is a hurdle that they enable content assets to be acted upon has to be overcome quickly in order for On by metadata that hasn’t originated with the Demand content to evolve into the next asset’s initial introduction. A collection generation product everyone desires. could consist solely of metadata introduced halfway through a movie’s lifecycle offering Problem 2: Different Requirements from a discount if the movie is viewed in tandem Edge Equipment with a related new release feature new to the video server. If the collection metadata Similar to the previous problem, but refers to an asset ID and that ID is not potentially trickier to solve, is the differing recognized the same way by every system level of capabilities and requirements of nationwide the collection is not usable and hardware throughout the On Demand again revenue is lost because of a disconnect architecture. These differences can be seen on how best to interpret data. in areas as diverse as encoding specs, data requirements or EPG display capabilities, The features and benefits of moving yet they all have a solution within the realm of asset management. everyone into the same ubiquitous data set are easy to see. The most basic is simply the On Demand Asset management doesn’t ability to track where your content is and mean simply tracking data but also the who is using it. Building on that enables the construction and movement of the media introduction of ads. If you know where a hit itself. This area has encountered some movie resides and everyone knows the same difficulties because of different requirements ID structure it’s easy to introduce an ad into across different platforms. a collection or playlist available to the end- user. It also enables the main content to be A perfect example is that the encoding re-used without requiring re-pitching and specifications within Cablelabs3 have set the baseline for asset construction but there are beginning to emerge through the Cablelabs far more areas for missteps within the grey group. area of encoder setup itself. A configuration such as Program Stream has a choice within Solution: the specification but if an asset encoded on a valid, compliant stream arrives on some So what can be done to standardize the servers it will fail to ingest correctly even requirements for different vendor equipment though it is a “legal” encode. Obviously a in the On Demand environment? Tough spec in a new environment cannot possibly question. Anyone who has spent any time in be expected to have uncovered every nuance the space knows that not only are the normal across all manufacturers’ equipment, but the political elements involved in equipment introduction of elements such as Gig-E purchase, vendor relationships and sales switches on the VOD network and other incentives but the element of speed to increasingly common architecture changes market also comes into play much more require a bit more standardization. heavily in new technology centers. Manufacturers are forced to design on the fly and implement upgrades and rollouts As mentioned in the previous section, before they’re ready in order not to lose billing systems currently present a few market share to a competitor. But this problems in the next generation On Demand shouldn’t stop the industry as a whole, environment. Many are set up to deal with through Cablelabs and other like ID’s from the PPV world. This was fine organizations, from beginning to implement when there were assets numbering in the standards that go beyond the data and dozens arriving at a system each month and baseline encoding specs that are currently the only difference was time of day they deployed. were being watched. On Demand content has already reached the level of thousands of While there will always be differences pieces of content per month and is growing between rival commercial products, which rapidly. No provider is going to allow his or obviously should be encouraged, there is no her content to be used forever without reason not to ensure certain quantifiable, meaningful revenue in some form. That measurable benchmarks be achieved in more requires either direct usage-tracking for paid areas. In order to be Cablelabs compliant a transactions or click-data for advertisements. server and a distributor need to meet certain Many current billing systems cannot provide criteria in metadata creation and mpeg this data based off the 1.1 and 2.0 ID settings. Other settings need to be structures. If this continues, the introduction implemented as well, such as minimum of playlists, collections and increased capabilities on an EPG or minimum data shortform data will come to a standstill. compliance from the billing side. There is no reason a billing system cannot be certified as EPG’s also need to be able to offer the compliant in the On Demand environment same minimum usage experience regardless the same way a server vendor must be. As of the platform they rest on. Currently each more and more small MSO’s get into the On EPG is capable of different data capture and Demand space, and as more and more new displays. While there will always be companies bring their equipment into it, the commercial reasons to offer different natural propensity will be to become more capabilities, a certain baseline set is disparate in the technology sets, not less. Broadening the standards to include more ingests into the video server and resides equipment, encoding, display and interface during its license window for streaming to requirements is the necessary next step to the customer’s set-top box. Every one of ensure advancement and interoperability. these touch points is susceptible to either the easy duplication/theft of content or at least Problem 3: Content Protection the appearance thereof. The studios will not differentiate between the two and will As a content distributor, it has become require all touch points to have the same obvious over the past few months that one of level of security. the biggest waves cresting on the horizon in 2005 is content protection and encryption. The fact the studios haven’t been more Providers are increasingly focused on focused on this issue so far is due to a ensuring their content is as safe as possible myriad of reasons, one of which is the as it moves through its digital lifecycle. In seemingly safe environment of the secure an On Demand environment, where content headend. However this won’t last. is being passed from studios to distributors, According to a study by AT&T and the sent over satellites across the country and University of Pennsylvania,4 77% of taken down in hundreds of headends with trackable movie theft has been traced back almost as many differing security protocols, to studio insiders. If the actual film it’s easy to see why a major motion picture companies themselves have been so studio or television network would be compromised, an MSO headend, where the concerned about their content falling into the files are already digitized, is an even easier wrong hands. The seemingly endless mark. downward spiral of profits within the music industry due to peer to peer file-sharing and Solution: illegal downloading is nothing if not alarming to content owners, and with good reason. So how do we prevent such a potential platform-destroying problem from taking The question obviously isn’t whether to root? The issue, obviously, is to protect the protect and encrypt content but how to do it content as thoroughly and as long as in a way that makes managing the protection possible. But with all the previously protocols and the content itself as easy and mentioned touch points for any single piece as transferable as possible. Without some of content how does that take shape? format for doing this, managing the asset throughout the VOD chain will not be The best and most scalable solution is to possible. encrypt any piece of content from its moment of creation (encoding) and offer the The biggest problem implementing such choice to leave that encryption on, or a format is the number of places the content decrypt, at the natural handoff points is touched in an On Demand environment in throughout the distribution chain. With a order to reach the end-user. It starts with the modular, strippable solution in place the dub house, moves to one or more content is never clear-text except in a closed distribution/transmission companies where it network, while it is simply changing cipher, is encoded and multicast and ends up at the not eliminating it. So how does this look? MSO/Telco’s headend. It is there that it There are multiple types of encryption The encrypted data packets are and no two MSO’s will want to implement it transparent, so firewalls, proxies etc. can see exactly the same way. Some will favor pre- the encrypted packets and pass them through encryption and some will favor session- as if they were unencrypted. This allows any based. Many factors will go into this, each encrypted file to pass itself down through as specific to the MSO, and as valid, as any the value chain and be acted upon other. The best way to enable all of them is (transferred, streamed, trick-file creation) as to offer the content to them in a way that if it were clear text. If this end-to-end pre- makes either possible. encryption is carried through it is up to the distribution company to manage the key-list A distribution company is generally the server and enable all of its MSO’s to receive first player in On Demand to touch the these keys. It also has to ensure that the content in a digital format. (This is a encryption scheme is standards based. generalization as many content providers do Getting all the encryption companies to their own encoding.) The first thing any maintain a standard scheme, and potentially distribution company must do is create an a universal cipher, is underway and will go a MPAA approved library storage system that long way to making the transition to a is approved from a physical access completely encrypted product easier. standpoint and also stores encoded files solely on a private network. This obvious If an MSO chooses to receive the content step will keep access to the files to as few as clear text the obvious choices are session- people as possible. based encryption, with its higher degree of safety but commensurately higher costs, or At this point the files should be to re-encrypt on the server. This requires the encrypted, but how to do that this early and same client decryption on the set-top. still fit into the modular scheme mentioned above? A standard 128 bit AES encryption There are many complications as scheme can be applied at this, or any, point encryption enters the On Demand and passed down the chain. As the content is environment. By adhering to a standards- received at an MSO’s headend the receiving based, modular, flexible encryption scheme device can be configured to either decrypt that allows the cipher to be removed or the content prior to handing to the video passed through at any point, the studios and server or pass it through in its encrypted networks will be satisfied their content will state. If the decryption takes place the be protected as thoroughly as they demand. content must be handed to the video server This will remove a major roadblock from across a private network to ensure its safety. managing On Demand assets throughout If the MSO chooses they can receive the their lifecycle content still in an encrypted state and pass it all the way to the set-top. This of course Problem 4: Immediate Data Access requires the set-tops to host a decrypt client that can receive the decryption keys. It is in A problem that has not derailed the the MSO’s best interest to get set-top rollout of On Demand content but will soon manufacturers to certify as many of these pose a big problem is the inability for clients as possible to ensure competition and programmers, distribution companies and an open standards-based solution. even MSO’s to act upon the data surrounding their content in a real-time, meaningful fashion. As any new technology to take advantage of it. rolls out to customers, the wow factor carries it through its initial hiccups. This Solution: fades quickly and soon the product needs to meet consumers’ up to the minute demands The easiest was to prevent this data or it will fade into obscurity. Once the restriction from adversely affecting the technology is settled the marketing takes programming in On Demand is to open up over. The On Demand product chain has the usage data and give everyone along the been very limited in what the marketers can distribution chain access to changing it. This do because the data they can get their hands consists of two major steps. on has been very delayed. Even the small amount that comes to them quickly can’t be updated or changed in an easy, ubiquitous The first is to allow a standardized fashion. This will have to change quickly for interface into the server usage data. This the On Demand product to gain relevance to doesn’t necessarily mean giving providers or the end-user. It will also be a requirement distributors private customer information. for advertisers. Nobody wants an ad running What it does mean is at least passing back that is out of date by the time he/she has the generalized buy rates and click data. This ability to update it. can be done through a standardized reverse interface from the server outward. A With the upcoming release of the distributor can link into the server and Cablelabs 2.0 standard the ability to wrap extract (or more likely be fed) a set of usage metadata around content differently will data that can filter back to the distributor for help bring some marketing capabilities into storage in their database. This data can then the On Demand environment. An advertiser be mined to determine what pieces of can use either a collection or a playlist to content are performing better or worse than attach ads to free On Demand content or aim expected. Then, assuming the MSO’s targeted ads at a willing viewer. An asset equipment is updatable (which by CL 2.0 it price or window can be changed, or that would have to be) the data can be used to asset can be packaged with like material to extend, change or refine the offering to the use one as a loss leader, after the material consumer, making the asset more valuable. has been in its viewing window for some This interface has begun to get some traction time. There are still some major hurdles that within the Cablelabs consortium. need to be overcome before this is anything more than a technical capability though. This information will be worthless There is no performance data available to however without the ability to view it, make providers to determine how, for example, a decisions and act upon it. A distributor New Release movie is performing from receiving this information from a server market to market until the early heavy-usage needs the ability to grant any provider using period has passed. Even with this data at his/her system access to the data and update their fingertips the providers have no way of functionality in real time. A distributor’s acting upon it remotely without involving system needs to be able to author content another vendor or distribution partner. The within it and send that content to the only way to aim these ads correctly, or know destination headend’s but that’s not all. In what offer to adjust on a new title, is to have order to make data management meaningful, immediate access to usage data and a system any data brought back to the provider through the reverse interface needs to be Moving Forward editable by the same provider without having to go headend by headend. One On Demand (in one form or another) will centralized system, able to update data, eventually be how most people watch most nationally or singularly, is necessary. This content. Many more issues than discussed system needs to manage content preparation, here will eventually pop up, some of which transmission and headend interfacing at take everyone by surprise. But planning for every level, in other words an overall layer the next generation of asset management sitting on top of the entire value chain. This starts now. Done right, everyone involved enables any interested party, anywhere on can benefit, from consumers through the the value chain, access (with permissions of smallest distributor. course) to their content, be they provider, REFERENCES: distributor or MSO. Without this universal overlay the job of manipulating content will 1. http://www.cablelabs.org/projects/meta grow exponentially larger and quickly get data/specifications/specifications11.html out of hand. This will have the effect of making content quickly grow stale, thereby 2. http://www.cablelabs.org/projects/meta seeming less interesting to the consumer. As data/specifications/specifications20.html advertising and short-form assets continue to be a larger part of On Demand content and 3. http://www.cablelabs.org/projects/meta as the number of files associated with each data/specifications/content_encoding.html piece of content grow, one system capable 4. http://www.aeanet.org/Government of managing this content form end to end is Affairs/gamb972_ATTReport_MoviePiracy. a necessity. asp?bhcp=1

OPTIMAL QAM ASSIGNMENT IN THE PRESENCE OF MIXED SD AND HD STREAMS

Jiong Gong andYasser Syed Cable Television Laboratories, Inc.

Abstract of a modulo in the streaming bit rate with the division remainder to be zero. When VoD systems need to accommodate both SD and HD streams, the traditional When VoD systems need to accommodate capacity engineering rule for deploying QAM both SD and HD streams, the traditional modulators face a new challenge. Two issues stream capacity assignment rule for QAM arise from this new paradigm. One is the modulators faces a new challenge, in that it is streaming bit rate for HD, and perhaps both possible to incur blockage on each of the HD and SD streams to optimize the system QAM modulators while jointly they have the performance. The second issue is the QAM capacity to support the arrival of a new allocation algorithm to minimize system stream request. In other words, the posterior inefficiencies. We propose a solution that allocation of the streams is suboptimal. Both has the potential to significantly improve the busiest and the least-busy rule tend to system performance to accommodate a have suboptimal allocations. This brings mixture of SD and HD VoD streams about the issue of finding an alternative compared to prevalent methods. algorithm to improve the allocation efficiency. We propose in this paper an optimal solution that is a significant INTRODUCTION improvement over current methods.

With the rapid penetration of HD TV sets VOD STREAMING BIT RATES in the consumer electronics market, cable companies have been aggressively deploying Before getting into QAM allocation High Definition (HD) cable channels in algorithms, it is important to first look at the recent years. A powerful competitive issue of VOD streaming bit rates. Design of response to the DBS offerings is the HD VoD bit rates should consider both the issue of service. When the VoD streams consist of fully using QAM resources and the quality only one type that is of Standard Definition issue perceived by viewers. For each type of (SD) TV, the Quadrature Amplitude stream to fully utilize the useable bandwidth Modulation (QAM) assignment algorithm is capacity of a QAM, there must be a modular not critical in affecting overall system relationship between each type of stream. performance. Commonly implemented Additionally, if a different capacity QAM is algorithms include starting from the busiest used then a modular relationship must still QAM and starting from the least-busy QAM. also exist for full utilization. In reality, the Either way is not going to affect the blocking limits of this relationship are dependent on rate of the system. To make full use of a the modulation types in use. Since quality is QAM channel capacity, bit rate is often dependent on bit rate, there is an additional chosen so that all possible numbers of limit to designing the modularity factor streams in a channel form a congruence class between the two types of streams. current pratice in many VoD systems is to There are some real practical limitations tear down the stream, if the stream incurs on this relationship. The data rate used in a more than 5 minutes of pausing. One futher QAM-designated VOD service today is 37.5 complexity is whether the torn down stream Mbps for 256 QAM and 26.25 Mbps for 64 needs to have any priority over new stream QAM. The additional capacity in the data requests, if it needs to be resumed again. If rate is reserved for in-band traffic. Today a so, this would affect how the bandwidth is typical SD VOD stream is at a constant bit allocated and the amount of time the rate of 3.75 Mbps which is good quality for bandwidth is reserved (e.g. a 2 hour movie MPEG-2 paid movie content. If one wanted may typically have a reserve time that has an to determine the HD MPEG-2 stream bit rate extra 20-30 minutes). based on modularity on a fully-utilized QAM, then HD date rates would be 7.5 Mbps QAM RESOURCE ALLOCATION (2x of SD stream) or 18.75 Mbps (5x of SD stream), which is either too low in quality or In this section, we describe a QAM too much data rate used. Alternatives to this allocation algorithm that we believe would be 11.25 Mbps (3x of SD stream) or represents a significant improvement over 15 Mbps (4x of SD stream). Both of these current prevalent methods. We start off by offer acceptable quality, but they are not describing a mathematical framework to fully-useable “pure” QAMs, requiring either model the problem. Suppose a collection of a 2 HD/2 SD- 256 QAM (or 1 HD/2 SD- 64 n QAM modulators is deployed to serve a

QAM) stream configuration or 3 HD VoD service group. Let qi , i = 21 , ..., n, , stream/1 SD stream-256 QAM (or 2HD/1SD- denote the used capacity of each QAM 64 QAM) configuration because of a modular modulator i. Total capacity, Q, is assumed to relationship. be the same for all QAM modulators. Therefore, the remaining capacity that can be In typical systems, a VOD service to a used for new stream requests on that QAM node is allocated in 4 QAMs (for now let’s modulator is then − qQ i . Let rs and rh assume 256 QAM) or an integer multiple of denote the streaming bit rate respectively for it. This has to do basically with fiber SD and HD streams. The two types of distribution to a node. For a pure SD VOD streams may arrive at a collection of QAM service, this would be about 40 streams that resources according to two distinct random could service about 400 homes, assuming a processes, such as the Poisson process, but 10% peak capacity. For a pure HD VOD exit the system based on the same holding service, this could be either 8 to 12 streams time distribution. We call the snap shot state that could service from 80 to 120 homes, of all QAMs [ q ] at a particular time as an assuming a 10 % peak capacity. If a QAM i allocation algorithm is not properly allocation. We define an allocation as configured, this would either lead to blocking inefficient, if, of HD streams to 4 to 8 streams by the wrong placement of 4 SD streams. That leads to − < rqQ hi , ∀i , and ∑ ≥− rqQ hi only 40 to 80 households for HD VoD service, assuing 10% peak usage rate. In other words, none of the QAMs The amount of time of this blocking individually has the capacity, even though would depend on length of overlap that the the sum of all available resources on each bandwidth is reserved for each movie. The QAM is able to support a HD stream request. Note that while each type of stream is in prioritize over the states of available QAM itself modulus in its own bit rate, they jointly modulators for stream allocation. The first are not when they are mixed together in a priority is to go with a non-mixing QAM of QAM. As a result, inefficiency tends to arise the same stream type. The next priority is to when streams are mixed together. Both go with an empty QAM. Then go with a busiest and least-busy algorithms tend to mixing QAM. The last resort is to create create mixing QAMs (i.e., both SD and HD another mixing QAM – going with a non- streams carried by the same QAM) as the mixing QAM of the opposite type. joint process of the two stream types is a mixture of two random processes. Naturally If there are multiple QAM modulators our improvement over the current methods is available within the same state class, priority in the direction of minimizing the probability is given to those QAM modulators that have of a mixing QAM. a larger likelihood of becoming a non-mixing QAM or an empty QAM once some streams The state of each QAM modulator can be start to drop. This implies the following categorized into four possible types: rules:

• It is entirely empty. If multiple non-mixing QAMs are • A mixture of SD and HD streams are available to a stream request of the same occupying it. stream type, priority should be given to the • Only SD streams are occupying it. busiest non-mixing QAM, because other non- • Only HD streams are occupying it. mixing QAMs have a higher likelihood of being empty. Mathematically we denote these four types accordingly by defining a state function If multiple mixing QAMs are available to as: a SD or HD stream request, priority is given to the busiest mixing QAM, because other mixing QAMs have a higher likelihood of  , if qi = 01  being non-mixing or empty.  ,2 if q xryrx ihisii y i ≠≠+= 0,0, qS ii )( =   ,3 if q xrx ihii ≠= 0, If multiple non-mixing QAMs are  available to a stream request of a different  4 q, yryif ihii ≠= 0, type, that is if a stream request will have to

create a new mixing QAM, priority is given where x and y are positive integers i i to the least busy QAM, because it has the representing the number of SD and HD highest likelihood of becoming non-mixing streams occupying QAM modulator i. In the again. above four states, we call a QAM in state 1 an empty QAM. We call a QAM in state 2, With these rules explained, the selection that is i (qS i ) = 2 , a mixing QAM. QAMs in algorithm of a particular QAM modulator by state 3 and 4 are called non-mixing SD and a new stream request is then based on the HD QAMs respectively. following sequence. We take a SD stream Since our algorithm is based on the request as an example. principle of minimizing the probability of mixing two types of streams within a QAM modulator, it is then straightforward to 1. Identify a set of I , s.t. Q ≥− rq si for alternative algorithms. If our algorithm Ii, i ∈∀ Ii, generates least blocking under the same load, 1.1 If I is empty, reject the stream it would then verify the theoretical result. request; Proposition: The QAM allocation 2. Identify a subset of J , ⊆ IJ , s.t. algorithm described above is more efficient than the busiest and the least-busy 3; j)(qS ∈= J, jj algorithms. 2.1 If J is empty, go to the next step; 2.2 If J has multiple elements, select Proof: Suppose the system has only * Q-qMin j = arg Q-qMin j ; empty and non-mixing QAMs. At this point, ∈Jj * the system has capacity to accommodate new 2.3 If there are multiple j , select streams. As more streams are added to the randomly among j* ; system, the algorithm can only incur mixing when the last QAM is called for to meet the 3. Identify a subset of J , ⊆ IJ , s.t. demand. In other words, mixing occurs when the system is close to full capacity except the jj 1; j)(qS ∈= J, last QAM. Note at this point, all other 3.1 If J is empty, go to the next step; QAMs are non-mixing. Let j denote the only 3.2 If J has multiple elements, select mixing QAM. When it is full, Q − < rq j* randomly; hj implies <−=− rqQqQ . This is ∑ i hj 4. Identify a subset of J , ⊆ IJ , s.t. because all other QAMs are non-mixing, and each type of stream is modulus in its own bit jj 2; j)(qS ∈= J, rate. Therefore there would be no 4.1 If J is empty, go to the next step; inefficiency according to our definition. 4.2 If J has multiple elements, select * Therefore our algorithm is more efficient. Q-qMin j = arg Q-qMin j ∈Jj 4.3 If there are multiple j* , select On the other hand, mixing of QAM in the * busiest and the least-busy algorithms is a randomly among j ; random event. Therefore inefficiency is likely to result with a higher probability. 5. Identify a subset of J , ⊆ IJ , s.t.

jj 4; j)(qS ∈= J, CONCLUSIONS 5.1 If J has multiple elements, select * This paper presents an alternative QAM Q-qMax j = arg Q-qMax j ; ∈Jj resource allocation algorithm to 5.2 If there are multiple j* , select accommodate a mixture of SD and HD VOD randomly among j* ; streams. Our analysis indicates the need to first design the stream bit rates so as to be We conclude this paper by presenting the modular with respect to the full QAM major result of the paper as in the following capacity, as well as to be modular with proposition. The result waits to be verified respect to each other. The subsequent by simulation tests. The test scheme will allocation algorithm then calls for avoiding generate two arrival random processes and the mixing of different types of streams to the document blocking rates for the three extent possible. This principle is very much like the principle of ocean freight container shipment, where modularity is critical in making full and efficient use of the ship capacity.

In the future when there are more types of streams in the same QAM set (e.g., alternate codec versions like MPEG 4-AVC or VC-1, and each with an SD or HD version), one should consider the modularity for selecting the bit rates for these. Additionally if we consider supporting VBR streams instead of CBR streams. Some parameters may need to reflect modularity, start time difference in QAM streams, and value of the streams.

PREPARING FOR THE NEXT GENERATION OF VOD TECHNOLOGY – A CONTENT PROVIDER ’S PERSPECTIVE

Dave Bartolone, Vice President, Technology TVN Entertainment

Abstract per title. Then, in October 2002, TVN began including VOD content from cable network Video on Demand has advanced from a providers. We ended the year with simple process of managing 50 hours of approximately 250 hours of free On- movie content with 7 day lead times to a Demand content with average run times of robust collection of long form and short approximately 54 minutes each. Today, the form content with lead times reduced to TVN network distributes over 3000 hours of hours instead of days. Add to this the next content a month from these and other VOD generation of VOD functionality that will categories with average run times of provide capabilities that could increase the approximately 49 minutes each. amount of assets to be distributed and managed significantly putting greater Delivery lead time is defined as the demands on providers’ Asset Management amount of time that a title must be pitched and Distribution Systems and processes. from TVN and caught by a VOD site before the view start date of that title. In the This paper explains and illustrates how beginning, having only feature and library VOD has changed in the past 4 years and content to deliver, the VOD delivery process where it is headed based on certain trends required delivery lead times ranged from 7 and new capabilities introduced by the next to 10 days. As more time sensitive content, version of the CableLabs VOD specification. such as current events, highlights from This paper examines the impact these recent sporting events, etc., was added, we advances will impose on content providers now have some delivery lead times reduced and how they can prepare for it in the to as little as 6 to 12 hours. future. Most of TVN’s content growth is a result of adding over 70 content providers to the BACKGROUND mix of aggregated programming available through TVN. Under the current CableLabs In the past 4 years, TVN has seen the 1.1 VOD asset structure, where a package amount of VOD content distributed grow at contains one metadata file, one to two a steady and predictable pace. In April MPEG files and possibly a graphics file, this 2001, TVN began pitching VOD content to volume equates to between 5,000 and its first commercial VOD site with 10,000 files a month to be distributed and approximately 50 hours of movie content managed by TVN. If you consider that at with an average run time of around 100 any one time, these 5,000 to 10,000 files minutes per title. That was soon followed by may be multicast to between 80 and 100 the launch of approximately 300 hours of sites at once, that equates to an average of SVOD content from the premium networks 675,000 distributed files managed per with average run time of around 70 minutes month. Knowing the current simplicity of a VOD Breaking up Asset packages - An offering that contains a single version of a intended feature of the new specification is title at a single price, we have been able to to allow a content provider to break up a accurately predict the rate of increase and CL1.1 asset package into individual scale up our support systems and operations elements and assign individual window accordingly. Our next task, however, is to dates to each element. Each package examine the new capabilities introduced by element can now be introduced to a VOD the next generation of VOD and plan system independently and live on its own. accordingly. For instance, a content provider would have the capability to efficiently distribute and NEW CAPABILITIES FOR VOD display the preview or poster art of an offering before the movie is made available Currently, the majority of VOD for viewing. installations operate under the bounds of the CableLabs 1.1 VOD specification (CL1.1). The diagram below shows the difference The next version of the specification, between the capabilities using CL1.1 versus CableLabs 2.0 (CL2.0), will introduce much those capabilities using CL2.0. needed flexibility in forming, distributing and offering content. Following are a few examples:

CL1.1 ASSET PACKAGE TRAILER MPEG POSTER MPEG MOVIE MPEG

CL2.0 GROUP ASSETS TRAILER MPEG POSTER MPEG MOVIE MPEG

Ts - 2 WK Ts - 1 WK Ts Ts + 1 WK Ts + 2 WK

Movie License Window Under the CL1.1 scenario, the viewable This simple but useful concept will allow window dates of each element of the asset for the promotion of upcoming VOD titles package opens at the same time at Ts. with the goal of increasing awareness and However, under the CL2.0 scenario, in this buy rates for the movie asset. example, the preview and its metadata is introduced at Ts – 2 weeks, the poster art Using Playlists to Insert Ads – The and its metadata is introduced at Ts - 1 playlist concept has been around for many week, and the movie asset is introduced at years and is a very versatile tool that will Ts. allow content providers, and eventually consumers, to specify a list of individual MPEG files and a sequence in which they should be played out. This one feature alone attribute of each piece of content such as will create an enormous opportunity to genre or actor. In fact, the person defining a dramatically change the way VOD playlist can pick from a number of assets programming is being offered today. Even that have been placed on the VOD server, though there is more work to be done to and, within the bounds of business rules define business rules, resolve potential agreed to by the Content provider and the billing system issues, etc., playlists offer a MSO, create a multitude of different offers. huge potential in the ability to create unique and appealing offers to consumers. One For example, you might have two ads, opportunity for both content providers and two movies, and two promos that a content MSOs is the ability to gain additional provider has placed onto the VOD server. In revenue through placing advertising within a Offer 1, you can combine one of the movies VOD offering. Additionally, a single asset and two promos to make an offering. Or, in can be shared across many playlists allowing Offer 2, you can combine a different ad with for create marketing and pricing discount a different promo and two movies to create a schemes such as “two for ones” and “many “two for one” offering. for one” offerings based on a common

Inventory at VOD Server

ADVERTISEMENT A MOVIE A PROMO A

ADVERTISEMENT B MOVIE B PROMO B

OfferCL1.1 1 - Traditional ASSET PurchasePACKAGE with Advertisement and Promos

ADVERTISEMENT A MOVIE A PROMO A PROMO B

OfferCL1.1 2 - "Two ASSET for One" PACKAGE with Advertisement and Promos

PROMO A MOVIE A ADVERTISEMENT B MOVIE B

This is an example of a static playlist that random order. Furthermore, since a playlist has been pre-determined prior to the view will be an asset having their own window start date assigned to the playlist. Future dates, different barker videos can be applied versions of this feature may include the at different times of the day or week. ability to create dynamic playlists based on subscriber viewing behaviors or stored Adding Chaptering Information – Similar demographic information. to a DVD presentation, an intended feature of the new specification is to allow for each Other uses for playlists include the asset to be indexed and then have chapter formation of both category and main page information and chapter graphics applied to barker videos. In this scenario, instead of the indexed locations. Once a subscriber has editing a series of movie trailers into one navigated to and ordered a particular title, video that then gets encoded, a playlist this feature will allow the viewer to now enables content providers and MSOs to navigate within the asset. Viewers would reuse existing movie trailers to be played also be able to skip past or skip to chapters back, as a barker video, in either a set or as desired. Managing Menu Categories – Currently, of hours. In other words, it may take four content is displayed in menu categories with times the effort to distribute and manage no regard to the display order other than four 15 minute titles than it does to alphabetically by title. An intended feature distribute and manage a single one 1 hour of the new specification is to allow content title. By studying the trend of VOD providers and/or MSOs to define a list order operations using the current CL1.1 of the menu category contents as well as specification over the past three years, we time durations for the listing within the found that the average run time per title has menu category. been steadily decreasing which inversely increases the number of units per hour of Adding Keywords – In an environment content to be distributed and managed. where the amount of content on a VOD system exceeds the practical navigational The following chart shows TVN data ability of the subscriber, keyword search gathered over the last three years illustrating functions will become an attractive feature. how the average run time per asset has been In this scenario, a user can enter keywords, decreasing. In December of 2002, TVN perform a keyword search, and then select managed 2100 hours of in-window assets from a list of results that might not have with an average run time of 57.11 minutes otherwise been displayed on the VOD user resulting in approximately 2225 titles. In interface. December of 2003, TVN managed approximately 2650 hours of in-window FUTURE TRENDS assets with an average run time of 49.71 minutes resulting in approximately 3200 New Requirements based on Current titles. In December of 2004, TVN managed Capabilities approximately 3500 hours of in-window assets with an average run time of 48.78 When considering the challenges of minutes resulting in approximately 4200 Asset Management and Distribution, content titles. This equates to a 17% decrease in providers must consider the number of files, average run time over the three year period. or units, distributed rather than the number

VOD Trends

4500 58

4000 56 3500 54 3000

2500 52

2000 50 1500 48 1000 Number of Hours/Units

46 Average Run Time (min) 500

0 44 Dec '02 Dec '03 Dec '04

Total Hours Total Units Average Run Time

Another trend that we have noticed over limitations introduced during the past year is a decrease in the delivery implementation. If we look at the most lead times. This is due to the introduction of basic, easiest to implement playlist, it would time sensitive content such as sports consist of one or two promos together with a highlights and current events. main movie asset. In this scenario, there are no dependencies on back-end process such If these trends continues, which we think as those associated with ad-viewing they will, and there are no new capabilities confirmation and reconciliation. Using this introduced, then an unprepared Asset example, given a slight increase in total run Management System will be challenged to time, up to two additional metadata files and distribute and manage this increased unit two MPEG files will need to be distributed volume of content quickly and efficiently. and managed. The result, again, is at least Adding on top of that the new intended two fold increase in the number of units to capabilities introduced by CL2.0, you may distribute and manage given a similar run find an exponential increase in the number time. of units per hour to manage. Adding Chaptering Information – New Requirements Based on New Although the indexing tags for the location Capabilities of each chapter will be contained within the existing metadata file, the addition of Following is a review of some of the optional chapter graphics will increase the increased capabilities introduced by CL2.0 number of units to be managed in direct from an Asset Management and Distribution proportion. perspective: MANAGING ASSETS Breaking up Asset Packages – Using the example of a CL1.1 asset package New Demands for Asset Management containing a movie file, a preview file, and a Systems metadata file, the management of the asset package consists of the management of three It may appear obvious that the new individual files. Breaking up the package capabilities introduced by the new CL2.0 into multiple distributed elements of the specification also introduces new demands preview one week, the art work two weeks for Asset Management and Distribution later, and the movie asset as the final systems. With asset packages being broken delivered asset will result in the creation, apart, tracking the relationships of these distribution, and management of six separated assets will be a key feature of any individual files. The result is a two fold future Asset Management and Distribution increase in the number of units to distribute system. New systems must keep track of and manage given the same run time. each package element each having their own window dates that must all overlap at some Using Playlists to Insert Ads – Under the point in order to produce a viable offering to current proposed specification for playlists, a subscriber. Furthermore, as individual there is no technical limitation to the elements of playlists are identified, their complexity and resulting amount of assets viewable window dates must, at minimum, that can be strung together to form a playlist. match the start and end view dates as However, there will no doubt be practical defined in the playlist. What Content Providers can do to Prepare • Adding Keywords – Start creating and adding keywords to your Although the timeframe for content. implementing these new features is • Chapter Graphics – Start identifying uncertain, it is important for content and storing graphics that can be used providers to prepare for them in advance. to identify chapter locations. Following are a few suggestions: CONCLUSION • Asset Management - Approach the new capabilities from an Asset The new capabilities brought on by the Management perspective paying introduction of CL2.0 are exciting and are close attention to the challenge of bound to launch the VOD product into its managing more units per hour and next generation of usefulness. However, one what may seem to be disparate assets must not underestimate the demands the new that come together on a VOD server capabilities will have on Asset Management through the use of playlists. and Distribution systems. The sooner • Segmenting Files - For the short content providers can intelligently anticipate term, and only after playlists are the use of CL2.0, the better they can prepare implemented, consider separating by upgrading or selecting an Asset content into segments so that other Management and Distribution system or elements can be inserted in between. service that is capable of satisfying these new demands.

RF FINGERPRINTING: AN OPERATIONALLY EFFECTIVE METHOD TO REDUCE CABLE TELEVISION SIGNAL AND EQUIPMENT THEFT

Lee Pedlow Sony Electronics, Inc.

Abstract initiative for digital television, as mandated by the FCC. Presented is a robust method to self-detect the unauthorized relocation of digital cable As the cost of implementing digital television appliances, especially one-way or decoding capabilities in consumer products CableCARD-based devices, as a deterrent to rapidly declines and the prevalence of digital signal (service) and MSO-provided programming on cable television systems equipment theft. The method offered has grows, the cable industry is marching toward high resolution, yet requires no additional removal of all remaining analog television hardware to be added to the products in services from their systems to reclaim which it is implemented. Implementation of spectrum, reduce operational costs and the concept uses an innovative application of reduce signal theft. resources already present in all digital cable compatible devices plus the real-time In the foreseeable future cable operators analysis of characteristic data obtained by will need to supply their existing customers the subscriber device through direct having legacy analog televisions, VCRs, etc. observation of its environment. large quantities of digital converters in 3to

maintain continued operation of those analog A system for automated device devices in an all-digital network and as part management is also presented wherein of the operator’s compliance with federal subscribers could self-activate attached regulations. devices without need for manual intervention by the cable system operator under normal circumstances. Industry estimates indicate that there may be four or more legacy analog devices attached to the cable system in a typical household in addition to existing digital cable INTRODUCTION converters for premium service access and CableCARD enabled products. Because of Digital Cable Appliances the sheer volume of digital converters that Digital cable television appliances are the cable operators will need to deploy in becoming mainstream devices in the modern support of the analog devices presently in home. These devices may be stand-alone their subscribers’ homes and the fact that “set top boxes” that are either leased from the analog devices connected directly to the cable operator or purchased by the consumer cable network do not pay for advanced through retail channels. Alternatively, the services such as electronic program guides, functionality of a digital cable appliance may video-on-demand or pay-per-view services, be integrated directly into new television cable operators have no method to recover receivers as part of the “plug and play” the huge additional capital outlay required for supplying the advanced, two-way digital cable signal either through an unauthorized cable boxes currently available. connection by tampering or because there hasn’t been a costly dispatch of the cable As a result, attention is now focusing upon operator’s field personnel to the premises to providing very inexpensive, one-way digital implement a disconnect will no longer suffice converters for this purpose, delivering for basic tier customers to receive services current analog subscribers like-for-like for which the cable operator is not digital service at a significantly lower cost to compensated. This also applies to new the operator than would be encountered using digital television receivers if the owner has the presently available advanced two-way not obtained a CableCARD from the cable devices. The cable operators, for regulatory operator and had it electronically authorized and other reasons, intend to provide these for service. one-way converters at no additional cost to their subscribers and believe that the cost of The typical conversion scenario for the providing these devices can be more than all-digital transition would be for a cable offset through the recovery of valuable cable operator to upgrade a headend serving a spectrum, elimination of signal theft and community or city to carry basic tier content reduction of operational costs such as truck in digital form in addition to the analog rolls for service connect/disconnect. format presently carried. Next, all current

two-way devices deployed for decoding These simple digital converters are premium digital services are provisioned intended only for the most basic of service with new channel maps, directing them to tiers, ones that are presently delivered in receive only digital forms of content, analog form and therefore have been left including the new digital replacements for unprotected against unauthorized reception, the analog tier instead of the present mixed unlike the current premium services which format. In parallel, the operator will begin employ modern digital encryption. A distribution of the new, low-cost one-way conservative estimate is that more than one converters to existing subscribers based upon third of the channels carried in modern cable the number of cable outlets in the home that systems are presently analog basic services. are reported by the subscriber as connected to A May, 2004 press release from a major a legacy analog device (VCR, TV, etc.). cable system operator, Cox Communications, There is no way for the cable operator to indicates that roughly 11.5 million U.S. determine the analog device count in a home households steal these cable services each without either surveying the subscriber or year at an industry cost of $6.5 billion in lost performing a physical audit inside their revenue annually. premises. The operator will likely deploy these new converter devices en-masse as The transition of the basic subscription each node served by a cable headend is tiers from analog services to exclusively converted from mixed analog/digital format digital services having encryption applied to all-digital through the removal of analog will eliminate most of the present forms of services. A network node typically serves signal theft that occur because these new from 500 to 2000 customers and the digital converters will be individually converters must be available to all addressable by the cable operator. Unlike subscribers in a targeted node prior to today, merely having physical access to the cutover in order to avoid service interruption QAM Demodulator

I Channel RF Tuner X Filter Mixer

Forward Cable Input Input SAW Local Adaptive Phase MPEG A-D Error VGA X VGA Oscillator Equalizer Recovery Transport Output Filter Mixer Filter Converter Correction

Q Channel Local X Filter Oscillator Mixer

AGC & Tracking Loop Control

Tuned Frequency Command Figure 1. Generic digital cable network interface

While the introduction of all-digital Other one-way devices that attach to the services and low cost digital converters cable network also suffer from the same would seem to address all the issues of vulnerability. The new CableCARD device unauthorized viewing and signal theft, a new for digital television is an example of such a opportunity to deprive cable operators of fair device that suffers the same susceptibility to payment for service emerges, made even unauthorized redirection. Two-way devices, more challenging because these low cost such as existing digital cable decoders for converters are likely only one-way devices,. premium services, are less likely to suffer When subscribers are contacted to determine from this issue because there are means to the quantity of converters necessary for electronically detect the location of these supporting the analog appliances in their devices through headend interrogation and home, the subscriber may intentionally response, with the time delay to respond “over-report” the quantity of analog being measured to determine the cable appliances in the home. They can later distance to the device. In such an provide the excess converters received from application, the response time values for two the cable operator to friends, family, etc. to devices assigned to the same address can be “split the costs” of basic cable service. This compared for similarity and physical is but one example of how one-way proximity inferred. While such a method converters can be redistributed without the provides some protection from unauthorized knowledge or consent of the cable operator in relocation, it will be shown later that it is an order to deny the operator of payment. inferior method, suffering from poor resolution and other problems. Since these new devices are issued by the operator to a valid subscriber, they remain Sony’s has developed technology to authorized, but the cable operator is deprived address the issue of detecting unauthorized of the full value of subscription revenue relocation of one-way equipment consigned because the devices are present in locations to the subscriber by the cable operator. The other than the home of record for the method in which it is accomplished uses authorized subscriber and the operator is resources already available in the appliance therefore not compensated by the additional, and adds no additional hardware cost to the unauthorized viewers. product.

The Network Interface for Digital Cable range of valid cable audio/video services (54MHz to 863MHz) and the variable gain Regardless of the end use of a particular amplifier (VGA) is automatically adjusted so device, all appliances attached to the digital that the RF signals passing through the tuner cable network share a common front-end and demodulator remain at optimum levels at topology. The elements that make up the all times. The final stage of the RF tuner is network interface are available from a the surface acoustic wave (SAW) filter, number of different manufacturers and may which is an electromechanical device be offered in different configurations designed to only let a small band of signals featuring flexibility, integration with other centered at the IF value pass and all other RF elements, support of multiple interfaces, etc. energy to be heavily attenuated. The SAW to serve as the differentiation between only passes a standard 6 MHz wide channel products. and effectively rejects all others. The signal that emerges from the tuner is therefore only The typical network interface is shown in the channel carrying the service of interest Figure 1. The cable network interface and it has been downconverted to a fixed, consists of two major sub elements, the RF standard (IF) frequency for processing by the tuner and the QAM demodulator. The QAM demodulator. function of the RF tuner is to receive all signals on the digital cable system and to The QAM demodulator receives the exclude all but one desired RF channel, incoming 6 MHz wide signal at the containing the digital service of interest. The intermediate frequency, typically 44 MHz, method used to select the desired channel is and again amplifies it to a constant and called heterodyning and this process is used optimum level through a second variable gain to convert an entire block of incoming amplifier. The VGA is automatically signals to a lower intermediate frequency adjusted by a closed control loop within the (IF), with the signal of interest centered on a QAM demodulator. The signal is then fixed, constant value, which is passed processed by an analog to digital converter through a fixed, narrow filter to eliminate the (ADC), which converts the incoming stream unwanted carriers. The QAM demodulator of time-varying voltages to a serial stream of processes the tuner’s IF output, converting it binary bits representing the voltage levels of to an error free digital stream of MPEG the signal at discrete time intervals. The transport data carrying the compressed audio ADC typically has 10 or more bits of and video services. resolution.

Inside the RF tuner the local oscillator, The digital stream is then split into two controlled by the host processor, varies in components, the in-phase component (I) and frequency such that the nonlinear the out-of-phase component (Q). The Q term combination of the local oscillator signal and is used because the signal is in quadrature the incoming spectrum from the cable with respect to the I signal, meaning it is network inside the mixer results in the signal shifted 90° in phase. Phase separation occurs of interest emerging from the mixer centered simultaneously with down conversion to a at the fixed, lower intermediate frequency. baseband signal, where the lowest frequency The IF typically might be selected to be a is 0Hz (DC) and highest frequency 6 MHz. value such as 44MHz. The input filter This is in contrast to the incoming 44MHz IF eliminates extraneous signals outside the signal, which has its content symmetrically centered ±3 MHz about the IF signal. The architecture is common between them and downconversion is accomplished through the takes the form of a classic feed use of a balanced mixer and the I-Q forward/feedback digital filter. A typical separation occurs because one of the two digital filter for such a purpose is shown in halves of the balanced mixer has a local Figure 3. oscillator signal output that is shifted 90° in bm phase relative to the signal applied to the an b1 other half of the balanced mixer. The outputs f(k) -1 -1 -1 -1 -1 y(k) + Z Z Z Z Z b0 + of the balanced mixer, I & Q, are then passed through identical channel filters that provide an-1 an-2 the appropriate shaping and attenuation of am a1 undesired processing artifacts occurring a0 above the 6 MHz passband. Figure 3. Adaptive equalizer

TX RX Post Equalizer The structure of the filter is centered about -1 Amplitude Amplitude Amplitude a cascaded chain of delay stages (Z ), where the discrete time samples of the voltages seen

fHZ fHZ fHZ at the demodulator input, converted to binary (a) (b) (c) digital from by the ADC, are successively Figure 2. Digital cable channel spectrum stored. The output of each delay stage tap, in Next, an adaptive equalizer is applied to addition to feeding the next cascade, may be the outputs of the channel filters. The fed back to the input or fed forward to the adaptive equalizer is an automatically output. The tap feedback may be may be in self-varying digital filter network that conjunction with feed forward and either path continuously alters its filter characteristic may be employed exclusively on a tap-by-tap (shape). Its purpose is to compensate basis. Each feedback or feed forward path automatically for echoes, reflections, has associated with it an independent dispersion, tilt, intersymbol interference and coefficient term. This term (a & b blocks in other distortions that alter the signal from its Figure 2) may provide amplification or ideal, original form (Figure 2a) as it is carried attenuation of the tap output, depending upon by the cable operator’s hybrid fiber-coax the value of the coefficient. Because the distribution network to the receiving device equalizer is adaptive, the coefficients (Figure 2b), possibly over very long dynamically change under the control of a distances. By approaching the ideal of a microprocessor or state machine. The values matched filter, waveforms distorted through are varied based upon the characteristics of the communication path to the subscriber can the equalizer output, as seen by the next be recovered and the data error rates for processing stage, phase recovery. Typically transmitted data reaching the phase recovery a least mean square (LMS) algorithm is used element (derotator) significantly reduced. to vary the tap values and converge upon the This allows the system to operate optimal solution. Adaptive equalizers in successfully under non-ideal conditions, QAM demodulators vary in implementation which are typical of real world applications. between manufacturers. One design may The details of how the adaptive equalizer have a total of 22 taps, where another may is realized differ between different QAM have a total of 40 taps – 16 feed forward and demodulator manufacturers. The general 24 feedback.

Equalizer Coefficients

Weighting Cable Plant HFC Transmission Digital Cable Other & Fingerprint Fingerprint Pass/Fail Headend Environment Network Interface Data Threshold Value Comparison Function Memorize Command AGC Memory Values Distortion Dispersion Reflections Digital Cable Network Appliance at Subscriber’s Premises Figure 4. RF Fingerprint system The output of the adaptive equalizer is operator inserted into the corresponding then processed by the phase recovery block, QAM modulator at the headend for also known as a detector or derotator. The transmission. purpose of the detector is to decode the combination of I and Q signals into a single Further processing is done to decrypt, data stream. The detector is able to expand demultiplex, decompress and convert the the incoming data streams by a factor of content to a form suitable for display on a log2(Modulation Order). This expansion is a television. These steps, while vital to the factor of 6 for 64-QAM and 8 for 256-QAM, proper function of a digital cable appliance, the two typical forms transmitted in digital are beyond the scope of this document. cable. This expansion is the reason high transport data rates can be efficiently carried SONY’S “RF FINGERPRINT” in relatively low spectrum bandwidths that TECHNOLOGY appear to violate the Nyquist criterion. The coefficient values of the adaptive equalizer The ability to detect changes in location of and the frequency setting of the QAM a one-way digital cable receiving device is modulator local oscillator are both controlled based in large part upon the adaptive by a microprocessor or state machine based equalizer. The equalizer, as indicated, acts as upon the success of the detector to “lock” i.e. a matched filter to the communications to recover valid data. channel. As a result, the values contained within the equalizer’s coefficients can be The last processing stage, the forward mathematically manipulated to show the error corrector (FEC), applies a variety of transfer function of the communications algorithms to the raw recovered digital cable channel that influences signals passing data stream to reduce the likelihood that any through it. Stated differently, the values of of the data has been corrupted in addition to the coefficients, taken as a set, represent at a formatting it appropriately for recovery of specific point in time the sum total video and audio services as an MPEG knowledge of all mismatches, reflections, transport stream. It is in this stage that de- phase variations, gain variations, echoes and interleaving, Viterbi (trellis) decoding, de- other perturbations of the transmission media randomization, Reed-Solomon error upon the transmitted signal. The fact that the correction and MPEG formatting occur. QAM demodulator is able to achieve and Some overhead data unique to the operation maintain signal lock under a given of these stages are removed from the stream environment validates that the state of the so that the final MPEG transport emerging equalizer at that time is such that it from the demodulator is identical in form, accurately reflects the knowledge of the content and data rate to what the cable plant’s effect upon the system so it can negate those effects and lock successfully. If we let an equalizer coefficient be The tolerance to a suboptimal equalizer represented by a±jb, then H1, the matrix of all configuration is low, given the small vector equalizer coefficients representing the state error radii for either the QAM-64 or of the system at one point in time, i, can be QAM-256 formats used in digital cable. The represented by: vector error radius is the composite of effects ab00 due to both amplitude and phase distortions  H1, i = MM (1) upon a received signal.  abnn

Since the filter coefficient set is directly Likewise, if we let the gain value of one representative of the transmission of the multiple nested AGC loops be environment, it responds dynamically to any represented by k, then H2, the matrix of all changes in that environment. The low order AGC coefficients representing the state of the feedback taps are most affected by high system at one point in time can be frequency trends, such as impedance represented by: variations at the connection or connector on k0  the back of the appliance, reflections within H2, i = M  (2) the cable from the house splitter(s), etc. The  kn middle taps are more predominantly affected by variations in the characteristics of the If one were to capture the equalizer tap cabling to the tap and distribution amplifier, coefficients and AGC data from a digital while the highest order taps are sensitive to cable appliance, then applying an algorithm channel tilt, dispersion, etc. This data, when to allow the weighted summation of the combined with the AGC information which coefficients a weighting function, based upon indicates total gain required for a constant the expected statistical variance, to create a signal level input, provides the basis for a single scalar, a unary value representing the very characteristic fingerprint of the unique “fingerprint” of the environment of environment where a specific cable appliance the device could be expressed. The threshold is installed. and weighting functions could be made unique to a particular operator and are kept secret to reduce the likelihood of tampering. Research by Sony indicates that the equalizer is so sensitive to such changes that The algorithm for these operations then one can distinguish between the short cables looks like: coming from different ports of an RF splitter Fingerprintii= Y,(HH1, 2, i) (3) to a bank of attached digital cable appliances

fed by a single common source. In this case, This fingerprint value is evaluated and the devices were all within one meter of each stored in the digital cable appliance memory other and had identical cable lengths, yet the upon receipt of a command message, such as values observed for each device were unique an EMM, from the cable operator. The and over time were relatively invariant. stored value should be secured through encryption and digitally signed to detect tampering. The Fingerprint Algorithm the Sony cable test network, an elaborate system serving the entire San Diego Most of our current research on the RF corporate campus closely emulating a Fingerprint concept is focused upon refining commercial cable television network. the algorithm used to calculate the RF Devices were sampled at three minute Fingerprint, specifically the coefficient intervals and remotely tuned to services on values of the weighting matrix, based upon multiple frequencies. The data gathered was experimental observation of actual cable subsequently compiled into a large database appliances in-situ. The generalized form of for further analysis. the algorithm to calculate the fingerprint

value based upon one manufacturer’s One of the first items confirmed was the implementation of QAM demodulator is: repeatability of the equalizer configuration

39 39 for a given subscriber drop and cable device. Fingerprint=•+•+∑∑()() Tapi, a W i,, a Tapi b W i, b SF (4) Repeatability was judged based upon the ii==00 measurement of the standard deviation over a four day period. Figure 6 shows the results Where W represents a value in the of one such data collection experiment where weighting coefficient matrix associated with two different digital cable devices were a particular equalizer tap and SF represents attached to the test network for a statistically the equalizer scale factor that is used to meaningful period. At the end of that period normalize all equalizer tap coefficients. the two devices exchanged locations and data In order to determine the appropriate collection was restarted for the same period. values for the weighting coefficients, tools As can be seen in the graph, not only is the were created by the team to remotely collect data generally quite repeatable for a given equalizer coefficient and other data from device, but the dispersion tracks the location, digital cable appliances installed throughout not the device. 10000 Population: 23557 Samples Interval: 3 minutes Duration: 4.83 Days

1000

100

10 Standard Deviation Standard

1 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85

0.1 192.168.40.100 192.168.40.97 192.168.40.97 Swapped 192.168.40.100 Swapped 0.01 Equalizer Tap Figure 5. Positional uniqueness

100 Population: 23557 Samples Interval: 3 minutes 80 Duration: 4.83 Days

60

40

20

0 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 Mean Value -20

-40

-60 192.168.40.100 192.168.40.97 -80 192.168.40.97 Swapped 192.168.40.100 Swapped -100 Equalizer Tap Figure 6. Equalizer coefficient repeatability

100 Population: 653 Samples Interval: 3 minutes 90 Duration: 11.95 Hours

80

70

60

192.168.40.100 50 192.168.40.122 192.168.40.15 Mean Value Mean 40

30

20

10

0 1 4 7 1013161922252831343740434649525558616467707376798285 Equalizer Tap Figure 7. Uniqueness under worst case conditions Using the data collected during the same also require some form of two-way experiment, Figure 7 shows that each of the communication. The RF fingerprint scheme four cases (two digital cable devices in two is immune to issues plaguing methods solely different locations) possesses a unique related to cable length and works in a purely signature, allowing each to be distinguished unidirectional environment as well as from another device/location. The purpose bidirectional environments. of the weighting matrix is to selectively amplify those equalizer tap and AGC Subscriber coefficients that express “uniqueness” terms Receives New and to attenuate those terms that contribute Appliance little in the context of distinguishing devices or tend to be unrepeatable. Connect Product

Call AV&V System In order to evaluate a worst case scenario, three cable appliances were connected Enter S/N through identical one meter long cables to a No common splitter fed by the test network. The Phone # Match? QAM performance of the three test units Yes were monitored using four discrete Challenge Code frequencies at three minute intervals for a Displayed twelve hour period. The results of the Enter Code # investigation are shown in Figure 5.

No Confirmed was that even in an apparently Code # identical RF environment and close physical Match? Yes Forward Call to proximity to ensure consistency of other CSR environmental factors, the three different Activate Services devices were distinguishable based solely Memorize & Store upon equalizer and AGC coefficient values. Fingerprint

Figure 8. Digital cable appliance activation process In practice, an installation having only one meter service drop lengths would seldom be Operational Scenario seen and could actually be disregarded by the algorithm comparing fingerprint values to Regardless of the specific details avoid false alarms due to product relocation associated with the implementation of the from room to room within a home. However, algorithm to calculate the fingerprint value, it does prove that the concept is robust and an operationally practical means of deploying applicable to high density dwellings such as a system employing RF fingerprint apartments and dormitories where even two technology is necessary. Such a system must way methods, such as the one used in be automated to the greatest extent possible DOCSIS, fail. The DOCSIS method, for in order to reduce operating costs and to example, can only resolve location to within maximize flexibility. The practical 63 meters, much more than the cable length implementation of a system employing RF differential between adjacent apartments. fingerprinting is as follows: This is because DOCSIS and other proposed schemes use time delay measurement as the determining criteria. These other schemes 1. A subscriber receives a fingerprint 4. If the ANI and phone number on equipped appliance from the cable record match, the AV&V system then operator. The product contains labeling sends a control message (EMM) to the indicating “Call XXX-XXX-XXXX from appliance having the serial number the your home phone after connecting the subscriber entered by phone. This EMM device to both the cable network and commands the appliance to display, on the television for activation”. This is identical subscriber’s television screen, a challenge in concept to the process now followed for number sequence contained within the activation of home satellite television EMM message and generated at random receivers and credit or ATM cards issued by the AV&V system. The AV&V through mail by the major financial system then instructs the subscriber to institutions. enter into the telephone, the number displayed on the screen using the keypad 2. The subscriber follows the on the telephone the and to press “#” key instructions and calls the number on the upon completion. appliance label after installation, as indicated. An automated validation and 5. Upon receiving the “#” key input, the activation (AV&V) system at the cable AV&V system confirms the validity of the operator receives the call and prompts the entered challenge number and if subscriber to enter the serial number of the unsuccessful, refers the call to a customer cable appliance using the keypad on the service agent. This step validates that the telephone and to press the “#” key upon authorized subscriber is attempting to completion. activate the appliance issued to them by the cable operator at the home on record. 3. Upon receiving the “#” key input, the AV&V system confirms the validity of the 6. If the challenge sequence is entered appliance serial number. The successful, the AV&V system sends system then looks through its subscriber another EMM to the now validated database and finds the record for the appliance, commanding it to perform two subscriber issued the appliance having the steps entered serial number. It then reads the A. Activate the services authorized for subscriber’s home phone number from the that subscriber database record. Using Automated Number Identification (ANI), a non- B. Calculate the RF fingerprint for the maskable form of caller identification appliance at the present location and store used for logging calls to toll-free it in persistent memory. telephone numbers (and 911 calls), the AV&V system then confirms a match At periodic audit intervals, determined between the number of record and if either by EMM or through self-initiation, unsuccessful, refers the call to a customer which uses a timer resident in software, the service agent. This step validates that an appliance collects data and calculates an authorized subscriber is attempting to electronic fingerprint value, comparing it to activate the appliance issued to them by the reference value stored in memory. If the the cable operator. calculated value is within predetermined message for appliance reactivation. This message occurs because the appliance has determined that an unauthorized relocation Digital Cable Appliance has possibly occurred. When the subscriber Power-up or calls the displayed telephone number, the Reset AV&V process is executed and upon Calculate Current validation, the location of the device is re- Fingerprint Value evaluated. Fetch Reference Fingerprint Value from memory CONCLUSION

Yes No RF fingerprinting technology is one Values Match? element in the toolkit available to address issues of service and equipment theft in cable Continue Normal Disable All television systems. Implemented in a digital Booting Process Services decoder and coupled with simple content Display ‘Call AV&V’ encryption techniques, a complete solution OSD Message providing both the quality of service seen in Figure 9. Digital cable appliance location audit process at reboot an all-digital network as well as the system security and compensation for services delivered previously encountered only in limits, no further action is taken until the next premium digital services. All this is possible audit period. If the new value for the in the lowest possible cost, one-way fingerprint is sufficiently different from the customer premises equipment. The stored reference value, then the reference implementation of RF fingerprinting does not value in memory is updated with the new reference value. add hardware cost to the cable device and can be implemented in any digital device attached to a cable television network. Whenever the appliance is rebooted or otherwise reset, signifying a lapse in network connectivity where the appliance may have On-going development of this technology been relocated without the authorization of continues to focus upon optimization of the cable operator, the appliance collects data weighting matrices as the technology is and calculates an electronic fingerprint value, matured. One major U.S. cable operator has comparing it to the reference value stored in already specified the inclusion of this memory. If the calculated value is within technology in their current system upgrades predetermined limits, the device continues to all-digital delivery and draft specifications the booting process and services are restored. of the management and control aspects have If the match is unsuccessful, all television been recently completed. It is quite likely services are automatically self-deauthorized that within the next 18 months, a commercial on the appliance, with an on-screen message example of RF fingerprinting technology will generated and displayed on the subscriber’s be available and in the hands of the public. television screen indicating that the cable operator must be contacted at the AV&V telephone number contained within the 2. ISO/IEC 13818-1:2000, Information ACKNOWLEDGMENT Technology – Coding of moving pictures and associated audio – Part 1: Systems. The author would like to thank Eric Geneva: International Organization for Holcomb and Aran Sadja, both of Sony Standardization/International EMCSA in San Diego, for their significant Electrotechnical Commission, Dec. 2000. contributions to the development of the RF fingerprint technology as well as the tools 3. B. P. Lathi, Signals, System and Controls. and systems implemented to perform New York: Harper & Row, 1974, pp. automated data collection in order to validate 207-214 & pp. 428-456. the concept. 4. A. Bruce Carlson, Communication rd Systems, 3 ed. New York: Mc Graw- REFERENCES Hill, 1986, pp. 514-517 & pp. 550-554. 5. Edward A. Lee and David G. 1. ITU-T Rec. J.83:1997, Digital multi- Messerschmitt, Digital Communication, programme systems for television, sound 2rd ed. Boston: Kluwer Academic, 1994, and data services for cable distribution. pp. 442-550. Geneva: International 6. Richard E. Blahut, Digital Transmission Telecommunication Union, Apr. 1997. of Information. New York: Addison- Wesley, 1990, pp. 159-170.

SEAMLESS MOBILITY BETWEEN HOME NETWORKS AND CABLE SERVICES

Jay Strater and Gordon Beacham Motorola

Abstract video, and voice. Nonetheless telephone companies are a competitive threat since Cable operators are now offering landline teaming up with satellite-TV providers to voice service in addition to video and high- offer their own voice, high-speed internet, and speed data. However, cellular service is video service bundle. The three largest local quickly becoming another important service telephone companies also own significant for cable operators to offer to ensure that they portions of the major wireless phone carriers, stay in a competitive position with telephone giving them a leg up on the cable providers by companies. also offering cell phone service. Consequently, as enticing as landline voice This paper provides an overview of service is, cellular service is quickly seamless mobile communications and its becoming another important service, a potential use by cable operators. It includes a “quadruple play”, for cable operators to offer description of seamless mobile applications to ensure that they stay in a competitive and their value, network approaches for position. seamless mobility, and considerations for power management, Quality of Serivice Cable operators are interested in offering a (QoS), security, and Network Address mobile service with access to cellular and Translation (NAT) traversal. The paper wireless local area network (WLAN) service concludes with a discussion of other seamless using seamless mobility. In general terms mobility concepts revolving around the seamless mobility is an approach that allows connected home. users to roam between application domains and communication networks without being aware of the underlying mechanisms that INTRODUCTION enable them to do so. This includes the scenario where a user moves between Through a careful development and environments where different networking standardization processes culminating in Data capabilities are present, but the network Over Cable Service Interface Specification provides negotiation to allow for seamlessly (DOCSIS) in the late 1990’s, cable operators transparent access. This differs from today’s have very successfully deployed high-speed environment where handover between data as a key element of the broadband heterogeneous networks is not supported in service package. Cable operators are leaders most cases, and users are required to stop one in residential broadband, and are now communication service and initiate another leveraging this penetration to provide landline between different networks. voice service using Voice-over-IP technology (VoIP) to POTS phones. Seamless mobile communications between cellular and WLAN environments refers to a Landline voice service provides the “third service in which a user receives voice and leg” of cable’s current “triple play”, data service on a dual-mode mobile handset completing a service mix consisting of data, when inside or outside a WLAN.1 Service value in an enterprise application longer term. within a WLAN is provided via VoIP over A few differences exist between residential unlicensed WLAN spectrum and DOCSIS in and enterprise applications. These include the the case of a cable plant. Service outside the number of access points, the call control WLAN is provided via circuit switch cellular technology and features, and the level of technology (GSM or CDMA) or VoIP over required security. cellular or a 3G data network. Moreover, a voice call or data connection is seamlessly Residential applications typically require handed off between WLAN and cellular only one WLAN access point (AP) in a user’s network as a user moves between them. home whereas enterprise applications require multiple WLAN AP in a user’s work location. This paper provides an overview of Having multiple APs adds to the complexity seamless mobile communications and its of a seamless mobile communications potential use by cable operators. It includes a solution by requiring micro-mobility description of seamless mobile applications management for call handover between APs and their value, network approaches for in addition to macro-mobility handoff seamless mobility, and considerations for between WLAN and cellular service. In power management, QoS, security, and NAT addition residential applications have phone traversal. The paper concludes with a service controlled from a central office discussion of other seamless mobility (circuit or packet switched) and CLASS5 concepts revolving around the connected features offered, whereas enterprises typically home. have phone service controlled by a PBX (often an IP PBX using SIP or H.323 SEAMLESS MOBILE signaling protocols) along with enterprise COMMUNICATIONS APPLICATIONS features. AND VALUE PROPOSITION What are the benefits of seamless mobile As described previously, we refer to communications? For residential users, the seamless mobile communications as a service benefits may include: in which a user receives voice and data service on a dual-mode mobile handset with • Reduced cellular bill resulting from off- cellular and WLAN capability. Moreover, loading the cellular air-interface when service in the WLAN is provided via VoIP calls are made from the mobile handset in over Wi-Fi and carried via DOCSIS in a cable the home’s WLAN2 plant. Service outside the WLAN is provided • Improved in-home coverage and via cellular technology such as GSM or reliability which is often limited with CDMA. A voice call or data connection is cellular service seamlessly handed off between WLAN and • Wireline audio quality because of a cellular networks as a user moves between higher-rate codec thanks to WLAN and them. broadband connections Seamless mobile communications may apply to residential or enterprise applications.

Cable operators are focused on residential • 2 Cell phone contracts typically bill for a bulk seamless communications initially but see number of minutes and residential VoIP service is typically an all-you-can-eat service. By switching 1 A dual mode handset here refers to a mobile handset your cell phone call to a Wi-Fi call, the minutes with functionality for cellular and WLAN operation. should now fall in the all-you-can-eat category.

• Convenience of a single mobile number For wireless carriers, the benefit of and voice mail service, whether inside or seamless mobile communications may outside the home include: • Mobile and landline voice service inter- working, e.g. allowing for a shared • Cellular capacity relief by off-loading of “family” number as well as “individual” air-interface when user is in a WLAN at mobile and landline numbers home or in the office • Ability to offer residential phone service, Studies indicate that users will still want in the case of independent wireless access to their existing landline telephone carriers (those without a wireline carrier service in addition to a mobile handset. affiliation) Consequently, residential applications should • Increased customer satisfaction from also allow for a mobile handset and landline improved in-home coverage, a key user phone to inter-work. In the example noted network quality metric above a “family” number implies a user • Improved customer retention (reduced selected mix of mobile and landline phones, churn) through unique value added allowing for group ringing and possibly line- services extension service. By contrast an “individual” number implies a unique number For cable operators, the benefit of seamless for mobile handsets and landlines. Having mobile communications may include: “family” and “individual” numbers requires • A “quadruple play” service offering of that distinctive ringing be supported. data, video, voice, and wireless as in the For enterprise users, the benefits of case of cable operators, allowing added seamless mobile communications may service bundling include: • Greater pricing flexibly resulting from increased service bundling and migration • Cost savings from eliminating use of of customers to higher revenue / margin cellular calls in the enterprise, assuming wireless offerings cellular service is available and otherwise • Increased leverage of existing broadband utilized in the enterprise infrastructure • Operational cost benefits from carrying • Ability to offer superior QoS over cable VoIP and data on a single network; infrastructure (via DOCSIS 1.1 or higher although this may already be realized if an QoS) IP PBX service is in place • Increased customer retention through • Wireline audio quality because of a bundling and possible value added higher-rate codec thanks to WLAN and services, including interworking of mobile broadband connection and landline phone services • Convenience and productivity benefit of using a single device for all The presumption is that by bundling communications, whether inside or services, a broadband carrier will present a outside the office, whether at or away customer with only one bill, and offer some from ones desk service discounting compared with offering services individually. In addition to the benefits outlined above, users may receive advanced services and, Finally for vendors, the benefit of seamless through use of a common handset, consistent mobile communications may include: services inside and outside the enterprise and home. • Increased handset, modem, and access A phone number is assigned at the CMS to point /gateway sales each dual-mode handset as well as landline • Increased core network equipment sales phones served by the CMS. Calls destined to • New opportunities for integration and a dual-mode handset’s CMS number when the deployment services handset is inside a WLAN are served directly by the CMS. Calls destined to the dual-mode In summary, seamless mobile handset’s CMS number when the handset is communications appears to offer a win-win outside the WLAN (in the handset’s cellular solution for all stake-holders. And as stated WAN) are forwarded by the CMS to the in the introduction, having a “quadruple play” handset’s cellular number. Seamless is quickly becoming a service cable operators handover between (into or out of) WLAN and will need to stay competitive. cellular WAN is provided for any calls to the NETWORK APPROACHES FOR dual-mode handset’s CMS number, and any SEAMLESS MOBILE calls initiated by the dual-mode handset from COMMUNICATIONS inside the WLAN or to a phone controlled by the handset’s CMS. An important aspect of any seamless mobility solution is providing an easy-to-use For single number reachability and service that is transparent to the user. Ease of seamless mobility to be supported as outlined use and seamless service functionality can be above, all calls must pass through the achieved by including intelligence in the CMS/mobility management element(s). network or in end user devices. The most Consequently the handset’s CMS number likely point for intelligence to be implemented must be the only number advertised to friends is in the network with varying levels of and colleagues. support in the end user devices. The following figure illustrates an example We know of three fundamentally different of a call-forwarding network architecture. network approaches to providing mobile VoIP signaling and bearer transport protocols seamless communications. They include call are shown, data is not. forwarding, Unlicensed Mobile Access CMS/MGC (UMA), and Third Generation Partnership plus Mobility Manager Project (3GPP) IP Multimedia Subsystem (IMS) technologies.

Call Forwarding Managed IP Cellular Voice and Network Data Network In a call forwarding approach, a specialized mobility management element is WLAN integrated (logically or functionally) with a call management server (CMS) serving landline phones and mobile/dual-mode Figure . Example Network Architecture for handsets in a WLAN. For residential 1 Call-Forwarding Approach applications the CMS would reside in a central office whereas for enterprise The call-forwarding approach has the applications the CMS may reside in the benefit of making maximum use of existing enterprise as an IP PBX. In either case there CMS and media gateway resources if present. is no need for changes at a cellular Mobile As such new integration requirements with Switching Center (MSC). operator back-office systems may be minimized and use of common features for The following figure illustrates an example landline phones and mobile handsets may be UMA network architecture. VoIP signaling maximized. The call forwarding approach has and bearer transport protocols are shown, data the disadvantage of imposing inefficient is not. routes via dog-legged calls that must always be anchored at a CMS, even if the CMS is far from source and destination. This inefficiency may prove costly from a PSTN Cellular Voice and IPsec Data Network interconnect perspective and also from a tunnel transcoding, performance perpective, when transcoding is required.3 Also support for seamless mobile handover from cellular WAN WLAN to WLAN is not possible when calls are initiated by a dual-mode handset in the cellular WAN to a phone outside the control Figure 2. Example UMA Network of the dual-mode handset’s CMS. Finally Architecture seamless mobility call forwarding approaches are not currently standardized, raising the The UMA approach has the benefit of concern over single supplier cost premiums requiring little change to an existing GSM and limited product availability.4 cellular infrastructure, only requiring the addition of a UNC. As such the approach Unlicensed Mobile Access offers a quick time-to-market solution for cellular operators. It also has specifications With UMA technology, GSM and GPRS that are governed by a consortium of cellular signaling and RTP and IP data traffic are operators and vendors. The UMA approach carried over an IPsec tunnel between a dual- has the disadvantage of being limited to GSM mode handset and UMA network controller systems. And because it utilizes existing (UNC) when the handset is in a WLAN. The GSM infrastructure it is limited in terms of UNC emulates a GSM base station controller offering new, advanced IP-based services (BSC), providing a gateway between WLAN (limited to what an existing MSC offer). and cellular WAN signaling and traffic. GSM Moreover a major investment in current- signaling is used to signal when a user moves technology GSM infrastructure including between a UNC and a BSC for mobility MSCs would be required for those operators management and handover purposes. The without GSM infrastructure (most MSOs) to cellular number maintained at the cellular deploy a UMA solution.5 carrier’s home location registrar (HLR) identifies the user. IP Multimedia Subsystem

5 Otherwise, an unorthodox business arrangement

would be required in which a cellular operator Mobile 3 Transcoding between WLAN and PSTN can be Switching Center (MSC) allows access from a 3rd party avoided by utilizing G.711 while in the WLAN. operator’s UNC. In any case, cable operators could 4 Even if signaling from handset to mobility manager is still receive revenue from a UMA solution offered by SIP based, unique SIP extensions are likely required to cellular operators over a cable operator’s infrastructure. support specific mobility and handoff signaling In this case the cable operator could provide the requirements. To date no call forwarding approaches cellular operator QoS on the DOCSIS network and we know of have been standardized. However charge for it accordingly. Of course cellular operators Motorola, Avaya and Proxim have developed an can also use the DOCSIS network without permission enterprise solution that they are working to move into or knowledge of the cable operator and offer a best an industry standard. effort VoIP service. With IMS technology, VoIP is provided The following figure illustrates an example via SIP signaling and RTP traffic between a IMS network architecture. VoIP signaling dual-mode handset and call control elements and bearer transport protocols are shown, data in an IMS core network when the dual-mode is not.7 handset is in a WLAN. The same applies if a dual-mode handset is in a 3G cellular network that supports VoIP over its data network. When a handset is in a circuit-switched IMS Core cellular network, circuit-switched signaling and voice are converted to SIP signaling and RTP at an IMS signaling and media gateway Managed IP Cellular Voice and respectively. To neighboring MSCs in a Network Data Network circuit-switched cellular WAN, the IMS core looks like another MSC. Yet all signaling WLAN within the IMS core network remains SIP based with WLAN and cellular call mobility controlled from the core.6 In essence, the Figure 3. Example IMS Network Architecture IMS core provides WLAN and cellular network-agnostic access control with Compared to UMA, the IMS approach has the centralized application services. benefit of being RAN agnostic and facilitating rapid deployment of enhanced packet- A cellular number is maintained at the switched services. It also has the benefit of home location registrar (HLR) associated with allowing a cable operator the flexibility of the IMS core network and identifies the user. controlling a dual-mode handset when on a However additional numbers at the IMS core WLAN connected to their broadband could be used for identifying a dual-mode network. Compared to call forwarding, the handset when in the WLAN. This allows IMS approach has the benefit of being flexibility to support “family” and governed by a set of standards (developed by “individual” phone numbers when the handset 3GPP) and avoiding route, functionality and is in the WLAN. potential performance issues of the call forwarding approach. A limitation of IMS is Note that IMS, which is developed by the that it does not make use of existing Call 3GPP forum, applies to GSM-based cellular network technologies. In contrast, CDMA- 7 Of the IMS signaling elements: the HSS provides based cellular network techologies have an IP Home Subscriber Server with AAA and Databases. core solution like IMS as directed by 3GPP2 The Application Servers (AS) include a Session Initiation Protocol (SIP) AS, an Open Service Access Multimedia Domain (MMD) forum. (OSA) Service Capability Server (SCS) & OSA AS, Therefore, we refer to IMS as a cellular core and an AIN Interworking Server. The CSCF is the Call technology for supporting any cellular Session Control Function which has 3 flavors. network technology. Note also that seamless Serving-CSCF provides session control for endpoint handoff between cellular networks and/or devices, Interogating-CSCF is an entry point to IMS WLAN has not been specified as yet. Its from other networks, and Proxy-CSCF is an entry point to IMS for devices. The BGCF is the Breakout specification is anticipated as another SIP Gateway Control Function which selects networks to application serving an IMS serving call use for PSTN/PLMN interworking. The MGCF is the session control function (S-CSCF). Media Gateway Control Function which controls the MGW. The MRFC is the Multimedia Resource Function Controller which controls MRFP (media 6 3GPP SIP header extensions are used in IMS SIP server). And the PDF is the Policy Decision Function signaling. which authorizes QoS requests. Management Servers (CMS) and media inadequate. To understand what new 802.11 gateway resources for VoIP-based landline power management approach was selected for phone service. Dual-mode handset and VoIP traffic, it is first necessary to understand landline inter-working is expected to be the why legacy 802.11 power management was subject for Phase 2 of the PacketCable 2.0 found inadequate. cellular integration initiative. In 802.11, legacy power management In summary, of the three alternatives allows a station (dual-mode handset in our discussed, IMS appears to offer the most application) to receive data from an AP at promising solution to cable operators from a infrequent intervals and only when data is business, functional, cost, and performance available at the AP for delivery. AP data perspective. transfer may be scheduled to occur at Delivery Traffic Indication Message (DTIM) POWER MANAGEMENT, SECURITY, beacons, spaced say 300ms apart. Utilizing QOS, AND NAT TRAVERSAL this timing, a station need only wake its CONSIDERATIONS processor infrequently to listen to selected AP Aside from providing seamless roaming beacons and poll the AP for its data if the AP indicates its data is available (via Traffic between WLAN and cellular WAN and 8 enhanced services, seamless mobile Indication Map). However, VoIP traffic sent communications solutions must address between station and AP during a VoIP call several critical factors in order to provide a requires a short packet exchange intervals, say viable service. These include handset power 20ms. Consequently, if legacy 802.11 power management, security, QoS, and NAT management were used it would need to have transversal for operation from a WLAN. shortened beacon intervals of say 20ms to accommodate the VoIP traffic. In that case, Power Management processor sleep time would be very limited, particularly during idle call periods. Power management is required for a dual- mode handset in a WLAN to have active and The new 802.11 power management idle talk time that is comparable to that in approach is a method for allowing periodic typical cellular service. Without power frame exchange between a station and AP management active talk time may be reduced during a VoIP call while legacy power from hours to minutes and idle talk time may management is still utilized for data frame be reduced from tens of hours to single-digit exchange. A station activates the new power hours. management approach when it negotiates an admitted VoIP traffic stream (or any periodic Through the development of an enterprise frame transfer stream) with its AP. An AP seamless mobility solution, Motorola then buffers frames from the VoIP traffic identified several factors as necessary for stream destined to the station and, when a improving battery life in a dual-mode handset. station wakes up at a designated VoIP frame These include choosing a low-power 802.11 interval (presumably set to the VoIP chipset in the handset, running the handset packetization period), it sends a VoIP frame with one radio at a time (cellular WAN or with an implicit “poll” request to its AP. The WLAN), providing fast switching technology to select WLAN/WAN, and implementing a new 802.11 power management approach for 8 An AP responds to a poll with a buffered frame. The VoIP traffic. In the latter case, legacy 802.11 handset must poll for more buffered data if more exists. power management alone was found to be The AP will also send broadcast and multicast data right after DTIM beacons. AP responds to a poll request with buffered draft 3 for authentication and encryption. VoIP stream frame(s). Using WPA-PSK, mutual authentication, key verification and derivation between the This new power management approach has station/handset and AP is established using an been pushed into the 802.11e standard where 802.1x handshake. This is illustrated in the it is referred to as Unscheduled Automatic following figure. Power Save Delivery (U-APSD). U-APSD may be enabled when the U-APSD station Supplicant Authenticator issues a bi-directional Traffic Specification (STA) (AP) 9 (T-SPEC) to its AP at the start of a VoIP call . EAPoL-Key (Reply Required, Unicast, Anonce) The AP buffers frames matching the Access EAPoL-Key (Unicast, Snonce, MIC, STA SSN IE) Category (AC) for the T-SPEC according to the method described previously. The EAPoL-Key (Reply Required, Install PTK, Unicast, Anonce, MIC, AP SSN IE) following figure illustrates U-APSD from the EAPoL-Key (Unicast, Anonce, MIC) perspective of AP and station (subscriber unit).

A Figure 5. 802.11i WPA-PSK Example with C V AP K Supplicant and Authenticator

V+P SU 20ms After successful authentication, the Voice Frame Next Voice Frame AP – Access Point Temporal Key Integrity Protocol (TKIP) SU WLAN Power Consumption Profile SU – Subscriber unit V – Voice encryption algorithm is used to encrypt TX V+P – Voice + Poll RX Channel Sense Time session traffic. WPA2-PSK is another logical Doze 802.11 Post Backoff choice for residential applications. WPA2- Handset in Deep Sleep Mode during VoIP phone call PSK is an evolution of WPA-PSK, based on Figure 4. U-APSD Operation the final ratified 802.11i standard and includes use of the Advanced Encryption Security Standard (AES) encryption algorithm instead of TKIP. Security for a dual-mode handset operating from a WLAN is necessary to prevent user For enterprise applications, WPA or WPA2 traffic (VoIP and data) from being denied, using an operator selected Extensible stolen, or eaves-dropped. The level of Authentication Protocol (EAP) method is security offered should be commensurate with appropriate. EAP methods may include the level of threat and the impact of a mutual authentication schemes based on successful security attack. X.509 certificates (i.e., EAP-Transport Layer Security (TLS)), or other alternatives such as Security should be provided on the WLAN EAP-Tunnel TLS (TTLS), EAP-Subscriber between handset and AP and end-to-end Identity Module (SIM), etc. EAP message between handset and network infrastructure. exchanges are illustrated in the following For Wi-Fi access security, several standard figure. approaches are possible. For residential applications, wireless protected access with pre-shared key (WPA-PSK) is a logical choice. WPA-PSK is based on the 802.11i

9 802.11e draft 12 also provides a mechanism for a station to enable U-APSD in its reassociation request frame, without submitting a T-SPEC. Data Gateway (PDG). As part of the tunnel establishment, the PDG contacts a 3GPP STA AP AS 802.1X EAP Request AAA Server in the Home Public Land Mobile

802.1X EAP Response Identity Access Request (EAP Request) Network (HPLMN) for authorization of the 11 EAP Authentication Protocol Exchange mobile. IMS may require several IPsec

Accept/EAP-Success/Key Material tunnel attempts, starting with PDG associated 802.1X EAP Success with a handset’s Visiting PLMN and

IEEE 802.1X Port progressing to PDG associated with its Blocked f or STA HPLMN.12 Authentication between handset and AAA server uses either EAP-SIM or Figure 6. 802.11i Example with Supplicant, EAP-Authentication and Key Agreement Authenticator, and Authentication Server (AKA). For end-to-end security, handset Note that enterprise solutions may also utilize authentication and authorization is necessary separate VPN security for remote data access. to mitigate theft of service. In addition an QoS IPsec tunnel may be utilized to protect VoIP, data, and signaling traffic. UMA and 3GPP QoS for VoIP calls from a dual-mode IMS standards specify use of an IPsec tunnel handset when operating in a WLAN should be and EAP authentication. In the case of UMA, comparable to landline-like voice service to an IPsec tunnel is established between a differentiate service from competitive VoIP handset and UNC. The tunnel follows service providers. QoS should be provided authentication between handset and UNC over the WLAN between handset and AP, and secure gateway and AAA server. end-to-end between handset and network Authentication is performed using EAP-SIM infrastructure. within IKEv2. IKEv2 is used to authenticate a UNC to a handset and EAP SIM is used to The 802.11e standard provides QoS authenticate a handset to a network methods for a Wi-Fi network. It includes Authentication Authorization and Accounting specifications for prioritized and (AAA) server, as relayed by the UNC’s parameterized QoS in two basic access security gateway. UMA requires approaches: Enhanced Distributed Channel authentication and establishment of an IPsec Access (EDCA) service and Hybrid tunnel with at least two UNCs as part of Coordination Function Controlled Channel handset discovery and registration.10 Access (HCCA). EDCA is a prioritized CSMA/CA access mechanism offering four In the case of IMS, an IPsec tunnel is differentiated service Access Categories (AC). established between a handset and Packet EDCA also offers a level of parameterized

11 The service authorization decision is made by the 10 UMA clients are provisioned with or derive the 3GPP AAA Server based on subscription information FQDN for their provisioning UNC and UNC SGW. retrieved from the IMS HSS/HLR. After resolving these addresses, a UMA client sets up a 12 A handset will initially construct an FQDN based on secure tunnel with the provisioning UNC to discover its Visitor Public Land Mobile Network (VPLMN) and the FQDN or IP address of its default UNC and UNC use it to discover (via DNS) a set of visited-network Security Gateway (SGW). The UMA client then sets PDG IP addresses to attempt IPsec tunnel establishment up a secure tunnel to its default UNC and attempts to with. Should these addresses not be reachable or the register. The UMA client’s default UNC may deflect AAA server reject access to them, the handset will then the UMA client to an alternate serving UNC. In this construct a new FQDN based on its HPLMN and case the UNC client will establish a secure tunnel this attempt IPsec tunnel establishment with a home- UNC and register. network PDG. QoS for time sensitive traffic flows such as In the case of downstream end-to-end VoIP. A T-SPEC with AP admission control traffic a reverse process would apply. A is used to establish a traffic stream. HCCA is DSCP assignment at the other end of the VoIP a more complex approach involving a mix of call (terminating handset, MTA, or media contention and contention free periods (CFP). gateway) could be used to classify a VoIP It provides T-SPEC parameterized QoS for packet such that it receives proper DOCSIS time sensitive traffic using scheduled downstream service flow assignment at the transmission opportunities in the CFP. CMTS and 802.11e T-SPEC at a Wi-Fi AP. Downstream service flows would also have Several subsets of the 802.11e standard been established at the start of a VoIP call.14 were identified by the Wi-Fi Alliance in order to expedite interoperability testing for WLAN The 802.1d or 802.1p user priority should QoS. Wireless Multimedia Extension (WME) also be assigned for priority queuing at an features and more recently the Wireless Multi Ethernet switch. The following table provides Media (WMM) features were chosen based on a listing of typical assignments for 802.1d, the EDCA. Compliance to these tests is 802.11e AC, and RFC2474 DSCP. mandatory for many solutions, making EDCA commonly utilized. With EDCA, VoIP calls Table 1. Typical Relationships for QoS are typically assigned a T-SPEC with QoS Assignments parameters that limit delay and jitter and with 8021D 80211e RFC2474 DSCP Bits Recommended Traffic User Priority Access Category AC value set to VO, the highest level. Category 1, 2 AC_BK 001xxx or 010xxx Background For end-to-end QoS, IP differentiated services per RFC2474 (DiffServ) should be 0, 3 AC_BE 000xxx or 011xxx Best Effort 4, 5 AC_VI 100xxx or 101xxx Video considered. DiffServ sets packet priority via a 6, 7 AC_VO 110xxx or 111xxx Voice Differentiated Services Code Point (DSCP) field of an IP header. For upstream traffic Note that there are various techniques that from dual-mode handsets in a Wi-Fi network can be used for associating signaling and a DSCP assignment could be used to classify media to an 802.11e AC. In one case a a VoIP packet onto an 802.11e VoIP T-SPEC station/handset could have a default AC with AC_VO, established at the start of the method for signaling and media. In another call. It could also be used to classify a VoIP case, a station could use UPnP to discover a packet onto a DOCSIS Unsolicited Grant UPnP capable AP’s QoS capabilities and Service (UGS) service flow in the case of adjust its signaling and media ACs cable modem service, also established at the accordingly. start of the call. In this case, PacketCable Multi-Media (PCMM) is needed for NAT Traversal establishing a UGS service flow for the upstream VoIP call.13 Dual-mode handset traffic (voice or data) should be allowed to pass through any residential gateway for a seamless solution to be widely deployed. This includes gateways with network address and port translation (NAPT). 13 PCMM QoS is initiated by an Application Manager which, for a seamless mobility solution, would need to 14 be affiliated with an IP PBX or CMS in a call In the case of a DOCSIS downstream, the service forwarding solution, the UNC in a UMA solution, or flow would need a minimum reserved rate to be an S-CSCF in an IMS solution. assigned via PCMM.

Two issues concern setting up and passing undesirable delay into the communication VoIP traffic through a NAPTing gateway. path. In addition, both the STUN and TURN The first issue is that protocols that embed solutions require application layer support on local IP or port information in signaling the handset and the STUN/TURN server(s) in messages (i.e., SDP in SIP) will not be the service provider network must be on the reachable through a NAPTing gateway public Internet, making them susceptible to without a solution that provides the device threats such as denial of service attacks. behind the gateway (dual-mode handset) with information about its NATed IP address and The second issue with pasing VoIP traffic port or without a network server that provides through a NAPTing gateway is that if IPsec address translation from local to routable tunneling is utilized, NAPTing will not be addresses.15 Simple Traversal of UDP possible without the handset providing UDP through NAT (STUN) or Traversal Using Encapsulated IPsec/ESP Tunnel Mode. This Relay NAT (TURN) are two approaches that is because IPsec tunneling without UDP provide a client device behind the gateway encapsulation only presents an unencrypted IP with information about its NATed IP address header field, providing no UDP field for port and port. address translation. UDP Encapsulated IPsec/ESP Tunnel Mode provides an outer IP There are four types of NAT and UDP field for NAPTing as shown in the implementations: full cone, restricted cone, following figure. port restricted cone, and symmetric as defined Before Applying ESP/UDP in [9]. STUN is a solution for full cone, Original IP TCP Data restricted cone, or port restricted cone NAT Header Header implementations. However, in the case of a After Applying ESP/UDP New IP UDP ESP Original IP TCP ESP ESP Data symmetric NAT implementation, STUN is not Header Header Header Header Header Trailer Auth an effective solution because it will not work Encrypted Authenticated with connection oriented media using RTP. This is because a symmetric NAT will have different NAT mappings for each destination Figure 7. UDP Encapsulated IPSEC/ESP IP address and port. The symmetric NAT Tunnel Mode requires a VoIP client to send and receive When a NAPTing residential gateway RTP packets to and from the same IP address receives a UDP Encapsulated IPsec/ESP and port. For symmetric NAT tunneled packet (e.g. an RTP VoIP packet implementations, TURN is a possible from a dual-mode handset when in a WLAN), solution. TURN involves using an RTP relay it conducts the following operation: server between communicating VoIP clients. 1) replaces the Source IP Address in the outer Unfortunately, the TURN solution has IP packet header with the WAN IP address of some disadvantages. It is processing the AP intensive on the RTP relay server, raising 2) replaces the Source Port in the outer UDP issues of scalability, and it introduces packet header with the UDP port selected by NAPT 15 A gateway ALG may also be used but it is limited to 3) recomputes the IP packet Header scenarios in which signaling is not encrypted, a Checksum in the outer IP packet header16 scenario that can not be relied on. An ALG can not be utilized with IPsec since there is no SIP port number in the IP payload for the ALG to key off. In addition, the inner IP header and contents are all encrypted and 16 The Header Checksum provides a verification that cannot be parsed or changed by the ALG. the information used in processing an internet datagram UMA and IMS require use of UDP Home portal Encapsulated IPsec/ESP Tunnel Mode for Monitor doors IP Gateway manages Home Internet Receive in-home devices and network IPsec tunneled VoIP from a WLAN. notifications from communicates to portal and access to the home from anywhere on the OTHER SEAMLESS MOBILITY Internet CONCEPTS

Control and schedule View front door camera on TV As mentioned previously, seamless HVAC systems Control mobility in general terms is an approach that Connect to the car appliances allows users to roam between application Figure 8. Home Monitoring and Control domains and communication networks Seamless Mobility Scenario without being aware of the underlying mechanisms that enable them to do so. This In this example an IP gateway device in the may include providing seamless mobility for home may be used to provide an “always on” voice and data communication from a multi- portal between in-home devices and mode handset. It may also mean receiving communications access. Home monitoring access to a “family” number from any phone devices may have wired or wireless in the house or getting access to caller ID, call connections to the gateway. These devices logs, and or billing information from a TV. In may use non-IP protocols, in which case they the most general sense, seamless mobility is are proxied on to the home network through a more than seamless handover from a single protocol translator to make them look like device but having access to a common they have native IP. What is key to the application or user experience from different concepts in this and other seamless mobility devices at different locations. Furthermore, solutions is that a user’s seamless mobile seamless mobility need not only apply to experience is consistent (same “look and communications but may apply to other feel”) across the different access platforms, applications and experiences such as home even with varying levels of features being monitoring and control, purchased content accessible at different platforms. management, and personal content In the case of purchased content management. management, seamless mobility implies a In the case of home monitoring and means of allowing users to purchase music control, seamless mobility is a means for and video content from home and cell phone, allowing users access to their home to store and manage content, to play content monitoring and control system while on their from a home stereo, TV, and cell phone, and home network, via a Web portal from to transfer content to CD, portable audio and anywhere on the Internet, from their cell video player, and automobile. The following phone, from their home TV/STB, and even figure illustrates these concepts. from an automobile assistance service. The following figure illustrates these concepts.

has been transmitted correctly. The data may contain errors. If the header checksum fails, the internet datagram is discarded at once by the entity which detects the error. Buy/rent and stream Order prints View on music online digital frame Synchronize to Rip/burn CDs Web automobile based

Take picture Internet Store on IP Home IP Gateway stores Home Gateway / Internet View, edit and Settop network content and hosts network print from PC Store on PC family Blog

Listen via media Synchronize content adapter or connected from Jukebox to Buy music, pull stereo Portable Audio from home storage Devices Friends and family can view on PC or cell phone View directly Push to on TV from storage via phone Browse and listen STB from connected TV or HMA Settop Synchronize View from IP Gateway on family room TV content from DVR to Media Monster Figure 10. Personal Content Management Figure 9. Purchased Content Management Seamless Mobility Scenario Seamless Mobility Scenario In this example an IP gateway device in the In this example an IP gateway or Digital home may provide “always on” content and Video Recorder (DVR) Set Top Box (STB) storage management as well as a family Blog. device in the home may provide “always on” Accessory devices may be utilized for content and storage management. A viewing photos. A partnership may also be partnership may be established with online established with online printing services for content providers for seamless content seamless print ordering. purchases. Networked playback devices may be standalone or integrated. Portable audio SUMMARY AND CONCLUSIONS and video players should be home network docked. Access from a gateway to Seamless mobile communications is an automobile will need to occur via a wireless approach that allows users to roam between home networking connection. Digital rights application domains and communication management (DRM) licensing requirements networks without being aware of the may limit which devices and which network underlying technology or mechanisms that types that purchased content may have access facilitate transparent mobility. It promises to to. provide value to residential and enterprise users, operators, and vendors. In the case of personal content management, seamless mobility implies a Three different approaches to achieving means of allowing users to store and manage seamless mobile communications were family digital photos or video from camera, presented: Call Forwarding, UMA, and IMS. camcorder, and cell phone, to display content An IMS solution appears to be the superior on TV, digital frames, or cell phone, and to choice for Cable operators. order prints for digital photos over the Any seamless mobile communication internet. The following figure illustrates these solution must also account for power concepts. management, QoS, security and NAT traversal to be a complete service offering. For a dual-mode handset solution to provide acceptable battery life, additional power management techniques based on U-APSD are essential for a dual-mode handset service offering to be competitive with user expectations of cellular standby/talk times. In addition, proper handling of QoS on the WLAN and end-to-end is important in order to provide high quality audio and differentiate 4. IEEE Std 802.11TM-1999 (R2003) and its service offerings between competitive VoIP amendments. service providers. Security of the WLAN and end-to-end is should be provided and WPA2- 5. IEEE Part 11: Wireless LAN Medium PSK and WPA2 are good solutions for Access Control (MAC) and Physical residential and enterprise WLAN applications Layer (PHY) specifications, Amendment respectively. NAT traversal is also an 6: Medium Access Control (MAC) important consideration due to the variety of Security Enhancements, Std 802.11i-2004. NAT implementations and detection 6. IEEE Part 11: Wireless LAN Medium capabilities of network equipment deployed in Access Control (MAC) and Physical residential and enterprise environments. Layer (PHY) specifications, Amendment : Finally, seamless mobility need not only Medium Access Control (MAC) apply to communications but may apply to Enhancements for Quality of Service other applications and experiences such as (QoS) P802.11e/D12.0, November 2004. home monitoring and control, purchased 7. IETF RFC 2474, Definition of the content management, and personal content Differentiated Services Field (DS Field). management. 8. IETF RFC 3948, UDP Encapsulation of REFERENCES IPsec ESP Packets. 1. UMA Architecture (Stage 2) R1.0.2 9. IETF RFC 3489, Simple Traversal of User (2004-11-03), Technical Specification. Datagram Protocol (UDP) Through 2. 3GPP Group Services and System Network Address Translators (NATs). Aspects; 3GPP system to Wireless Local 10. IETF Internet Draft, Traversal Using Area Network (WLAN) interworking; Relay NAT (TURN). Work in progress. System description TS 23.234 V6.2.0 (2004-09).

3. IEEE Wireless LAN Edition, Standards Information Network IEEE Press compilation based on SECURING DIGITAL CONTENT – STRENGTHS AND WEAKNESSES OF SOFTWARE AND HARDWARE IMPLEMENTATIONS

Robin Wilson Nagravision

Abstract HARDWARE VS. SOFTWARE SECURITY

Conditional Access (CA) and Digital “Overview” Rights Management (DRM) are implemented in a number of ways in software (SW) and The security system relies on a hardware (HW). Often these schemes are computation, or algorithm, to decode the described as either “HW” or “SW” based protected content. security or rights management systems. Since SW requires HW on which to execute, Most digital CA systems employ a unique and HW has necessarily SW running on it, key that enables a successful computation. the terminology is often thoroughly confusing The locations on the Set-Top Box (STB) for and misleading. Since we are not dealing the decoding program and key are a the with locks and keys or hypothetical systems, subject o the hardware and software security both hardware and software elements must designations. be present, and work together smoothly, in any sophisticated content security system. “Software Only” Security Further confounding the confusion is the frequent use (and abuse) of terms like This and other terms like “Hardware- “replaceable”, “renewable”, “obfuscated” less”, “Downloadable” and even and “tamperproof”. In this session, these “Renewable” are used to describe security terms will be explained in the context of systems where the security solution supplied content security. by a CA or DRM company does not include hardware. Here the product of the company may be limited to only software but the TERMINOLOGY AND TECHNOBABLE inference is sometimes wrongly made that no additional resources are required or costs are Before embarking on comparing various incurred. security implementations and trade-offs, we will first define the terminology used. This Software Needs Hardware Too step will help to overcome the ambiguous definitions and terminologies recently Obvious, but software still requires applied under the guise of marketing new hardware for execution. Conventionally this security concepts. hardware is referred to as the CPU. In a security system, hardware security is a significant concern.

“Hardware” Security Hardware can be secure or not. Likewise, security software can either be written As we are not discussing security using without concern for potential reverse tumbler locks or brass keys, it should be of engineering or cleverly concealed attacks no surprise that the hardware referred to here within the hardware/system. Efficient is electronic in nature and it runs software! implementations are often a hybrid mix of In many instances, the decoding algorithm is several SW and HW security techniques. programmed into the “hardware” device. SOFTWARE IN A SECURITY SYSTEM Hardware is free? Bug Free (and we really mean free) A further misrepresentation is sometimes made that a so called “SW only” systems A Bug in any software jeopardizes your have zero hardware cost. While it may be hard-earned customer relationship and true that there is no hardware cost in the STB potentially your corporate reputation. While to be passed on by the CA / DRM company, a bug in a security system can be catastrophic someone, usually the operator, pays. and possibly threaten your business. Even a single bug can quickly give hackers many The assertion that CPU cycles are more clues as to how a system operates incrementally free is also misleading. Any compared with months or even years of CPU on any STB is almost always maxed analyzing a bug free system. This is further out. Indeed prior to launch of a new STB or compounded in a security system where a service it is unusual to find less than 110% of bug can be classified as any unplanned CPU resources are already assigned. There is operation regardless of the stimulus. For this a significant opportunity cost and a real cost reason the security of SW has much more to freeing up cycles for security applications. stringent requirements including the need to Even if the security application takes only do nothing when stimulated by any of an 20% of an existing CPU that can become infinite array of malicious or accidental several dollars in additional cost. stimuli. Try requiring that of you web browser! As the computational power of the STB increases and the sophistication of the CA Beyond the impact on the CA system, any system is enhanced to meet ever-growing unplanned and unexpected operations will threats, there is the risk that legacy STB’s impact the viewer. This is likely to have a will be unable to support the CA system negative effect on the viewing experience. necessary to protect content. Small is Good To Summarize: For the reason above and for speed of • Security software runs on hardware operation, bug free security software is best • Security hardware runs software written in very small kernels by very small • Hardware is always required teams with exhaustive regression testing. • Hardware can be secure or insecure • You pay for security hardware even Throwing tens or hundreds of staff at the in a “SW” only system problem will not help. Neither will bloating the code-base with huge code footprint. It is in this area that a few very experienced a temporary or permanent halt to the security programmers can easily out-perform the huge processes. corporations who’s bug ethics are driven by “good enough to ship” or “let the end HARDWARE IN A SECURITY SYSTEM customers find the bugs”. It is easy to see that a large, complex, system presents many Having bug free security SW is useless if more opportunities for error and bugs. the operating states, registers etc, can easily be monitored. While it is well beyond the Isolation scope of this paper to discuss the security philosophies relating to hiding and keeping It is counter-productive to implement tight secrets, having a transparent hardware highly-secure bug-free security kernel only to platform like a generic CPU, where the have other applications sharing CPU or operation and architecture are well memory resources (planned or unplanned). understood, fatally undermines almost any That strategy will totally undermine security. security scheme. Security will be at the mercy of the application suite du-jour. In-turn the QA HW Obfuscation issues of requiring that all the applications sharing the same code and memory space are Just as with SW, in the context of HW the totally bug free make this flawed structure of silicon or functional blocks are implementation unworkable. Security deliberately made non-obvious or non- software needs to be isolated in a protected intuitive to either human or machine. The environment isolated from a hostile memory term camouflage may also be used. Here the and bugs. Denial of visibility and silicon structure is laid out in an apparently accessibility of the CA process is essential to identical manner for many of the building protecting its secrets. blocks and the critical differences are hidden deep inside an obscure silicon structure.

SW Obfuscation HW Tamper-proofing

This means that the operation or structure Here the goal is to detect any abnormal of the software is deliberately made non- probing of the silicon or functional block. obvious or non-intuitive to either human or Numerous techniques are employed from machine. Although this term has recently detection layers to radiation detectors to gained some use in referring to automated produce electrical anomalies. transformations applied to pure software products like games, the technique has been One common measure employs fuseable in use for more than a decade in CA or DRM links that can be burned away or destroyed systems. after the CA programming is loaded into memory. This makes reading that code and SW Tamper-proofing analyzing the memory structure far more difficult. Here the goal is to detect any abnormal operation in the SW due do any unwelcome As with SW tamper proofing, when an external stimulus. When detected, the attempted intrusion is detected, the tampering tampering will almost always result in either will almost always result in either a temporary or permanent halt to the security Countermeasures processes. The countermeasures feature within the RENEWABILITY – WHAT DOES IT security system allows for secure and REALLY MEAN? validated updates to the current executing version of the security solution. It also Downloadable assists the service provider in detecting and disabling compromised platforms. The downloadable feature within the security system authenticates or identifies a PERFORMANCE CONSIDERATIONS network element i.e. STB, securely communicates a downloadable solution, and Latency launches the solution into a secured environment. Providing an easy-to-use viewing experience is critical to keeping customers Replaceable happy and giving them no reason to look at other methods of content delivery into the This has two possible meanings: home. Although not often thought as a factor in subscriber retention or churn it is 1). Electronic - The replaceable feature important to ensure that viewers are never within the security system revokes the annoyed by additional channel selection current security solution, restores the secure delays. In the new competitive video environment, and securely enables the environment, channel change delays will downloadable feature for the replacement become a differentiating factor for service security solution. providers.

2). Physical – Here a physical device may be Latency in a CA system can be broadly replaced. Replacement is based on proper categorized as two issues: authentication, binding/paring and secure provisioning. Note: this does not always 1). The first is the time taken between a infer that the removal of the previous device. subscriber’s request to view a channel or Physically replaceable hardware cuts both view a stored file and the proper permission ways. It permits total replacement of a communicated to the security process. This compromised CA system but it also permits could be summarized as “checking the cloning of apparently legitimate hardware viewing rights”. In any modestly sized solutions to receive unauthorized service. broadcast system, there is not enough bandwidth available to broadcast all the rights for each viewer frequently enough to Renewable avoid annoying delays of seconds if not minutes. The CA system must provide The renewable feature within the security specialized configurations, storage of system suspends the current executing permissions, and communication capabilities version of the security solution, maintains the for the timely delivery of each customer’s secure environment, and securely enables the permissions. downloadable feature the a new or upgraded version of the security solution.

A system without these specialized Rights Management Matrix configurations and communication capabilities relying on a two-way out of band This is the operational heart (as opposed (OOB) network such as one with a dedicated to the security heart) of a CA or DRM DOCISIS / DSG return path, encounters system. It is the complex alignment and processing and round trip delays such that the communication of the various rights as system cannot scale into tens of thousands or mapped onto the marketing driven needs more subscribers and guarantee to operate including packaging, floating previews, Push with the required quality of service. In VOD rights etc. Much of this functionality addition such an architecture would have to must execute on millisecond boundaries, bear the cost of the BW and support costs for uniquely control individual subscribers in the burdensome continuous OOB two way multimillion subscriber systems yet require traffic. The same system limitation applies to very low communication bandwidth and a pure IP network. Pure IP networks are minimum latency. often limited in the bandwidth dedicate to an individual STB limiting the number of It is this complex functionality, tightly communications carriers available to linked to cryptography that is often communicate with the STB overlooked and misunderstood, particularly in the area of new and emerging IP network In order to achieve satisfactory low where it is naively thought the routers, latencies it therefore becomes necessary to CMTS’, DSLAM’s or other edge devices can selectively stream and cache the subscriber execute this function. rights. CA HARDWARE IMPLEMENTATION 2). The second relates to the real time EXAMPLES cryptographic time base(s) used. When a subscriber requests to view a new stream. It Secure Microprocessor is considered desirable that the channel change delay from security system is well A secure microprocessor is a specialized under one second it is considered desirable to CPU with numerous enhancements. The be under 100mS. A figure of 200 mS is principal enhancement is a hardened generally accepted as the delay threshold that hardware and software environment to causes irritation. safeguard against security attacks. Hardening features can include intrusion Latency in a CA system is a complex detection, camouflage cell structures, subject with operational considerations encrypted communication, and including the likelihood of network outages, cryptographically secured memory with installation immediacies, warehouse support randomized page, address, and value etc, but it is important that the base-line construction. Today’s modern secure micros operation of any large deployment is non- can include many security specific immediate making available resources so that enhancements including true random number installations and customer support can be generators, public key generation, and given high priority. several security algorithm accelerators. A secure micro is an optimized platform for implementing security solutions. A secure microprocessor is typically used in a BGA, SIM, Smartcard, MCM, USB Key and There are two SC interface types: Contact potentially CableCARD (see below). and Contactless. The contact SC uses ISO- 7816 standards pin connections to BGA communicate via direct electrical contact. A contactless SC does not have contact pins but A BGA (Ball Grid Array) is a popular communicates via radio frequency (RF) type of physical IC package that provides using an embedded wire loop. high I/O density, small footprint, and physical security features. Historically, IC TV pass I/O has been through pins on the perimeter of the IC package, the BGA provided I/O A TvPass card is a proprietary security through an array of solder ball connections card from Motorola / General Instrument on the underneath side of the IC package. system providing a renewable security Since the I/O connections are physically solution. sandwiched between the IC and PC board, this maximizes the I/O connections, Proprietary solutions face cost challenges minimizes the footprint, and physically presented by non-standard manufacturing restricts access. Secure microprocessors are requirements and hardware with limited often implemented in a BGA package volume.

SIM CableCARD

A SIM (Subscriber Identity Module) is the The CableCARD™ is a more substantial security module predominantly used in GSM device, similar to a PCMCIA Type II- card mobile phones. For content security designed for laptops, that slides into a slot on applications, a SIM can be considered to be many newer high-end or high-definition identical to a smartcard in functionality, the television receivers and DVR’s. The main difference being a smaller physical size CableCARD™ eliminates the need for a and different insertion requirements cable STB (Set-Top Box) at least for the compared to a smartcard. Because a SIM is decoding function. The CableCARD™ visibly smaller than a smartcard is sometime contains a secure microprocessor, one or assumed that it must be less expensive. In two-way data transceivers, and specialized fact the production process is nearly circuitry to process security information and identical for both, and in some instances decrypt the digital content. It is, essentially, additional steps are required to punch the the entire CA system, hardware and software, SIM form factor out of a larger smartcard on a removable device. carrier. The current CableCARD being deployed Smartcard has only one-way functionality. The standard for a two-way multi-stream CableCARD is in A Smart Card (SC) is a credit card size development. security card containing a secure microprocessor. A SC has a wide variety of MCM applications ranging from phone cards, digital identification devices, and standards- A MCM (Multi-Chip Module) is structure based satellite / cable renewable security. consisting of two or more integrated circuits interconnected within the same IC package. security solution spans the following An MCM allows for high-density features: secure digital ID passwords and implementations with security provided by digital credentials securely stored on the key semiconductor-level integration. A typical and automatically presented to applications configuration might be an audio-video as required, authentication for third party decoder and descrambler packaged with a verification including multiple factor secure microprocessor. MCM technology authentication utilizes a variety of attempts to keep critical interfaces within the authentication methods including biometrics, chip structure away from attempts to one-time-passwords, digital certificates, and compromise the security system traditional PINs and passwords. Again a secure microprocessor can be used. SOC CONCLUSION A SOC (System On Chip) is a general class of solutions allowing for high Implementing a complete security system integration of many of the major subsystems in a STB is always a complex blend of many within a digital STB (Set-Top Box) and hardware and software techniques. Securing likely including the security solution and the the STB, while critical is only one aspect of secure microprocessor. Today no realizable an overall security solution perhaps just the implementations are in current use. most visible “tip of the iceberg”. Developing General Purpose CPU entirely bug free SW and systems is critical for security. Throwing huge resources or A general purpose CPU mentioned groups at the problem is a recipe for because of the compelling quantity of interminable security compromises. consumer PCs. Most if not all Likewise utopian “SW only” systems that communications to and from the CPU are in claim near free functionality and perfect the open and can be easily accessed. replaceability are the technological Monitoring, debug tools and expertise are equivalent of diet pills. widely available. Not an option to secure in a security system of high value content. Beyond the scope of this paper, there are many additional security issues in the supply, USB Key fulfillment, and support chains that have equally challenging solutions. Remember a The USB security key is used to combine security system is only as good as its weakest a digital identity and security functions into link and cryptography is only a small but an integrated security device. The USB important part of a security solution.

SILICON INTEGRATION FOR NEXT GENERATION VOICE OVER CABLE CUSTOMER PREMISE EQUIPMENT

Tony Andruzzi, Bill Wallace ARRIS Patrick Hibbs Product Marketing Manager, T.I.

Abstract multiple service providers (MSOs). VoCable deployments, starting as early as 1995, utilize The voice over cable (VoCable) market is traditional constant bit rate (CBR) and are the defined by both constant bit rate (CBR) and dominant method of providing cable voice over internet protocol (VoIP) telephony service today. More recently, technologies. Operators that have previous deployments have started to utilize voice CBR VoCable experience have provided over internet protocol (VoIP) technology. significant input on the features required for VoCable deployments have reached nearly next-generation VoIP. VoIP has gone twelve million subscribers and less than 10% through an initial deployment ramp in 2004 of the total subscribers are served by VoIP with many of the world’s service providers systems. In both CBR and VoIP systems, a either rolling out a commercial service or key element to successful VoCable conducting extensive, large scale field trials. deployment is the client device installed at Feedback from this process along with the service subscriber’s site. This client industry input on the future direction of the device, referred to as customer premise market has provided a valuable set of equipment (CPE) has evolved over a decade features and optimizations to be included in of deployments. The CPE evolution has been next generation products. This paper will driven by customer feedback and market explore the features necessary in a next trends. Valuable experience has been gained generation EMTA and the integrated, cost during field trials, initial service effective silicon necessary to support these introductions and full scale deployments of features. The paper will explore the VoCable CPE. This experience and feedback integration options that can reduce the has provided CPE vendors valuable component count and cost of the design. The information to aid in the analysis and paper will focus on factors and tradeoffs development of cost effective silicon associated with silicon integration that can technologies. optimize power supply, battery charger and telephony interface costs. The paper will CPE Evolution also explore the integration level necessary to support add on of peripheral components When first introduced, VoCable CPE such as wireless LAN technology and devices used in CBR systems were expensive convergence with mobile voice technology and bulky because they were developed with a Carrier Grade telco mindset, and typically INTRODUCTION deployed outdoors. As VoCable deployments grew and field experience increased, CPE For the past ten years Voice over cable vendors developed newer generations of CPE (VoCable) has been offered by several based on MSO feedback. Key factors that needed to be addressed included cost, system communications processor and DOCSIS availability, power consumption, and MAC and PHY functions integrated into a terminal equipment (ie. black phone set) single ASIC. CBR systems have also compatibility. enjoyed cost reduction due to silicon integration, however the pace of additional VoIP technology offers an opportunity for cost reduction has slowed due to the R&D the MSOs to continue to lower the cost of investment shift to VoIP. Further opportunity VoCable deployments as well as provide an still exists for additional silicon integration open platform to add new future services. and lower costs for VoIP CPE. The earliest VoIP CPE were essentially DOCSIS Cablemodems with discrete voice CPE Features interfaces, provided for early lab evaluations and field trials. These specialized Experience gained from VoCable Cablemodems are now often referred to as an deployments in both CBR and VoIP, along embedded multimedia terminal adapter with industry input on the future direction of (EMTA). Many MSOs have completed lab VoCable has helped to identify a set of key evaluations and field trials and have begun features to be considered for next generation VoIP cable telephony service introductions CPE. This paper will focus on a few critical and plan large scale deployments of EMTAs. features and the cost effective silicon integrations necessary to support these VoCable CPE designed for CBR and features. The features to be explored are: VoIP systems consist of similar functional blocks. Both types of CPE utilize a cable - low product cost MAC and PHY, a communications - telephony line count processor, memory, RF tuner, telephony line - ILEC replacement interface and power supply. EMTAs for - installation environment VoIP systems include a voice digital signal - mobile voice processor (DSP) that is not required in CBR system CPE. LOW PRODUCT COST

CPE for CBR systems typically use As with most Consumer Electronics proprietary MAC and PHY silicon while products, cost is one of the most important EMTAs utilize DOCSIS cable modem MAC factors in designing a VoCable CPE. Cost is and PHY silicon. The communications also one of the most difficult challenges processor, memory, RF tuner, and telephony facing the CPE design engineering team. line interface provide similar functionality for Economics ultimately drives every design both types of CPE. The power supply decision so cost tradeoffs must be considered circuitry is dependent on the requirements of when developing the next generation of the remaining functions. VoCable CPE. This is especially important when considering cost reduction through Silicon integration has been applied to the silicon integration of CPE functions. various functional blocks of VoCable CPE for cost reduction. EMTA’s have also Silicon integration is not always the most benefited from the lower costs associated effective way to optimize the CPE cost. Care with large volume deployments of data-only must be taken to avoid adding silicon for Cablemodem as well. In particular, silicon features that may not be utilized in volume vendors already offer products with the applications.

Integrated circuit (IC) packaging must consistent in their requirement for the also be considered. IC power density and number of telephony lines for residential thermal considerations limit what can service. Early CBR system deployments affordably be integrated in a single die and a typically required two telephony ports on single package. CPE, although there are cases when up to four telephony ports are required. Four As the number of integrated features grow telephony lines provide a significant potential so does the complexity. Size and complexity service differentiator for the MSO over the determine design, test, and manufacturing of Incumbent Local Exchange Carrier (ILEC). ICs, all of which impact cost. The various MSOs deploying VoCable with CBR systems found that the actual number of Increased silicon density magnifies telephony ports being deployed per resident integrated circuit susceptibility to process ranged between 1 and 4 with the average parameter variations and noise disturbances. being around 1.2 lines. As the geometry of the IC decreases the designs may require more fault-tolerance As expected, CPE requirements for VoIP which could add cost. systems have mirrored those of CBR systems. Two telephony lines per EMTA In the near future, complexity will continue to be the volume preference. increase due to a larger range of applications Silicon vendors have recognized this and are and higher level of design abstractions in already taking steps to optimize their silicon integrations. Robust solutions will be products for two telephony lines through required for the integration of mixed integration. However, interest for CPE with technologies. Implications of increased a single telephony port is also being integration will result in increasingly difficult discussed as an option to provide a lower cost cost decisions. alternative to subscribers. Currently, single line CPE cost will be burdened by the silicon utilized to implement the prevalent two line TELEPHONY LINE COUNT CPE. An important feature of any VoCable CPE is the number of telephony ports that are Multiple Dwelling Unit Line Count supported. The number of telephony ports required is primarily driven by the type of Multiple dwelling unit (MDU) telephony service offering. Three types of services are service requires a higher line count than considered: residential services. MSOs prefer to use a multi-line CPE device for economic reasons - residential service including ease of installation, footprint and - multiple dwelling service equipment recovery. Modularity is also an - enterprise service important feature to the MSOs so that additional lines can be added easily as Residential Line Count subscriber penetration increases.

Residential telephony service is by far the Larger line sizes and modularity create a largest portion of the total deployment of particular challenge for silicon integration. VoCable CPE today. MSOs have been fairly Cost per line is a critical factor in multi-line CPE so silicon integration should be a - interfacing with a wide range of consideration, but volumes may not justify standard terminal equipment the development cost. Also, it is important - interfacing with existing in-house not to burden the cost of the residential CPE. wiring - broad range of technical CPE silicon optimized to provide two specifications telephony lines should contain “hooks” to - reliable battery backup with real-time provide the capability to easily expand the status number of telephony lines supported. Telephony line expansion is possible by Terminal Equipment Compatibility including the capability to support cascaded DSP ASICs. CPE silicon integration could Since the MSOs strategy for cable also provide the ability to increase processor telephony includes ILEC replacement it is clock speeds to support multi-line important that existing terminal equipment applications. Continued telephony line operate the same when switched to VoCable interface integration is also under CPE. This is a critical contributor to the investigation. subscriber’s experience. Much has been learned regarding terminal equipment Enterprise Line Count compatibility from the extensive deployments of CPE in CBR systems. Early Enterprise telephony service is targeted at CBR deployments were affected by a few small to medium size businesses with line terminal equipment incompatibilities, such counts between residential and MDU as: services. Telephony service can be provided to many small business subscribers with four - handset volume and side-tone line CPE. Modular multi-line CPE is ideal - fax machine issues for businesses requiring more than four lines - caller ID issues and also provides a means for expansion. A - analog modem issues key distinction of enterprise service is a tendency toward higher telephony traffic Early CPE telephony line interfaces for which puts an added burden on the selection CBR systems were designed to meet the of features for silicon integration. Additional Bellcore TR-909 specification. TR-909 DSP resources are typically required to specifies different values for some of the support frequent use of feature such as 3Way voice and signaling parameters than those calling, voice compression, and T.38 Fax typically seen on traditional ILEC lines. TR- Relay. 909 is a specification targeting short subscriber loops like those expected in cable ILEC REPLACEMENT telephony applications. Unfortunately, some terminal equipment is impacted by these Most MSOs that have deployed VoCable parametric differences and the equipment CPE have offered facilities-based telephony either worked unsatisfactorily or not at all. In services in order to compete directly with the CBR systems these issues have, for the most ILEC. A key part of this strategy is to deploy part, been resolved. VoCable CPE that provides a “primary line” telephony service. Primary line CPE It is not surprising that some of these requirements include: same issues are appearing in early deployments of VoIP cable telephony, and in lines and call features supported dictate the most cases for completely different reasons. voice codec and compression bandwidth This is primarily due to the fact that the CPE requirements. If three-way calling is a for VoIP systems is required to perform required call feature then the voice codec many of the audio functions that were processing requirements are doubled. Low handled by the public switched telephone bit rate (LBR) compression also adds to this network (PSTN) in CBR systems. These processing requirement. The number of audio functions include: simultaneous telephony lines requiring LBR compression must also be considered. - call progress signaling - tone generation The expected EMTA volumes support the - tone detection economic integration of the DSP with the - voice activity detection cable modem ASIC, which consists of a - echo cancellation communications processor and a DOCSIS - voice compression MAC and PFY. The size of DSP and - fax relay supporting memory depends on the audio function optimization for the number of lines Since VoIP is base on packet switched and call features requirements. DSP technology, a DSP is required in EMTA requirements are also impacted by the designs to perform these audio functions. amount of margin in performance desired. The DSP used in the MTA must be carefully Field upgrades for new or modified call selected to insure that the required audio features are possible if the DSP is sized features can be supported on all EMTA appropriately. telephony lines simultaneously. This is particularly true for two and four line The telephone line interface is also residential EMTAs. In multi-line instrumental in providing terminal equipment applications it may be possible to allow for a compatibility. The line interface circuits traffic model that requires simultaneous provide the BORSCHT functions: support for a percentage of the total number of available lines. This is more likely the - Battery feed case for multiple dwelling EMTAs but less - Over-voltage true for EMTAs expected to support - Ringing enterprise services. - Supervision - Codec Optimizing a DSP for integration with - Hybrid other EMTA functional blocks becomes a - Test challenge when the audio functions are considered. VoIP CPE deployment forecasts The BORSCHT functions are indicate that an optimal DSP would support implemented primarily with a voice signal the full audio feature set for two telephony processor and a subscriber line interface lines. circuit (SLIC). The line interface must be software configurable to adequately address Particular audio functions must be terminal equipment compatibility. The voice considered when determining DSP signaling processor provides channel requirements for silicon integration. The filtering, input impedance synthesis, number of simultaneous active telephony transhybrid balancing, gain adjustment, and voice path diagnostics. The SLIC provides service. Since the CPE power consumption the high voltage interface for battery feed and directly affects battery backup time, a highly ringing, as well as the two-to-four wire efficient, low power design is a must. The conversion and loop test. subsystems, including the cable modem IC, battery charger, telephony interface and RF Various options exist for telephony line tuner, must continue to provide high interface silicon integration. The voice signal performance while consuming as little power processor is often implemented with a DSP as possible. so that a suitable integration with the DSP is possible, thereby providing the audio features Many of the power system components described earlier. The difficulty is can be readily integrated into the cable determining if this is a cost effective modem IC for improved cost. To make this integration. Expanding the capabilities of a successful, attention must be directed to low DSP to provide both the audio features and noise designs, versatile circuit blocks, and the voice signaling may not provide a cost high power efficiency. advantage over discrete components. Integration of DSP functions can be a cost Linear regulation is perhaps the easiest advantage if the DSP is properly optimized function to integrate and is available in and volumes are high enough. Concerns with today’s cable modem ASIC’s. These integrating the DSP functions are possible regulators should operate with low dropout feature reduction due to memory size voltage requirements to maximize efficiency constraints and the number of telephony lines and provide excellent ripple and noise supported. rejection across a wide range of frequencies. The output voltage of these regulators should Alternately, the voice signaling processor be available outside of the cable modem could be integrated with the SLIC. This ASIC for powering external circuits. The option has the disadvantage of mixing low available current from these regulators should and high voltage silicon technologies but is allow powering of typical CPE functions. ideal for multi-line telephony applications. Thermal limitations should also be addressed Service Availability within the ASIC.

An important requirement of primary line The efficiency demands of certain voltage service is to provide telephony service during rails will preclude the use of a linear utility power outages. In CBR systems this regulator. For these requirements, the was implemented with three distinct CPE integration of a generic Pulse Width powering schemes: Modulator (PWM) function will allow the power systems designer to design a highly - network power efficient (>90%) switching converter using - local external uninterruptible power external power components. This PWM supplies (UPS) function must be versatile to allow - internal battery backup implementation flexibility for the system designer. Within the cable modem IC, care For today’s CPE, the power system design must be taken to minimize noise effects from and the use of low power, high performance the PWM and associated external power subsystems has the greatest effect on components on the cable modem functions. achieving the powering goals of primary line The battery charger and monitoring requirements due to compensation circuits circuits are excellent candidates for silicon and expanded test times. There is also the integration. Today’s chargers are possibility that certain integrations are not implemented with simple 8-bit cost effective at extended temperatures. microprocessors, A/D converters, Op amps and similar circuit functions. These items are MOBILE VOICE inherently simple to integrate into the silicon and are low power by design. The Several additional features related to implementation within the silicon must mobile voice services are worth examining maintain flexibility to allow charging of related to EMTA silicon integration. These multiple battery chemistries, voltages, features include implementing cellular currents and algorithms. codecs in the audio feature DSP and adding wireless capabilities to the EMTA. Finally, silicon integration should take advantage of power management features. Each time a voice packet is translated The integrated components can be controlled from one codec format to a second codec to provide the lowest power consumption format, voice quality is degraded. This codec during various states of operation. During format translation is referred to as idle modes, selected IC functions could be transcoding. A call between a cell phone and shut down providing maximum battery back an EMTA goes through at least two up efficiency. Processor clock speed should transcodings. For example, the call would be also be controllable to allow optimal power translated from GSM to G729 and then from consumption. The processor clock could G7.29 to GSM. As the number of transitions then be slowed during idle modes to save increases the voice quality decreases. power. Instead of compressing voice traffic using INSTALLATION ENVIRONMENT the standard set of cable-centric codecs that are transcoded at network gateways, an VoCable CPE deployed in CBR cable EMTA could use a cellular based codec. telephony systems are principally installed This would eliminate a transcoder operation outdoors, either on the side of a building, on that contributes to lower voice quality. a pole, or occasionally strand-mounted. MSOs preferred this outdoor method to Codecs are audio functions so they are simplify installation and maintenance. implemented within the DSP. The addition Outdoor CPE are required if network power of cellular based or any other native is the preferred powering scheme. equipment codecs impact the DSP requirements and must be considered when VoCable deployments have shifted planning silicon integration. towards indoor EMTA installation to take advantage of lower product, but outdoor At some point market economics will EMTAs are being considered for some support the integration of silicon for wireless residential and multi-line deployments. ASIC local area network (WLAN) applications. operating temperature becomes an important Broadband-enabled phones will support high- aspect of silicon integration if outdoor quality audio, live video streaming, full- deployments are supported. Silicon cost is speed web access, automatic and impacted by extended temperature instantaneous synchronization with address books and email services, and real-time functions into single ASICs cost effectively. internet gaming. The key to a successful cost reduction is to carefully select the features, scaling An EMTA’s ability to support voice-over- capabilities, upgrades and future expansions. WLAN also enables convergence of the cable Timing is also critical in cost effective silicon and cellular networks. Subscribers will use integration. It is important to apply the the traditional cellular network when they are proper level of integration for optimal mobile, and access the VoIP network when performance for a given market. they are in the office or at home, utilizing the same dual-mode cellular/Wi-Fi handset. REFERENCES

SUMMARY 1. M. Paxton, “Cable Telephony Service: VoIP Finally Shows Up”, In-StatMDR Much has been learned during the past Report No. IN0401244MB, December, decade of VoCable deployments, and there is 2004. still more to learn with the advent of VoIP 2. W. Maly, “Cost of Silicon Viewed from cable telephony. What has been learned can VLSI Design Perspective, Electrical and readily be applied to designs for the next Computer Engineering, Carnegie Mellon generation VoCable CPE. The next University. generation EMTA designs must take 3 . W. Maly, “A Point of Vie on the Future advantage of silicon integration opportunities of IC Design, Testing and to meet the cost requirements of the MSOs. Manufacturing”, Electrical and Computer Many options exist for integrating EMTA Engineering, Carnegie Mellon University. SOME CONSIDERATIONS IN THE USE OF FORWARD ERROR CORRECTION (FEC) IN BROADBAND VIDEO DISTRIBUTION OVER WIRELESS LANS

Walter Boyles Consultant

Abstract Table 1 - Summary of Some 802.11 Standards MSOs seeking to deploy broadband video services to subscribers who are using 802.11 Modulation Freq Max Theo Theo wireless LANs should consider the use of Standard Ban Link Max Max d Rate TCP UDP forward error correction (FEC) technologies Rate Rate as well as the pros and cons of the use of streamed video versus file-based video 802.11b CCK 2.4 11 5.9 71. GHz Mbps Mbps Mbps delivery. 802.11g OFDM/ 2.4 54 14.4 19.5 (w/ CCK GHz Mbps Mbps Mbps 802.11b) 802.11g OFDM/ 2.4 54 24.4 30.5 INTRODUCTION (11.g only ) CCK GHz Mbps Mbps Mbps 802.11a OFDM 5.2, 54 24.4 30.5 5.8 Mbps Mbps Mbps As MSOs look to deploy new services to GHz broadband subscribers, they find a number of home networking products and technologies The actual performance of the wireless in place in their subscribers homes in much LAN, of course, will be complicated by a the same way that MSOs find a variety of huge array of factors including (but not PCs with different CPUs, memory limited to) the distance within the home, the configurations, and operating systems. The configuration of the room, the number of MSO is likely to seek to deploy services such walls and other solid objects between the as broadband video to as much of the access points and the client, and even RF equipment in place as is reasonably possible interference. The 2.4 GHz modes of 802.11 to accommodate their subscribers. can incur interference from a variety of other appliances in the home including cordless 802.11 phones and microwave ovens.

The majority of WLAN products in home In fact, even other factors such as packet use employ one form of the 802.11 standards. size will affect the performance of the The more well known versions of 802.11 are wireless network, as well.1 Of course, the summarized in Table 1. networking equipment itself which relies on chip sets in different versions from a variety of vendors will be one of the largest determining factors in the performance of the wireless network.

In open office environments, the observed performance of 802.11a using equipment

m based on previous generation semiconductor where p b is the bit error rate (BER) of the technology generally maintained at least 1 PHY mode at a given SNR (signal to noise Mbps for distances of up to almost 60 feet ratio). and approximately 3 times that distance for equipment using a newer version of the Bypassing most of the mathematics, in chipset. 2 this discussion Shannon introduced the idea of codes as ensembles of vectors that are to The performance of the more common be transmitted. If the number of vectors is 802.11b in a similar open office environment K=2k each vector can be described with k generally stayed above 1 Mbps for up to bits. The vectors are assumed to be of equal about 140 feet. 2. However in general usage length and we refer to this length as the block such as in enclosed offices with walls, closed length. For a length of vectors “n”, then n doors, and other solid objects common times k bits has been transmitted and the 802.11b home networking equipment, resulting code of has a rate of k/n bits per commercially available in 2003 and 2004, channel.3 was not reliable at data rates above 700 kb/s. Shannon went on to prove the existence of Forward Error Correction: these codes that allow us to approach the capacity of the channel and since his The idea of using forward error correction groundbreaking work, a variety of coding (FEC) in communications is of course well- techniques have been developed. This ended established and dates back to the birth of the concept that reliable communications information theory and its legendary inventor meant that transmission power must be Robert Shannon. Shannon, of course, proved increased and/or messages must be sent that there is a capacity for any repeatedly. By the early 1990s, the best communications channel and that reliable coding solutions had achieved actual communications is possible for rates capacities within 3-5 dB of the theoretical approaching that theoretical capacity. That limit shown by Shannon. theoretical capacity is determined as follows: A breakthrough occurred in 1993, when C= W log2 (1 =P/N) two French engineers, Berrou and Glavieux, introduced turbo coding that resulted in a where: 3dB improvement over existing coding schemes. Turbo codes can be classified as C is the channel capacity in (in bits/sec) turbo convolutional codes (TCC) and turbo W is the bandwidth (in Hertz) product codes (TPCs). 3 P is the transmitter power (in Watts) N is the noise (Watts) As coding technology advanced and the desire for better and simpler codes increased From information theory, techniques a class of codes previously discovered by emerge to model a channel that can be used Robert Gallagher some 30 years earlier to predict the probability of an error for N reemerged. This coding scheme called LDPC bytes of packet length such as below: – or low-density parity checking codes perform closer still to the Shannon limit. m m 8N P e (N) = (1-p b)

A comparison of LDPC code to PCCCs LDPC coding works particularly well for (parallel concatenated convolutional codes), file distribution. There is an added advantage SCCCs (serially concatenated convolutional in the distribution of large files since the use codes), and TPCs is shown in Table 2. of a large file mean your code is operating over a large number of bits. Thus, the loss of Table 2 - Comparison of Turbo Coding any small number of bits is likely to be Schemes 4 corrected quite easily with an extremely high probability. Attribute PCCC SCCC LDPC TPC Code Rate fair poor good poor Take the example of a one hundred (100) Block Size good good poor poor Modulation good best good good packet message, that is transmitted, using an Complexity fair fair good good LDPC code with 10% overhead. The loss of (throughput) say any 7 packets has less than a 10-6 chance Performance poor/ good/ fair/ fair/ best fair good good of resulting making the video file unviewable. Or put another way, the A large number of researchers worked on probability that any particular 7 message LDPC codes which has also resulted in a packets that are lost would result in a variety of vendors for coding products that correction that could not be performed is less 5 use LDPC. Some of these additions to the than 1 in a million. basic LDPC theory bear the names of the researcher or company that has developed Now, of course, there are other issues with them and each puts forth some advantages regard to packet loss. One view is that over the base LDPC coding developed by packets lost will typically be the same from Gallagher. However, LDPC coding (as one subscriber as another in an IP Multicast shown in Table 2) provides a good distribution scenario. So, the idea that you compromise of attributes for use in file-based can request replacement packets that were broadband video distribution over wireless lost and correct the files that way has been LANs. suggested.

The codes can be implemented to utilize In this author’s view, while there are variable amounts of overhead to provide undoubtedly network incidents that occur different degrees of error correction which result in the loss of the same packet or depending on the noise on the channel – packet for some number of subscribers, there overhead may typically vary from 5%-50% are also random events not only at different and provide resistance to data loss. points in the network but on each subscribers computer, as well. These random and Considerations of Actual Use of FEC: singular losses are better corrected by using an LDPC coding scheme with say 5%-10% Now, in actual use, the success of overhead. This amount of overhead is barely achieving data integrity with a coding noticeable and in the case of IP Multicast scheme such as LDPC is due to a number of distribution the overall efficiency of content factors (some mentioned here) as well as the distribution is so higher anyway as opposed overhead and the particular coding scheme to unicast file distribution that an additional chosen. Of course, one critical issue is the 10% of overhead is a very small price to pay. way the video content itself is distributed.

Of course, FEC can also be used in video CONCLUSION: streaming. The one penalty incurred in using FEC for streamed video is that a delay is Overall, however, the one factor that can imposed on the stream to perform the error not be overcome in a wireless LAN is of correction calculations over some number of course the network performance. If you can packets. The issues regarding what is an not reliably distribute data at over 700 Kb/s, acceptable delay are dependent upon a due to whatever factors, then the use of FEC number for factors including amount of delay is not going to allow you to distribute video is tolerable to the viewers. at a higher data rate whether it is streamed or delivered file-based. Then again, in a streamed video environment, the loss of a few packets may For streamed video, the limitations of the not be significant in the same way as it is in wireless LAN performance will of course file-based video delivery. The loss of few also limit the encoding rate you can choose packets might only momentarily degrade the for streamed video. You will always be signal and still make the quality of the video limited in video resolution by the lowest experience acceptable to the subscriber. common denominator of what data rates the network can tolerate. Again, the question of what is acceptable is dependent upon the many factors from the A file-based video distribution system has video quality that is trying to be achieved to no such limitations. In fact, even across a the wireless LAN performance from the network with a 700 kb/s or less bandwidth access point to that particular subscriber to limitation, it is easily possible to distribute the length (in time) of the disruption, as well files that are encoded at high definition as what packets are lost and how frequently quality. The use of FEC and IP Multicast the loss reoccurs. makes such distribution across home wireless networks in place today entirely possible and The delay considerations may also make implementable. the preferred coding scheme in streamed video different than that of the coding Table 3. summarizes some of the scheme in file-based video distribution. considerations in video delivery methods and the use of forward error correction of current One such proposal for video streaming is wireless LANs. From the table, it can be the system for packet loss protection for the seen that the combination of FEC with file- H26L-FGS. This employs Reed Solomon based delivery and is very efficient – coding along with a system of requests and especially when these files are delivered with acknowledgements. The combination of IP Multicast. feedback and error correction may be more suitable than in streaming than the FEC coding schemes used in file-based video delivery.6

Table 3 - Comparison of Video Delivery and Bibliography: Quality w/ FEC. 1. Overhead Constrained Packet Video, Video Tolerance to FEC FEC Achievable Distribution Minimal Delay Value Video Multimedia Communications Laboratory, Method Packet Loss to Quality University of Texas at Dallas, 2002. Video over Most Quality WLANs 2. 802.11 Wireless LAN Performance, Streamed Fair None N/A Worst Atheros Communications, 2003. w/out FEC 3. LDPC: Another Step Toward Shannon, File-Based Poor None N/A Fair w/out FEC Tony Summers, CommsDesign.com (EE Streamed w/ Good Worst Fair Fair/poor Times), Oct., 2004. FEC File-Based Best Not Best Best 4. Forward Error Correction, TrellisWare W/ FEC Notic Technologies, 2005. ed 5. Private Conversation, Bhavan Shah, 2005. 6. Packet Loss Protection of Scalable Video Bitstreams Using Forward Error The author can be contacted at Correction and Feedback, Charfi and [email protected]. Hamazoui, 2003.

SUB-BAND DIVISION MULTIPLEXING (SDM) INCREASES BANDWIDTH EFFICIENCY AND PROVIDES HIGHER TOLERANCE OF COMPOSITE DISTORTIONS

Mark E. Laubach, Yi Ling, William J. Miller and Tracy R. Hall Broadband Physics Inc.

ABSTRACT I. INTRODUCTION Bandwidth needs of customers on cable plants have increased dramatically over SDM stands for Sub-band Division recent years and will continue to rise in the Multiplexing. It is a technique of dividing near future. Increasing the order of QAM RF spectrum into multiple and equal-sized modulation has been the most popular way sub-bands using filter bank structures. When to satisfying the bandwidth needs until the basic sub-band filter is designed to have recently when nonlinear distortions and steep roll-off property, the spectrum of each limited dynamic range in HFC systems have sub-band has little overlap into the proved to be an obstacle for reliable 256 neighboring sub-bands. Combined with QAM service. Hence, it is of both practically SDM’s inherent frequency and time necessary and theoretically interesting to orthogonal basis, each sub-band is highly investigate approaches other than QAM to independent of other sub-bands, in both increase bandwidth efficiency and to frequency and time. For this independency provide higher tolerance of composite property, SDM can be used as a digital distortions all at the same time. Sub-band signaling scheme to transmit data stream in Division Multiplexing (SDM) is one of these parallel over the multiple sub-bands that it new approaches. In this paper, we give an creates without having large inter sub-band introduction of the Sub-band Division interference. Used as such, SDM is a multi- Multiplexing (SDM) technique based on band carrier-less modulation. Each sub- filter bank scheme and wavelet mathematics. band operates in the same manner, but with SDM represents a philosophical change a different offset frequency. comparing to standard QAM in terms of baseband signal formulation and alphabets SDM can be applied to digital selection. To show this change, the communication over many different types of fundamentals of SDM will be overviewed. media, such as cable [1], power line and Unique characteristics resulting from the wireless. The first proposal of applying SDM fundamentals will also be presented. It SDM in high-speed data communication will be shown that these characteristics was made almost ten years ago by Miller implying multiple advantages of SDM over [2], the founder of Broadband Physics Inc. equivalent QAM on cable applications, Later, similar ideas of using filter bank especially the tolerance of composite techniques for data communication were distortions. Simulation results and proposed in other literatures, such as [3]. measurements on the Broadband Physics However, since the 1990s, Broadband prototype system from the lab and field Physics Inc. has been the active leader in trials also will be presented in the paper to developing the SDM technology to be verify the theoretical results. implemented in a variety of applications over different channels. Currently, II. SDM AND ITS CHARACTERISTICS Broadband Physics Inc. is focusing on developing SDM modems for cable Based on a well-designed band-pass filter downstream applications. It will be argued with high stop-band attenuation, a filter in the sequel that SDM has numerous bank can be constructed by frequency benefits in the cable downstream shifting this filter prototype and combining applications. The most prominent two are them as a polyphase filter. Assuming that increased bandwidth efficiency and higher the transfer function of a prototype band tolerance of composite distortions. In its pass filter is implemented form, SDM also demonstrates many advantages over popular digital L = )()( znhzF −n (1) modulation schemes other than QAM, such 0 ∑ 0 −= Ln as Orthogonal Frequency Division Multiplexing (OFDM). These advantages where = ez sT and T is the sampling time, if include but not limited to achievable higher bandwidth efficiency with lower system we over-sample it at M times the original complexity, less sensitive to phase noise, sampling rate, then the over-sampled version highly resilient to multi-path impairment. of F0 (z) is All these advantages make the SDM particularly well suited for wireless MML −+ 1 = )()( zkhzF −k applications as well. However, this paper ∑ −= MLk discusses all the said characteristics of the (2) SDM but focuses on its high bandwidth efficiency and tolerance of composite T s distortions for cable applications. The rest of where = ez M . the paper is organized as the following: Section II gives an overview of the For = = − LLnnMk ,,...,0,...,, we have fundamentals of the SDM and discusses some unique characteristics resulting from = = nhnMhkh ).()()( those fundamentals. This section further 0 discusses the advantages of the SDM closely (3) connected with its characteristics. Section III discusses the bandwidth Separating (zF ) into its polyphase efficiency of SDM and through an example, components, we obtain shows that SDM has higher actual bandwidth efficiency than the theoretically zF )( = L equivalent QAM. −kM kM +− )1( Section IV discusses the resiliency of SDM ∑ ++ )1()(( zkMhzkMh to composite distortions in depth. −= Lk Section V presents the results from a + ... (4) Broadband Physics Inc. prototype system ( −++ zMkMh MkM −+− )1( ))1 test in a simulated cable plant. −1 zFzzF ++= ...)()( Section VI concludes the paper. 0 1 M −− )1( + M −1 zFz ),(

where 0 (zF ) is the same as defined in (1) These data streams can come from because of (3), and its polyphase versions different alphabets or constellations. For are instance, Aj comes from a four state Amplitude Modulation (4-AM) alphabet, L −kM taking values from {-3, -1, 1, 3}, while Aj+1 1 = ∑ + zkMhzF ,)1()( −= Lk comes from 8-AM, taking values from {-7, - L 5, -3, -1, 1, 3, 5, 7}, etc. Delaying A with = + zkMhzF −kM ...,,)2()( (5) +dj 2 ∑ T −= Lk respect to Aj d −+ 1 by M for 0 ≤ ≤ ld , and L combining with all zero data streams for the = ()( −+ zMkMhzF −kM ,)1 M −1 ∑ unused branches, we have the following −= Lk M T combined data stream with rate s T for = ez M .

jj + aaaA lj −+ 0)1(0)1(0 ,0,...,0,,...,,,0,...,0: Here we can construct a filter bank using j aa lj −+ 1)1(1 ,0,...,0,,...,,0,...0 (7) the prototype 0 (zF ) and its frequency- ,...0,...,0 shifted versions of F 21 M −1 zFzFz )(),...,(),( by choosing the prototype so it’s frequency shifted versions form an M-band quadrature- Transmitting this combined data stream mirror filter bank, the impulse responses for through the complete filter bank (zF ) is each sub-band filter are wavelets orthogonal equivalent to transmitting M in both time and frequency. The polyphase ,..., AA ljj −+ 1 (expanded to rate T through construction makes it computationally expanders) separately through the efficient. With this filter bank, we can corresponding branches, delaying by one transmit data in parallel through all or part clock from each other and combining them of the M branches of the filter bank with at the output as shown by the equation each branch being delayed by one sample below T clock M from the previous branch. Due to the clock delay between each branch, we can lj −+ 1 ])[( = − j AzFzAzF ][)( combine the outputs of each branch to form ∑ k ke = jk a single transmitting signal. To express this (8) combined transmitting signal in equation, we assume that branches where the square bracket denotes the

filtering operation and ke , ≤ ≤ + ljkjA −1 j lj −+ 1 zFzF )(),...,( ,0 1,.., ≤−+≤ Mljj −1, are the expanded A ,..., A . j lj −+ 1 are used to transmit data. The data streams We can also illustrate this process in 1 spectrum plots. with rate T for these branches are defined as Figure 1-1 shows an example spectrum of the output of a single branch. aaA jjj 10 ,...,:

aA ,...: jj ++ 0)1(1 (6) ...

aA ljlj −+−+ 0)1(1 ,...: 70 orthogonality of the filters can be compromised if, for example, there is a 60 ) group delay variation across the subbands. B d 50 ( e The resulting self-interferences are relatively d 40 u ti n easy to remove if the prototype filter has g 30 a very steep roll-offs Another useful view of M m 20 u the above transmitter and receiver structure rt c 10 pe is the wavelet transform. The transmitting S r 0 e filter bank is a form of inverse wavelet w o P -10 transform.. Each data symbol is amplitude -20 modulating a wavelet, the data stream to be Frequency transmitted is a signal vector comprised of Figure 1-1: Single sub-band spectrum modulated wavelets or equivalently, a vector expressed by a set of base functions in the Figure 1-2 shows an example spectrum of wavelet transform domain. Hence, the the combined outputs from all active receiving filter bank just needs to be a branches. wavelet transform [2], [4].

70 Comparing to other typical digital modulation schemes, such as QAM, SDM 60 ) has several unique characteristics. All the B d 50 ( e characteristics of the SDM are the results of d u ti 40 the basic construction of the SDM as shown gn a 30 above. Furthermore, these characteristics are M m ur 20 the reasons behind the advantages of SDM ct pe comparing to other typical digital S 10 r e modulation schemes. Here we discuss some w o 0 P of them. For more detailed comparisons -10 between SDM and other modulation Frequency schemes, e.g. QAM and OFDM, please see Figure 1-2: Combined multi sub-band [5]. spectrum First, depending on the design of the In view of the spectrum plots in Figure 1- prototype filter, 50dB plus stop-band 1 and 1-2, we see why the name “Sub-band attenuation is achievable by the SDM Division Multiplexing”, each branch creates spectrum. This characteristic of the SDM a sub-band of the entire spectrum; the makes an extra pulse-shaping filter overall spectrum is the combination of all unnecessary at the transmitter. With typical the sub-bands. The performance shown in additional 10% bandwidth required for a Figure 1-2 has been achieved in digital pulse-shaping filter, (e.g. raised cosine filter) hardware by Broadband Physics, Inc. for QAM, SDM with the same theoretical With such transmitter structure, we can bandwidth efficiency has more effective easily see that the receiver structure is bandwidth efficiency. essentially the same as the transmitter with the same filter bank structure and Second, each data symbol is confined to correspondingly matched filters filtering the its respective sub-band by the filter bank as received signal into separate sub-bands. The seen in Figure 1; consequently, the abrupt Fifth, recalling that the SDM transmitter transitions of the data symbols do not cause and receiver pair is essentially a wavelet ringing in the channel unlike the multi- transform pair. Any equalization to be done carriers in OFDM. Since no ringing effect in the receiver while following the receiving exists, cyclic prefix is not needed in SDM. filter bank is not in time domain. Thus the In OFDM the length of the cyclic prefix can equalizer structure can be made significantly sometimes exceed 30% of the symbol less complicated than the time domain duration. Thus SDM can achieve adaptive equalizer typically seen in QAM significantly higher effective bandwidth systems. efficiency comparing to OFDM. Like OFDM, the bandwidth of each SDM sub- The above five properties and advantages band and the number of sub-bands are of SDM are not the only benefits of using design parameters and can be optimized SDM, rather they are the five most essential accordingly for the channel. Furthermore, by and also most intuitive to describe at this selectively choosing active and inactive time. sub-bands, we can tailor the entire transmitting spectrum to the actual operating A note of notation is helpful here and for channel. The benefits of such flexibility are the rest of the paper, depending on the numerous, including easy fitting under number of bits a symbol in each subband emission masks and easy mitigating narrow represents, the corresponding SDM scheme band interference [4]. Because of the very is called L-SDM for L bits per symbol in a high stopband attenuation of SDM subbands single subband, or equivalently, for an there is an advantage compared to OFDM, alphabet of size S = 2^L for the subband. which typically has only 13 dB attenuation between adjacent bands. In the next two sections we will discuss advantages of SDM in cable applications, Third, SDM can transmit a real signal, especially its capability of providing actual i.e. signal with only I component but no Q higher bandwidth efficiency and mitigating component without being limited to lower composite distortions. bandwidth efficiency. This can be achieved by choosing the data stream for each sub- III. SDM PROVIDES INCREASED band from some amplitude modulation ACTUAL BANDWIDTH EFFICENCY (AM) constellations (same constellation is not required for different sub-band). Since Since one of the current major driving the transmitted signal is a real signal, the forces of digital cable technology decision region of the receiving slicer is development is the need of bandwidth, any single dimension, it can stand up to higher modulation scheme that can provide higher phase noise than signals with both I and Q than current standard bandwidth efficiency components carrying the information. (measured in bits/second per Hz or bps/Hz) Fourth, SDM is orthogonal in time allowing will increase the raw digital capacity of the the overlap in the transmission of one spectrum without increasing the actual symbol with previously transmitted symbols bandwidth. SDM happens to be such a [5]. The amount of symbol to symbol modulation scheme. overlap is a design choice. To study the bandwidth efficiency of the SDM, we can start with an example for cable applications. Each current cable channel is 6 MHz wide. To formulate a guard bands, so the actual number of baseband transmitting signal on one channel subbands for data transmission is 60 – 2 = using SDM, we can first choose the digital 58. For the convenience of presentation sampling frequency to be 2f (f > 6 MHz to here, assume that each subband uses the satisfy the Nyquist theorem). Then, we same 16-AM alphabet with 16 states on a divide the spectrum of 0-f Hz into M single real axis. Each subband has subbands, each subband has bandwidth of bandwidth efficiency of f/M and the symbol rate for each subband is log2(16)*2*100,000/100,000 = log2(16)*2 2f/M. For convenience, f can be chosen such = 8 bps/Hz, the same as 256 QAM. The that f/M is an integer. We can turn on any actual overall combined channel data rate block of continuous 6*10^6*M/f subband to (accounting the two inactive guard create a 6 MHz cable channel. Two of the subbands) is 58*8*100,000 = 46.4 Mbps, active subbands at both edge of the channel the actual overall channel bandwidth can be turned off to avoid interfere with the efficiency is 58*8*100,000/(60*100,000) = neighboring channels. 58/60*8 = 7.7 bps/Hz, which is less than 4% lower than the theoretical bandwidth Again, for the convenience of efficiency of 256 QAM. However, presentation, we assume that each active considering the excess bandwidth of QAM subband has the same alphabet that has S from the pulse shaping filters such as root states. So the bandwidth efficiency of each raised cosine filter, the actual 256 QAM subband is log2(S)*2f/M/(f/M) = 2*log2(S) bandwidth efficiency can be 10% less than bps/Hz, therefore, the bandwidth efficiency the theoretical value of 8 bps/Hz. The other of the combined channel is also 2*log2(S) possibility is that each subband uses the bps/Hz if we just ignore the two inactive same 32-AM alphabet with 32 states on a guard subbands at the channel edges for single real axis. With this set up, the actual now. channel data rate is 58*log2(32)*2*100,000 = 58 Mbps, the actual bandwidth efficiency With the general derivations above, we is 58*2*log2(32)/60 = 9.7 bps/Hz, which is can look at some actual numbers for the about 3% less than the theoretical bandwidth example. Assume that the chosen digital efficiency of 1024 QAM, and could be more sampling frequency is 51.2 MHz, the entire than the actual bandwidth efficiency of 1024 useable bandwidth for SDM is 51.2/2 = 25.6 QAM if the 10% additional bandwidth MHz. By using SDM technique, we can required for pulse-shaping filter is divide 25.6 MHz into 256 subbands; each considered. subband is 100 kHz wide with a symbol rate of 200 kHz. To create a 6 MHz wide cable Our observation from the example is that channel, we can choose any 60 continuous based on the first characteristic of the SDM, subbands out of the entire 256. Please note no pulse-shaping filter is needed at the that the entire 256 subbands are for signal transmitter; hence SDM can achieve higher formulation purpose only, no actual energy actual bandwidth efficiency than equivalent being put on them except the 60 chosen QAM. subbands, the real occupied spectrum is still only 6 MHz wide, the other unoccupied In view of the example derivation above, spectrum is free for other uses. Suppose we we have choose 60 subbands from 9 MHz to 15 MHz, two edge subbands being turned off as SDM actual channel bandwidth efficiency = -2 Number of active subbands * subband bandwidth * bandwidth efficiency of each -4 subband / Channel bandwidth = Number of -6 active subbands * bandwidth efficiency of -8 each subband / (Number of active subbands R E -10 + Number of guard-bands) B 0 -12 (9) g1 ol 3SDM 4SDM 5SDM -14 Using (9) and noting that the number of -16 the guard-bands is always 2 regardless of the -18 theoretical bit-true simulation channel bandwidth due to the extremely lab-measured -20 23 27 31 35 39 43 47 high stop-band attenuation of the SDM SNR (dB) subbands, we can see that the actual bandwidth efficiency will get better when Figure 2: AWGN BER performance of the number of active bands increases with SDM any increment of the digital cable channel bandwidth (e.g. 12 MHz or 18 MHz). Noting that the choice of alphabet for Thermal noise performance for SDM can be each subband is independent of other calculated and simulated by Additive White subband, we can in fact choose the alphabet Gaussian Noise (AWGN) model. Since the for each subband differently. The freedom theoretical AWGN performance for a SDM of doing so allows us to set the bandwidth subband with S-AM alphabet (S = 2, 4, 8, efficiency/modulation density for each 16, 32, 64, …) is equivalent to a S^2 QAM, subband according to the channel condition, so 1SDM, …, 5SDM and 6SDM with thus finely tune and optimize the trade off corresponding 2-AM, …, 32-AM and 64- between error performance, channel signal AM alphabets have the same AWGN to noise ratio and over all bandwidth performance as QPSK, 16 QAM, …, 1024 efficiency. As we will see in the next QAM and 4096 QAM, respectively [5]. section, the ability of changing bandwidth Bit-true simulation results and lab-measured efficiency on a fine frequency scale instead data on the baseband prototype system in of on a whole channel scale will help SDM Figure 2 show that both BERs are close to providing higher tolerance to composite the theoretical values. distortions.

SDM channel also has higher capacity/cost ration than combination of logical QAM channel. One of the major reasons is that SDM receiver has less complexity than comparable single QAM receiver for the fifth characteristic of SDM. In the final version of the paper, we will give more detailed comparisons between SDM and logical combination of QAM channels.

IV. SDM MITIGATES COMPOSITE that forms a sharp notch at the interferer DISTORTIONS frequency. However, to achieve a sharper notch through adaptive equalization, higher Composite distortions are produced by number of equalizer taps is required. amplifier nonlinearity caused Naturally, higher number of equalizer taps intermodulation of analog TV carriers. The requires more demodulator complexity and dominant components of the distortions are more system throughput delay. In the case of Composite Triple Beats (CTB) and non-blind equalization, which uses a training Composite Second Order (CSO) [7] [8]. sequence to obtain the optimal equalizer These distortion components typically have taps, higher number of equalizer taps also average power levels 12~15 dB below the requires longer training sequence. The thermal noise level. Even though their low longer training sequence again causes more average power levels appear to be harmless, throughput delay and overhead. due to their statistical properties, the random peak envelope power can be significantly Adopting the SDM approach for higher to cause large performance baseband modulation will achieve actual degradation for 256 or higher order QAM bandwidth efficiency higher than 256 QAM [8]. and obtain inherent capability of mitigating composite distortion effect. In the event of Other than asking operators to control excessive composite distortions, as we will and improve CSO/CTB levels through discuss in the following, SDM does not need carefully choosing channel frequency offsets to fall back to a lower bandwidth efficiency and maintaining head-end transmitter mode completely unlike 256 QAM has to aggregate noise power at low levels, etc., the fall back to 64 QAM for the entire channel. main methods to mitigate composite Instead, SDM can fall back to a lower distortion at the baseband digital modulation modulation density only at the subbands level that have been proposed include being affect by the composite distortions increasing interleaver depth and improving most severely, thus allowing reliable service adaptive equalizers [8]. Just by looking at without significantly lowering the the performance data, these two methods bandwidth utilization. In addition, SDM is appear to be adequate. However, if we look not contradicting with those proposed closely, there are problems associate with improvement done by the operators or each of them. proposed longer interleaving depth. When these improvements are available, SDM can For longer interleaver depth, first we work with these methods to provide an even know that the prices of increased higher tolerance to the composite interleaving depth include increased latency, distortions. When the limits of the HFC which affects the quality of service in plants or other restrictions render these another way. Second some longer methods unusable, SDM alone still can interleaving depth required to handle provide a more reliable service. CSO/CTB transients are not even supported by lots of set-tops. To fully understand the effect of composite distortions on SDM, we first look For improved adaptive equalizers, at some properties of CSO/CTB. according to [8], it is possible to have adaptive equalizers to converge to a state Since CSO/CTB are produced by 6/8*38% + 8/8*62% = 90% of the normal intermodulation of analog carriers, they are channel capacity, but still gain better of narrow frequency nature. The typical reliability. power bandwidth of an individual beat is about 10- 20 kHz. Remember that the To look further at the error correction and bandwidth of each subband of a SDM interleaving protection, we note that the channel is a design choice, we can choose it duration of a random composite distortions to be convenient for overall system pile up burst is often inversely proportional requirements yet wider than a typical to the distortion power bandwidth [8]. Given composite beat component. In the example a 10 kHz power bandwidth, a distortion of Section III, we used 100 kHz as the burst can be 100 µs long. Since these bursts subband width. It is indeed wider than the are highly localized in subbands, they only typical bandwidth of a composite beat affect the symbols in one subband. Recall component. from Section II, the interval between two consecutive symbols in the same subband is Another important property of the 5 µs for 100kHz wide subbands. So a CSO/CTB distortions is that the locations of distortion burst of 100 µs only covers about all the beat components can be calculated 20 symbols in the same subbands. The fact [1], [9]. Using the information of the beat that a distortion burst only affects a low components locations, we can check the number of symbols implies that an composite distortion locations within a 6 appropriately chosen Reed-Solomon (RS) MHz cable channel. Calculation shows that type error correction coding can correct with 100 kHz wide subbands, only 14% of most errors caused by composite distortions sub-bands will experience direct beats bursts. In the final version of the paper, we products. Due to the narrow bandwidth will have more detail about the effect of RS nature of the composite distortions and high code on SDM and we will also show that a independency of SDM subbands, only the simple standard length interleaving will two immediate neighboring subbands will have the similar results as well. be affected, thus the maximum number of potential neighboring subbands to be V. TEST RESULTS OF A SDM affected will be no more than 24% of the PROTOTYPE SYSTEM overall subbands, leaving a minimum 62% of sub-bands clear of any CSO or CTB beat A Broadband Physics, Inc. SDM impact [5]. So even without protection of the prototype system has been tested over RF error correction codes and/or interleaver, in channel with some simulated composite the event of excessive composite distortions, distortions. The system modulation density we can lower the bandwidth efficiency on was set at 3SDM, which according to the those 38% affected subbands without previous sections provides a theoretical 64 changing the 62% unaffected subbands. As QAM equivalent bandwidth efficiency of 6 an example, we can lower the bandwidth bps/Hz. The actual system bandwidth efficiency of the 38% affected subbands efficiency is 5.8 bps/Hz (see Section III). from 256 QAM equivalent 8 bps/Hz to 64 The 6 MHz channel occupies RF spectrum QAM equivalent 6 bps/Hz to maintain the of 582~588 MHz and is centered at 585 reliability, while leaving the rest 62% MHz. An approximate “spread” interference subbands still at 8 bps/Hz. The end results is tone was generated with an occupied that the overall channel capacity is about bandwidth of 20 kHz. This interference tone was modulated at different frequencies neighboring subbands causes less BER within the channel. The overall uncoded degradation than interferences falling in the channel BERs were measured under middle of a subband. Because of the SDM different levels of interference tone and subband independency, the increased bit Additive White Gaussian Noise (AWGN) at errors due to the narrow band interferences constant level of –37dBc. Parts of the results are from one and two subbands, are shown here in the chart below. respectively, for the interferences falling in the middle of a subband and on the boundary of two neighboring subbands. Let -4.00 us assume that the average number of -5.00 increased errors caused by an interference with a certain power level falling in the -6.00 middle of a subband is Ed. Another interference with the same power level but -7.00 falling on the boundary of two neighboring -8.00 subbands will actually have half the

uncoded log10 BER interfering power in each affected subband, -9.00 or equivalently, 3 dB higher signal to interference ratio. In view of this, the -10.00 -38 -40 -42 -44 -46 average number of increased errors in both Tone (dBc) affected subbands will be Ed1 << Ed/2, and the overall number of increased errors will 584.50 MHz 584.55 MHz be 2*Ed1 << 2*Ed/2 = Ed. Hence, the 583.25 MHz 586.75 MHz overall BER degradations caused by narrow interferences falling on the boundary of two Figure 3: BER Performance of 3SDM neighboring subbands are less than over RF Channel with AWGN and degradations caused by those falling in the Narrowband Interference middle of a subband.

The most interesting feature of the chart The simple analysis above is not valid for is that for interferences with the same power the single band modulation or multi-band level, their effects are also dependent on modulation without subband independency. their locations within the channel. This In those cases, error degradations are not distinctive feature, which is not available determined by the interference power within with single band modulations, is a direct an individual subband or a spectrum result of the multi-band approach and the subsection, so the type of results in Figure 3 subband independency of SDM. can only come from multi-band modulation Recalling that the 6 MHz RF channel is schemes, with highly independent subbands, divided by 60 SDM subbands, each subband such as SDM. occupies 100 kHz bandwidth, we find that three of the four interferences, namely the In the near future, we will present more tones at 583.25, 584.55 and 586.75 MHz, test results to further verify the CSO/CTB are in the middle of a subband and the other mitigating capability of SDM. one (at 584.5 MHz) is on the boundary of two neighboring subbands. The chart shows that the interference on the boundary of two VI. CONCLUSION ACKNOWLEDGEMENT

In this paper, we have introduced the basic The authors wish to thank Mr. Steve concept of Sub-band Division Multiplexing Anderson and Mr. Rama Nagurla for their (SDM). Several essential characteristics of effort in constructing and testing the the SDM have been presented. As a digital Broadband Physics, Inc. SDM prototype modulation scheme, advantages of the SDM system and generating the test results. associated with its characteristics were also discussed. These advantages include: REFERENCES

1. M. E. Laubach, “Moving towards 1. Effective bandwidth efficiency is Shannon’s Limit, Sub-Band Division improved, as no pulse-shaping filter Multiplexing could help MSOs pump up is needed. existing cable”, CED Magazine, Reed 2. Transmitting can be tailored easily to Business Information, September 2003. fit under transmission masks since 2. W. J. Miller, U.S. Patent # 5,367,516, sub-bands are highly independent, “Method and apparatus for signal nearly orthogonal, and can be turned transmission and Reception”, November active or inactive easily. 1994. 3. Transmitted signal can be made real 3. G. Cherubini, E. Eleftheriou, S. Olcer to have higher phase noise resiliency. and J.M. Cioffi, “Filter bank modulation 4. Less receiver complexity can be techniques for very high speed digital achieved, as no time domain subscriber lines”, IEEE Communications equalizer is needed. Magazine, May 2000. 4. G. Strang and T. Nguyen, “Wavelets and In the latter part of the paper, we focused filter banks”, Wellesley-Cambridge on the application of SDM in cable Press, 1997. channels. We discussed that due to its 5. Broadband Physics, Inc, “SDM characteristics, SDM can provide higher Technology”, www.broadbandphysics. actual bandwidth efficiency than QAM with com, 2001-2004. the same theoretical bandwidth efficiency. 6. E. A. Lee and D. G. Messerschmitt, We further discussed that also due to various “Digital Communication”, 2nd Ed. advantages of SDM, especially its subband Kluwer Academic Publishers, 1994. independency, SDM can effectively 7. W. Ciciora, J. Farmer, D. Large and M. mitigating composite distortions caused by Adams, “Modern Cable Television CSO/CTB, thus relax the constraints on the Technology”, 2nd ed., Morgan RF channels and lessen the burdens on Kaufmann, 2004. operators. 8. R. D. Katznelson, Statistical Properties of Composite Distortions in HFC In Section V, we also presented the test Systems and Their Effects on Digital results for a prototype SDM modem to Channels, Cable-Tec Expo Proceedings, verify the theoretical results. SCTE, 2002. 9. T. B. Warren and J. Kouzoujian, “Some notes on composite second and third order intermodulation distortions”, Matrix Technical Notes, MTN-108, Matrix Test Equipment Inc., 2003. THE INTELLIGENT NETWORK: DYNAMICALLY MANAGING BANDWIDTH AT THE OPTICAL LEVEL

Gaylord A. Hart and Steven Robinson Mahi Networks, Inc.

Abstract between peak and average usage) and the migration to IP based services (which are CATV network bandwidth requirements connectionless, and therefore sometimes are evolving at a rapid pace, driven by unpredictable in traffic requirements). Layer deployment of new services, increasing onto this the need to maintain QoS across a penetration of existing services, and the wide range of services, and bandwidth ongoing transition from analog to digital management becomes even more critical to services. As services migrate to IP based and tomorrow’s cable network. on-demand content delivery, bandwidth requirements vary dynamically in real time, There are two key components to and a minimum QoS is required to ensure managing bandwidth in the future: the adequate service performance. All of these naturally changing bandwidth requirements place stringent requirements on the network. as the network and service penetration evolve over a period of time and the real-time Managing evolving & dynamic bandwidth bandwidth management requirements requirements is complex, but emerging imposed by everything-on-demand (EOD) reconfigurable optical add/drop multiplexers and the converged IP network. To minimize (ROADMs) allow transport bandwidth to be long-term CapEx and OpEx costs (and hence managed effectively at the optical layer. remain competitive), MSOs must plan for Moreover, ROADMs enable fully automated these future requirements today and deploy optical transport systems which eliminate systems that evolve with the network without design and cut-over errors, accelerate forklift upgrades and without compromising service delivery, and lower network costs. service quality. This paper explores evolving services, optical transport technologies, and bandwidth The basic HFC architecture is a long way management mechanisms in the context of from running out of bandwidth: increasing migrating to an all-digital, IP based network bandwidth in the future is relatively easy and that lowers both CapEx and OpEx costs. inexpensive to accomplish simply by subdividing optical nodes (i.e., reducing the number of homes served per node and INTRODUCTION thereby increasing the bandwidth available per home), and the equipment for doing so Bandwidth requirements in CATV exists today and is relatively inexpensive. networks are rapidly changing, driven by Alternatively, a significant amount of deployment of new services (VoIP and network bandwidth can be made available by HDTV), increasing penetration of existing converting the broadcast analog TV services services (DTV and cable modems), and the on the network to digital services (roughly a ongoing transition from analog to digital and ten to one savings in bandwidth). from circuits to packets. This picture is further complicated by the transition to on- Because the HFC access architecture demand services (with large variations largely relies on transparent transport pipes, deploying new services on the network the lowest network costs while maintaining frequently does not require any upgrades to an acceptable QoS. the HFC network itself, thus greatly lowering new service deployment costs and shortening Network engineers are used to designing time to market. These new services do, optical transport paths as static circuits however, place new requirements on the providing dedicated bandwidth to highly transport and switching components in front predictable traffic. As new CATV services of the HFC plant. are deployed, as penetration increases for existing services, and as services migrate to Over the next few years, the greatest connectionless delivery via IP, new demands changes in the CATV network will occur in are being placed on the transport network and front of the HFC access plant at those points transport engineering. At the same time, in the network where services are aggregated, rising customer expectations for service switched, and transported in the purely digital reliability are requiring more redundancy in domain. For the MSO, this includes ISP and the network. To keep up with these demands telephony POPs, content storage and and to address emerging requirements for origination points, regional and metro real-time bandwidth management (brought headends, primary and secondary hubs, and on by EOD and IP services), it is necessary large businesses where services will be the detailed optical transport network design, delivered directly via fiber. configuration, migration, and operational processes be automated, essentially masking A network’s architecture is bounded by much of the network complexity from the the optical transport paths connecting it, both engineering and operational processes while physically and from a bandwidth perspective. enabling accelerated network evolution. These paths can be a bottleneck to delivering sufficient bandwidth and QoS as the network In its fullest implementation, such an evolves. While it is common to engineer optically reconfigurable network would these paths to provide sufficient bandwidth dynamically modify itself in real time to under a pre-defined set of conditions, this can respond to changing network conditions and be costly, either requiring constant re- service requirements. This would be engineering to add incremental bandwidth as analogous to the way a router automatically needed or resulting in stranded bandwidth if discovers paths through the network and current traffic loads are far less than the dynamically routes packets to their deployed transport bandwidth. destination based upon changing network conditions and traffic requirements. It now This problem only worsens as services begins to make more sense to treat the optical migrate to content on demand and IP based transport layer (Layer 1 in the OSI model) as delivery. Both of these natural evolutionary an extension of the switched layers residing steps, though very efficient in only requiring above it, supporting integration of bandwidth bandwidth when it is actually needed, result management and QoS across the optical, in bursty traffic, which makes traffic Ethernet, and IP layers. engineering even more complex. Needless to say, network complexity and traffic All of these objectives for optical variability are only increasing, and new transport automation and a dynamic optical approaches are required for network design layer can be accomplished with ROADMs and operation in the future if we are to ensure using DWDM and GMPLS. When these technologies are coupled, an underlying real time based upon changing service and optically reconfigurable network is possible content requirements, as well as subscriber which flexibly and dynamically supports on-demand service usage. future network evolution as well as bandwidth and content on demand, and ROADM TECHNOLOGY without having to manually re-engineer and re-configure each component of the network OADMs have been deployed widely in when changes are made. optical transport networks over the last few years. Most of these rely on fixed Presently, dense wave division optical wavelength components (lasers, multiplexers, add/drop multiplexers (DWDM OADMs) filters, etc.) which require a significant allow MSOs to collapse multiple parallel amount of manual network design as well as service transport networks onto a common manual configuration and provisioning. A optical network that supports multiple new class of fully reconfigurable OADMs protocols and services over a single fiber. has recently emerged which enables With modern network planning tools, the automation of these processes as well as design of both the optical and service layers automated and dynamic network operation. may be fully automated. This simplifies and These ROADMs are built around flexible speeds up network design, eliminates the optical components which can be controlled need to memorize and understand complex via software and an intelligent control plane design rules, and reduces design errors. which supports process automation. The primary underlying technologies which OADMs, however, typically rely on fixed define ROADMs are outlined below. configurations and components which are determined at the time of the initial design. ROADM Architectures This restricts the degree of automation which may be applied in the future to network Several architectural alternatives, based on evolution and slows down the upgrade a variety of markedly different optical process itself. ROADMs, because they allow technologies, exist today. Early ROADM their configuration to be controlled remotely technology, based on discrete optical- via software, enable provisioning and mechanical switches, filters and variable operations to also be fully automated. This optical attenuators (VOAs), is shown in includes network topology discovery and Figure 1, below. service turn-up, as well as the dynamic monitoring and setting of network parameters for optimized operation. Modern O O P P ROADMs already support these capabilities M M today.

But as optical switching speeds increase OPM and the GMPLS control plane becomes more content-aware and tightly coupled to Layers 2 and 3 above it, ROADM based CATV optical transport networks will become even more powerful, being able to dynamically re- Figure 1. Discrete ROADM Architecture configure transport paths and bandwidth in While simple to implement, since most of design integrates the add Mux with the the technology is commercially available, wavelength blocker. This design eliminates this approach utilizes many discrete optical the extra add Mux, but at the expense of components. The result is very high insertion requiring additional optical switches. loss, very high cost, and large size—thus preventing its widespread acceptance.

The second architecture to be used for Splitter / Drop filters ROADMs, the wavelength blocker O P architecture, is shown in Figure 2, below. M Essentially, this design splits the incoming DWDM signal into a drop and through path.

An integrated DWDM demultiplexer OPM Fixed OPM (Demux), VOA and multiplexer (Mux) form drops the core of the wavelength blocker. Typically blockers are implemented using Micro-Electrical Mechanical Systems Figure 3. Combined Blocker / Adder ROADM (MEMS) or Liquid Crystal Display (LCD) To be cost effective, this design typically technologies. integrates all of the filters and switches into a

single module (similar to the wavelength blocker architecture). However, this again WB Module forces the MSO to pay for all wavelengths on Splitter / Add Filters/ Drop filters Combiner day one. While this design now manages add O wavelengths, it still does not manage drop P M wavelengths. Furthermore, this design now permanently locks the add wavelengths to a fixed wavelength design—thereby making OPM fully tunable lasers only good for sparing (no dynamic wavelength management). Thus,

this design is also a “semi-reconfigurable” Figure 2. Wavelength Blocker Architecture ROADM.

While this architecture reduces the Planar Lightwave Circuits (PLCs) are one number of discrete components, it forces the technology used to implement this combined MSO to pay for all wavelengths at each node blocker / adder design. While they have the on day one. Furthermore, this design only potential to integrate complex optical manages the through wavelengths—not the components onto one or more substrates add or drop wavelengths. In fact, most (such as silicon), manufacturing yield and implementations of this architecture use power management are still a challenge. inflexible fixed filters for the add and drop wavelengths. Hence this design is actually a More recently, a new architecture based “semi-reconfigurable” ROADM. on Multi-Port Wavelength Selectable Switches (MP-WSS) allows for completely A variant of the wavelength blocker reconfigurable ROADM functionality. This architecture is the combined blocker / adder architecture is shown in Figure 4, below. architecture shown in Figure 3, below. This Wavelength switches have the ability to direct one or more wavelengths from an manually placed front panel optical incoming DWDM signal to one or more interconnect jumpers to be moved inside to output ports (usually with individual VOA- the rear of the chassis. Reciprocal optical like power control for each wavelength). connectors on the back of each transponder allow these optical interconnects to be made when the transponder is plugged into the

chassis. Without the WSS, however, an optical backplane would require all slots to use pre-defined wavelengths and thereby O MP-WSS P waste precious rack space. M An optical backplane frees MSOs from Flexible Flexible the tangled mess of jumpers interconnecting Drops Adds the lasers & receivers to the DWDM filters and amplifiers and from the need to carefully Figure 4. Fully Flexible WSS Based ROADM map and record these interconnects. Operators simply connect the client-side service fibers and walk away. Note that in this architecture, the MP- WSS is located on the drop side. In this Tunable Lasers configuration, the ROADM can manage both the drop and through wavelengths. It can The recent advent of widely tunable lasers also direct multiple wavelengths to a single has enabled an equivalent capability for full drop port for low cost optical ring reconfigurability on the transmit side of the interconnection or physical mesh nodes. This ROADM. Initially, MSOs were interested in design is inherently more efficient because tunables primarily for sparing purposes. This the same demultiplexer is used for both typically yields a 32:1 savings in spares and through and drop wavelengths. provides tremendous cost savings. More

recently, MSOs have realized that tunables Another interesting aspect of this design is also substantially reduce delivery lead times the use of broadband (wavelength and allow for significant equipment reuse. independent) optical add ports. In this design, dynamic wavelength provisioning with full C-band tunable lasers is now More significantly, however, widely possible. MP-WSS technology enables a tunable lasers, when coupled with “fully reconfigurable” ROADM architecture. wavelength selective switches with single lambda add/drop granularity, enable greatly Optical Backplane simplified provisioning as well as dynamic wavelength management. Combined with an With the advent of the fully flexible optical backplane, remote provisioning and ROADM (using the WSS), practical optical management are also possible. Operators no backplanes can now be implemented on an longer need to match fixed wavelength lasers OADM chassis. Because wavelength with fixed wavelength filters using manually selective switching eliminates the need for configured jumpers during installation. manually configured optical jumpers, an Instead, operators can simply point and click optical backplane permits the 100 or more to remotely provision or re-provision wavelengths. This capability delivers For rapid changes (such as a fiber cut), tremendous OpEx and CapEx savings. ultra-fast amplifiers can now provide robust transient control by quickly clamping the Optical Power Monitoring and Control gain change to less than 0.3 dB in less than 1us. This accuracy is needed since the Initial OADM implementations provided transient is additive with cascaded amplifiers. only broadband, non-real-time optical power Combined with the ROADM capabilities monitoring of the aggregate DWDM signal. already outlined above, both per-channel & However, to properly set up, manage, broadband optical power level management equalize, and optimize wavelengths, real- can be fully automated and dynamically time optical power monitoring (OPM) and controlled in real time, thereby simplifying power level control are required for each optical layer design and network lambda at each node. Real-time, direct management and operation. measurements are more reliable and provide for rapid fault identification as well as GMPLS Control Plane support for newer protection schemes such as shared optical protection switching. Ideally, One final technology is required to fully and for a fully robust system, all optical automate the provisioning—an integrated inputs, outputs, adds, and drops would be control plane that supplies the intelligence monitored in real time at each node for and communication needed to control the system optimization and full fault isolation. ROADMs in the network. More recently, DWDM vendors have begun implementing Robust Variable Gain Amplifiers GMPLS (Generalized Multi-Protocol Label Switching) control planes. Typically a Early DWDM amplifiers required dedicated wavelength carrying an Optical operators to operate them in fixed gain mode. Supervisor Channel (OSC) is used to To ensure a flat gain profile, operators had to communicate the GMPLS messages between manually “pad-out” every span. This nodes. This wavelength is completely unnecessarily added noise to the DWDM independent of the service bearing signal—limiting reach. As new wavelengths wavelengths and also supports provisioning were added or span losses changed, the communications from the Craft Interface and system ran the risk of becoming unbalanced. Element Management System (EMS). Furthermore, any upstream wavelength changes (such as fiber cuts) could cause GMPLS enables DWDM nodes to downstream errors (on both working and seamlessly work together in a network to protection wavelengths). provide resource and inventory auto- discovery, topology information, service Newer DWDM systems are now being setup, signaling, path computation (including implemented with variable gain amplifiers Routing and Wavelength Assignments), light with transient control. Variable gain path setup, and management. Following amplifiers can provide continuous, automated industry standards, a GMPLS control plane broadband gain adjustments to correct for enables true cross-network, A to Z slow changes in span losses or a change in provisioning. the number of DWDM channels.

BANDWIDTH MANAGEMENT roll a truck to each node when an upgrade is performed. When wavelength selective switching, tunable lasers, and an optical backplane are Many modern software-based network combined in a ROADM along with a planning tools allow the optical layer design GMPLS control plane, a powerful platform is process largely to be automated. This is a created with an extensive set of capabilities. critical component in enabling flexible Because of these capabilities, ROADMs are bandwidth management in the future. A much more flexible in supporting service good planning tool also allows the MSO to migration and network evolution and in flexibly control design options (if necessary) automating these. For the following and to optimize the design based upon discussion, this paper assumes a ROADM selectable criteria such as cost or wavelength has all the attributes listed above (i.e., conservation. genuinely is reconfigurable), though this is not always the case. However, if an OADM uses fixed filters or other components which are not Network Design dynamically compensated for by the network itself, these must manually be entered into Bandwidth management in the optical the network design tool during the initial transport network begins with the initial design and may require subsequent re- network design. Limitations built-in to the engineering of the network when upgrades initial design can limit future growth and are performed. By eliminating many of these network evolution, or at least make them fixed components or by being able to much more painful and costly to achieve. dynamically compensate for them, ROADMs Ideally, MSOs should plan for one-time simplify the initial network design process network engineering, which allows and any subsequent upgrades. unrestricted and non-disruptive addition or removal of services, wavelengths, and nodes Network planning tools should also to the network at any time. support automated design of the service layer as well, allowing an MSO to select protection The initial system design, for example, options for common equipment and should readily support the migration (or mix individual wavelengths, service and and match capability) from 2.5G to 10G to transponder types, client interfaces, and any 40G wavelengths without re-engineering and other number of options. Once these are upgrading the network at each stage. selected, the planning tool will automatically Similarly, the ability to add or delete nodes in design the optical and service layers. At the the network without re-engineering optical layer, OSNR and dispersion amplification or dispersion compensation is compensation budgets are created, and the a significant advantage. ROADMs, because planning tool selects appropriate optical they are reconfigurable and can dynamically amplifiers and dispersion compensation adapt to new network configurations, make modules to ensure one-time network this possible. ROADMs also further simplify engineering. At the service layer, common and lower the cost of network evolution equipment and transponders are selected and because they are remotely configurable, assigned to specific chassis slots. which in most cases eliminates the need to A well-designed planning tool, once it has ROADMs which can dynamically provided a network design, will also provide optimize performance and optical parameters detailed network drawings, span information, at each node also allow the typical chassis drawings with slot assignments for engineering design rules to be relaxed, which each shelf component required, a bill of enables transport over longer distances and materials with part numbers, and even costs. through more nodes. All of these provide the The planning tool should also support MSO better bandwidth utilization and at a exporting these items to an external lower cost. spreadsheet for subsequent analysis, manipulation, or record keeping. Because Collapsed Service Transport Networks the planning tool can reduce the time required to design a network to a matter of Most CATV networks and the services minutes, it is easy to create different network delivered over them have evolved so rapidly scenarios for comparative evaluation, over the last few years that many MSOs have allowing the MSO to optimize the network deployed parallel networks optimized for design based upon other service or each service. While this made economic and architectural factors. technical sense at the time, it is not a sustainable model for a competitive market ROADMs, providing their underlying because it is costly to maintain, manage, components have been designed to provide operate, and upgrade these networks and fast enough switching times, also enable because it makes inefficient use of network more flexible protection options. Because resources. wavelength selective switches and tunable lasers allow receive and transmit wavelengths For most MSOs, the first stage in network to be assigned on the fly, protection evolution is collapsing these parallel wavelengths need not be assigned until an networks onto common network elements actual failure occurs. Such protection allows and infrastructure wherever possible. Multi- more than one path to share a common service, multi-protocol ROADMs are ideal protection wavelength, which means more for collapsing the transport sections of these protected services may be placed on a given parallel networks onto a single, unified fiber. This conserves fiber while still DWDM network (typically a ring) which providing complete service protection. provides fiber relief and service protection. Because ROADMs enable non-disruptive Good bandwidth management also upgrades, follow-on network evolution can includes conservation of network resources be transparently accomplished as new and performance optimization of these services are added and old services removed. resources. In other words, you should use as Because ROADMs also enable flexible few resources as necessary to get the job add/drop wavelength assignment, lambdas done, and you should get the most out of may constantly be recycled with service them. A well-designed ROADM conserves changes, allowing full utilization of existing fiber and wavelengths on a fiber by fiber resources. eliminating stranded bandwidth caused by wavelength banding and by enabling ROADMs with a GMPLS control plane unrestricted wavelength assignment and offer another significant benefit, as well. reallocation (single lambda granularity). Because GMPLS based ROADMs combine reconfiguration capabilities, intelligence, and communication between nodes, they also with an OADM simply by adding support automated network topology and transponders, providing sufficient lambdas inventory discovery. This eliminates the are available to support the new transponders, need to manually provision these into each or by migrating from 2.5G to 10G to 40G network node and the EMS. At the same transponders, providing the network and the time, this capability allows the network to OADM have been designed to support these. automatically turn up the individual ROADMs configured for one-time network wavelengths and services to be carried over engineering are ideal for supporting this type it. of growth. New transponders may be installed without any additional network Another key aspect of bandwidth engineering, and the ROADMs will management for these unified networks is the dynamically adjust to the new traffic load. ability to deploy only that bandwidth which And unlike many OADMs with fixed serial is needed to support today’s services, but at filters or multiplexers, the ROADM permits the same time to provide migration capacity transponders to be added without any service and capability for tomorrow’s bandwidth disruptions. requirements. MSOs should not only expect a ROADM to support mixing and matching Seamless service conversion or migration 2.5G and 10G wavelengths, but 40G is also readily supported by ROADMs. A wavelengths as well. While demand for 40G good example of this is the migration from wavelengths may not be strong today, it will TDM voice to VoIP. In this case, new be in the future. In keeping with the one- wavelengths are deployed to support the time network engineering philosophy emerging IP service, while existing outlined above, this should be accomplished wavelengths continue to be used to support transparently, that is, without any additional the legacy TDM service. As customers upgrade costs or new restrictions on span migrate to the new service, the legacy budgets, node counts, or ring circumference. DWDM wavelengths may be torn down and This pay-as-you-grow approach optimizes recycled on the transport ring for new CapEx by tying network costs to service services. ROADMs enable flexible and revenue, but it also eliminates or puts off into unrestricted wavelength recycling and non- the future costly forklift upgrades and the disruptive migration for existing services and deployment of new fiber. customers.

Transport Network Evolution As part of the bandwidth management role, OADMs may also be used to help Once any parallel single-service transport manage QoS. Some MSOs have elected to networks have been collapsed onto a segregate at various points in their networks, common multi-service DWDM optical at least initially, services which have network, the next stage of network evolution different QoS requirements and traffic becomes a matter of supporting bandwidth characteristics. For example, IP data services growth, service conversion, and new service and VoIP services, though both are IP deployment, as well as ensuring sufficient services, have very different QoS QoS is provided for all these services. requirements with respect to latency, jitter, and lost packets and very different traffic Bandwidth growth and new service characteristics with respect to average to deployment are relatively easy to provide peak bit rates. It is possible to segregate these services in the network to maintain network. As IP services begin to dominate, better control over QoS for each. legacy wavelengths can simply be recycled as no longer needed, freeing up additional OADMs can be used effectively here for transport capacity for IP services. However, parallel transport of each service on a significant challenges remain in deploying a separate lambda over a single fiber, allowing fully unified IP network. unique bandwidth and protection options to be applied to each service. OADMs also IP is a Layer 3 protocol and still requires typically support multiplexing transponders, transport by a Layer 2 protocol riding over a which carry many multiplexed tributaries Layer 1 Physical Layer in the network. over a single wavelength and can therefore While most MSOs have deployed some ATM support this type of service segregation, as and SONET in there networks, it has well. Of course, in the long run, the typically not often been a significant amount, efficiencies of common transport and and MSOs are now focused on Ethernet for switching of all these services will require Layer 2 and Optical Ethernet for Layer 1. convergence. ROADMs can be of particular OADMs can readily and simultaneously value here by enabling flexible and non- transport all these protocols and services and disruptive changes in the network to support can ease the transition as MSOs migrate their dynamic service requirements and also the networks. future transport convergence of these separate services when suitable QoS Ethernet is a wise choice for building mechanisms (such as MPLS) are deployed in CATV networks. Ethernet now scales the network to guarantee QoS under worst- effectively and inexpensively across LAN case conditions. and WAN environments, is very easy to deploy and manage, and as a connectionless Converged IP Transport Network service is well-suited for IP, which is also connectionless. Recent Ethernet standards A significant amount of equipment has for virtual LANs (VLANs) make Ethernet already been deployed in CATV networks to even more powerful and support QoS and support existing services and will not go traffic prioritization capabilities in Ethernet away in the near future. Nevertheless, voice, for the first time. video, and data services are destined for IP delivery. Indeed, data services in CATV When coupled with multi-protocol label networks are already there, with VoIP rapidly switching (MPLS), which allows even tighter emerging, and IP video to follow. coupling between the Ethernet and IP layers, Ethernet clearly offers advantages that cannot As this transition occurs, MSOs will be found with ATM or other protocols. continue to operate parallel legacy service Similarly, optical Ethernet, when coupled networks, as already outlined above, until with GMPLS, offers advantages that cannot such time comes when all customers are be found in SONET, which though widely transitioned to IP based services and the deployed and quite capable of transporting legacy equipment is taken out of service. Ethernet, is still a TDM transport technology. Multi-service, multi-protocol ROADMs can provide significant help in easing this As legacy services fade away and Ethernet transition while still allowing all services to and IP dominate, more wavelengths on the be delivered over a common transport optical transport network will be dedicated to Ethernet. While services may initially be can be particularly true for video, which not segregated for transport for QoS reasons, in only has relatively large bandwidth the long run this will not scale effectively, requirements, but which is also sensitive to especially in light of modern VLAN Ethernet latency and lost packets. switches which also support MPLS and can prioritize traffic. So the network will not Because on-demand usage is also tied to only migrate to Ethernet transport, but to customer preferences and other uncontrolled multi-service transport in each pipe with each variables, the peak to average ratio may also service provided sufficient transport vary significantly over time. For example, bandwidth to ensure reliable performance. television viewing typically peaks in the evening, but there can also be significant Many benefits are still to be derived from traffic peaks created by special or the over-subscription that statistical unpredictable events. And this can be true multiplexing allows, especially when larger across voice, video, and data services. While pipes can be used for transport of a larger multicasting in an on-demand environment number of services. So one can also expect a can mitigate the effects of these peaks for migration to wavelengths with ever more video, it cannot eliminate them. transport capacity (2.5G to 10G to 40G per lambda). Ultimately, it is also more cost Traffic planners are faced with a difficult effective to use fewer (but higher capacity) choice: provide sufficient bandwidth for peak switch and router interfaces. MSOs should demand and let a significant amount of the therefore, as they make transport decisions bandwidth go unused most of the time, or today, look for solutions that provide one- provide less bandwidth and let the network time network engineering and transport congest under peak traffic loads. The first capability for 40G wavelengths. option is expensive, and the second results in poor service. Of course, most traffic planners As CATV networks migrate to IP based try to hit a reasonable compromise, but with services, a parallel migration to on-demand service usage varying dynamically over time services is also taking place. Broadband data (both in real-time and as a result of access and VoIP services are already demand demographic trends), this is not always based services, for the most part only possible. Clearly, it would be desirable to consuming bandwidth when the services are have the network intelligently monitor its actively being used. Video on demand own loading and dynamically apply network (VOD) is also widely deployed, but most resources where and when needed. Such an CATV subscribers still receive their video intelligent network would allow sharing a services via broadcast analog or digital TV. smaller number of network resources across a wider range of applications and conditions. On-demand services have the benefit of only consuming network bandwidth when a ROADMs, given their flexibility and on- service is requested by a subscriber, thus the-fly reconfigurability, are entirely capable allowing statistical multiplexing at the of supporting such a dynamically configured service level. However, on-demand services network in the transport domain. Under such can also result in a very high peak to average a scenario, network planners could deploy a bandwidth usage ratio, which requires careful sufficient amount of transport bandwidth to network planning to prevent network cover average bandwidth requirements and congestion and poor service delivery. This provide additional uncommitted wavelengths which could be brought to bear when and appropriate paths for moving datagrams where needed to prevent congestion as traffic through a network (and to compensate for demand builds. Load-sharing switches or failed routers), GMPLS and its associated routers would provide complementary routing protocols can route wavelengths bandwidth delivery functions at the Layer 2 through the optical transport network as and/or 3 levels. needed to optimize network resources and capabilities. While current GMPLS and Since network bottlenecks can occur in MPLS standards do not support this degree both the transport layer (insufficient of integration across layers 1 2, and 3, this bandwidth to carry the traffic) and switching capability will no doubt be implemented at layer (insufficient capacity to switch traffic), some point. a dynamic transport network could provide additional wavelengths when transport An intelligent network built upon these capacity is exhausted or route existing principles would require less equipment since wavelengths around congested switches resources are optimally applied only when when switching capacity is exhausted. and where needed and would simplify operations since network reconfiguration is In this intelligent network, coordination done automatically and dynamically, and not between Layers 1, 2, and 3 would be manually by traffic planners. Overall, this implemented and automated by a GMPLS would result in greater CapEx and OpEx and MPLS control plane, which allows label- savings for the MSO while providing greatly switched paths to be created and torn down increased service reliability for subscribers. as needed. Because the overall network status would then be known at all layers, SUMMARY AND CONCLUSIONS optimization could be applied intelligently where it makes the most sense. CATV optical transport bandwidth requirements are constantly changing as a At the ROADM level, as congestion result of increasing service penetration, begins to build in the network, the GMPLS deployment of new services, and the control plane would assign and route transition to digital and IP based services. At wavelengths between nodes on the ring, the same time, on-demand services are providing additional bandwidth in real time rapidly being deployed and will likely where and when it is needed. This displace broadcast services at some time. mechanism could also be used to route traffic On-demand services significantly add to the around a congested network segment or complexity of traffic planning and bandwidth element, relieving traffic from congested or management. New demands are also being near-congested areas and applying it where placed on the network to ensure adequate sufficient resource are available to handle it. QoS is provided for each digital service to This approach can also be used to provide ensure reliable delivery. Service and network protection switching within the network by evolution is occurring at an ever-increasing dynamically reassigning wavelengths and or rate, and new real-time constraints are being transponders to compensate for equipment placed on the network which require dynamic failures or fiber cuts. optimization.

Just as routing protocols dynamically MSOs also face an increasingly allow routers to discover the most competitive service environment which requires reliable and low-cost service and dynamic monitoring and adjustment of delivery, rapid roll-out of new services, and network operational parameters for optimized the lowest possible CapEx and OpEx costs. performance. ROADMs are ideally suited To achieve these objectives, MSOs must for MSOs to meet their existing and future deploy equipment today that optimizes network transport requirements and also offer performance and efficiency on the existing the most cost-effective solutions for optical network infrastructure and which offers low transport, significantly lowering overall first-in costs, yet can evolve with the network lifecycle CapEx and OpEx costs. without forklift upgrades. This equipment must also provide a high degree of REFERENCES automation for network design, service turn- up, and operation, which in turn will result in "The Dynamic Network: Managing lower costs, more reliable services, and Bandwidth and Content on Demand at the accelerated service deployment and network Optical Level," Gaylord Hart and Zouheir migration. Mansourati, Proceedings Manual: Collected Technical Papers of the 2004 SCTE ROADMs deliver the best approach to Conference on Emerging Technologies. managing changing bandwidth requirements and evolution in the optical transport Author Contact Information network. ROADMs support automated optical and service layer design, as well as Gaylord A. Hart one-time network engineering. They provide [email protected] automated topology self-discovery and 303.910.7743 service turn-up, unrestricted wavelength usage and reconfigurability, and automated Steven Robinson [email protected] 732.465.1000 ext. 1248

THE MODULAR CMTS ARCHITECTURE

John T. Chapman Cisco Systems, Inc.

Abstract The motivation for this is several reasons. First, there is the promise of cheaper The architecture working group of Next- downstreams. The cost of a downstream on a Generation Network Architectures (NGNA) traditional CMTS is ten to twenty times the posed the question to the industry if a cable cost of a downstream on an edge QAM modem termination system (CMTS) device. This is mainly due to the fact that the architecture could be developed that could CMTS downstreams come with four to eight leverage edge quadrature amplitude upstreams attached to them. It is also due to modulation (QAM) devices that have been the fact that the CMTS is a more complex developed for the video-on-demand (VOD) piece of equipment, and the real estate inside environment. The author of this paper started the chassis is an expensive place to locate an working on this problem with a team of RF power amplifier. engineers in December 2003. Another key motivation is to develop a The resulting design was submitted to CMTS architecture in which the number of CableLabs in May 2004 and formally adopted upstream and downstream RF channels can be by CableLabs in January 2005 as the baseline independently chosen and configured. Today, design for the Modular CMTS (M-CMTS) adding a second downstream to a DOCSIS specification. The author of this paper is now group typically means that the operator also serving as lead author for the Downstream has to add an extra four to eight upstreams as External PHY Interface (DEPI) working well—something that is usually not very group. practical.

This flexibility which will allow CMTSs INTRODUCTION to increase their downstream bandwidth and lower the cost of the downstream, is required A new architectural concept for building in order for CMTSs to more effectively Data Over Cable Service Interface compete with new high speed competitive Specification (DOCSIS) CMTSs is underway services such as newer digital subscriber line at CableLabs. It is known as the Modular (DSL) standards such as the 44 Mbps bonded CMTS (M-CMTS). The concept is to extract family of asymmetric digital subscriber line the Physical Layer (PHY) out of the CMTS (ADSL) standards—called "ADSL2”—or and locate it in a separate network element. fiber to the home (FTTH). That separate PHY network element would be an evolution of the edge QAM network Figure 1 on the next page shows a element that has already been developed and multiservice architecture where the edge deployed for the VOD market. QAM device is shared between a DOCSIS network and a VOD network.

The Edge QAM device manages two DOCSIS network” [1]. distinct types of traffic. The first is MPEG 2 (Moving Pictures Group 2) over MPEG-TS This new bonding technology is the (MPEG Transport Stream) video traffic. The highest priority for the upcoming DOCSIS 3.0 second is IP over DOCSIS over MPEG-TS working groups to address. traffic. Typically, an individual QAM channel is only carrying one of these types of traffic, REFERENCE ARCHITECTURE although the edge QAM itself may have QAM channels that are part of either service. The reference architecture for a M-CMTS system is shown in Figure 2.This architecture The DOCSIS component provides the contains several pieces of equipment, along “triple play services” of data, voice over IP with interfaces between those pieces of (VoIP) and video over IP. These services are equipment. This section briefly introduces combined in the CMTS core and transported each device and interface. Subsequent to the edge QAM device typically over a sections will go into more detail. switched Gigabit Ethernet network. The edge QAM device, or EQAM for short, has its origins in the VOD environment. The data capacity of the DOCSIS It is a chassis that typically has one or more downstream can be 40 to 50 Mbps for Gigabit Ethernets coming in and multiple traditional DOCSIS channels with a QAM modulators and RF upconverters on the 256 QAM downstream on a 6 or 8 MHz RF output. This EQAM is being adapted for use channel. The data capacity of the DOCSIS in a modular CMTS environment. downstream can also be on the order of 200 Mbps to 1 Gbps, based upon a new The outputs of these devices are often emerging technology referred to in industry as referenced as just a “QAM”, rather than the “bonding” or the “wideband protocol for a full “QAM Modulator and RF Upconverter”. It may be slang, but it has stuck. This paper direction, then extrapolation of the name will use the expression “QAM channel”. DEPI to the upstream results in upstream external PHY interface, or UEPI. UEPI is not The CMTS core contains everything a currently being defined in the CableLabs’ traditional CMTS does, except for the committees, so its definition currently is downstream PHY. Specifically, the CMTS proprietary. core contains the downstream media access control (MAC) and all the initialization and Downstream Radio Frequency Interface operational DOCSIS related software. (DRFI) is intended to capture all the current and future RF requirements for the Locating the MAC inside the CMTS core downstream direction for both integrated has several advantages. First, it permits the DOCSIS CMTS systems, modular DOCSIS maximum reuse of existing DOCSIS software CMTS systems, and VOD EQAM systems. code bases. Second, it provides a local piece This is quite the undertaking! of hardware that can rate shape the aggregate flow of all packets to each QAM so that the One of the goals of DRFI is to be able to output queues in the EQAM will not specify a simultaneous system with up to overflow. 119 digital carriers and the remainder; analog carriers. Another goal is to define a high This diagram currently shows the RF density RF connector as it is getting too section of the DOCSIS upstreams internal to difficult to use the standard F connector on the CMTS core. This has been done because high density front panels. The specification is at this time, the Modular CMTS specifications also looking at new modulation error rate only require the use of external downstream (MER) and out-of-band (OOB) noise QAM channels, mainly because those QAM specifications. channels are leveraged off of existing equipment. External upstream QAM channels DOCSIS Timing Interface (DTI) is an do not exist. However, there is nothing interesting interface. DTI is a point-to-point preventing an implementation of a Modular interface from the Timing Signal Generator to CMTS from using external QAM channels in all MAC and PHY components. DTI has the the upstream. concept of a DTI server and a DTI client. The DTI server is the Timing Signal Generator, The Timing Signal Generator provides a while each MAC and PHY has a DTI client. common frequency of 10.24 MHz and a The DTI client is light weight. The DTI server DOCSIS timestamp to all MACs and PHYs. distributes a 10.24 MHz frequency over unshielded twisted pair (UTP) with a The Downstream External PHY Interface timestamp modulated on it. The DTI returns a (DEPI), is the interface between the CMTS copy of the timestamp. The DTI server can core and the edge QAM. More specifically, it then measure the difference in the transmitted is an IP tunnel between the MAC and PHY in and received timestamps to measure the round a Modular CMTS system which contains both trip delay to each element. It then adjusts the a data path for DOCSIS frames and a control transmitted timestamp to each network path for setting up, maintaining, and tearing element so that they will all have the same down sessions. sense of time. Edge Resource Manager Interface (ERMI) If there was an external QAM burst is an interface that permits integration with a demodulator device for the upstream next generation VoD network. ERMI is used to interface to a Resource Manager. This Cable Modem to Customer Premise resource manager then allocates QAM Equipment Interface (CMCI) is also resources to either VOD or DOCSIS unchanged and is typically 100 Mbps applications. ERMI, however, does not Ethernet. directly manage within the DOCSIS QAM, so it is not a required interface on DOCSIS only DEPI OPERATION systems. DEPI is an IP tunnel that exists between Radio Frequency Switching Control the DOCSI MAC in the CMTS core and the Interface (RFSCI), is intended to manage an DOCSIS PHY that exists in the edge QAM. RF switch which would be used for DEPI’s job is to take either formatted redundancy. This would allow, for example, a DOCSIS frames or MPEG packets and bank of upstream or downstream “working” transport them through a Layer 2 or Layer 3 QAMs to be physically swapped out with a network unharmed. bank of “protect” QAMs in the event there was a failure in the “working” QAMs. This The base protocol that DEPI has chosen is interface is not being defined as part of the the Layer 2 Tunneling Protocol Version 3, or first round of definitions at CableLabs. L2TPv3 for short [2]. L2TPv3 is an Internet

Engineering Task Force (IETF) protocol that Operations Support System Interface is a generic protocol for creating a (OSSI) provides the management interface to “psuedowire”. A psuedowire is when a each system component. One of the Layer 2 protocol is passed transparently over interesting tasks the OSSI system has to a Layer 3 network. Examples of protocols define is which DOCSIS device initializes the supported by L2TPv3 include asynchronous PHY level parameters of the QAMs such as transfer mode (ATM), High-level Data Link modulation. Should the EQAM do this since it Control (HDLC), Ethernet, Frame Relay, and already has to do this for VOD QAMs? Point to Point Protocol (PPP). Should the CMTS core do this so that OSSI and CLI structures for the CMTS can be similar to what they are today? Whatever the Figure 3 on the next page shows the decision, one device should get configured, format of an L2TPv3 packet. There is a bit and the other device should learn these values called the “T” bit in the header of each packet over DEPI. to distinguish between data and control packets. There is also a 32 bit session ID. The The Network Side Interface (NSI) is UDP header is optional with L2TPv3, but is unchanged from some nine years ago. It the included to permit addressing of QAM physical interface the CMTS uses to connect channels with the UDP destination port. to the backbone network. This is typically 100 Mbps or 1 Gbps Ethernet.

L2TPv3 then permits a sub-header to exist The second encapsulation is straight whose definition is specific to the payload multiple: 188 byte MPEG-TS packets are being carried. There is a specific sub-header placed into the L2TPv3 payload with a unique for DOCSIS and MPEG-TS that currently are sub-header which contains a sequence number under definition. so packet drops can be detected. This encapsulation is intended to carry MPEG-TS The control channel allows for signaling based bonding or “traditional DOCSIS” where messages to be sent back and forth between it is not necessary to send MAPs separately. the MAC and PHY. Typical control messages will set up a “control connection” between the One of the technical considerations of the MAC and PHY, and then set up multiple Modular CMTS architecture is its impact on sessions. Each session can have a different the round trip REQ-GNT delay time. This is DiffServ Code Point (DSCP), and/or support a the time from when a CM launches an different encapsulation protocol. uncontended REQ to when it receives a MAP message with the GNT opportunity in it. There are two basic encapsulation techniques under consideration for DOCSIS. To prevent the MAP from being slowed The first is a something called the Packet down by data traffic, the MAP may be sent in Streaming Protocol (PSP) and allows an independent L2TPv3 session that has a DOCSIS frames to be both concatenated to unique DSCP. This DSCP will have a “per increase network performance, and hop behavior (PHB)” that will give MAPs the fragmented, in case the tunneled packets highest priority and lowest latency service. exceed the network MTU size. This encapsulation is intended to carry traditional DOCSIS frames.

EDGE QAM OPERATION The next interface supported is DOCSIS MPT. This is similar to Transparent MPT Figure 4 shows a high level block diagram except that edge QAM must search for of an edge QAM that is capable of handling DOCSIS SYNC messages and correct them either video MPEG traffic or DOCSIS traffic. based upon the edge QAMs internal The expression “MPT”” is an acronym for timestamp which has been derived from DTI. MPEG Transport. This mode is intended for DOCSIS frames

The first interface that is shown is the where the MAP is embedded into the stream VOD transport. VOD SPTS or MPTS streams and network latency is not a concern. This are received with a format of MPT over UDP. mode is seen as a transitionary mode which The video processing functions generally may be more of interest to early include de-jittering, PID remapping, signaling implementations. insertion, and PCR timestamp correction. These functions are not defined in the current The next interface is DOCSIS PSP. Here round of CableLabs specifications. DOCSIS frames and MAPs are received on different sessions. The DOCSIS frames may The next set of interfaces are the DEPI have been “streamed together” by PSP into a interfaces. The first one is called “Transparent uniform byte stream. The PSP reassembly MPT”. This is a simple mode in where the engine removes this overhead and recovers incoming MPT frames are copied directly to the DOCSIS frames. The PSP scheduler then the QAM channel without any interpretation allows MAPs to be placed in order, ahead of or modification. This mode is intended for data and SYNC messages to be inserted. The MPT based bonding algorithms such as output is then fed to a transmission described in [1]. convergence layer which converts the results to a DOCSIS MPT stream.

The last interface is the DTI interface ABOUT THE AUTHOR which provides a common frequency and timestamp. The reference frequency is used to John T. Chapman is currently a synchronize the downstream baud rate for use Distinguished Engineer and the Chief with DOCSIS 2.0 Synchronous Code Division Architect for the Cable Business Unit at Cisco Multiple Access (SCDMA) cable modems. Systems in San Jose, California. As a The timestamp is used for the DOCSIS SYNC founding member of the Cisco Cable BU, correction. John has made significant contributions to Cisco and the cable industry through his SUMMARY pioneering work in DOCSIS and development of key technologies and concepts critical to A new CMTS architecture called the the deployment of IP services over HFC plants. Modular CMTS architecture has been described. This architecture has the DOCSIS Included in these achievements are being MAC and PHY split into two network the primary author of significant portions of elements. This allows each network element the DOCSIS and PacketCable specifications to be optimized both for performance and as well as the originator of DOCSIS Set-top cost. Gateway (DSG) and evolving specifications for DOCSIS Wideband and Modular CMTS This new Modular CMTS will also architectures for the industry’s Next provide the foundation for an entirely new Generation Network Architecture (NGNA) class of CMTS which will have much higher initiative. John has also published a number of data capacities than anything out there today. ground breaking whitepapers on Multimedia These new machines will require multiple Traffic Engineering (MMTE), DSG, QoS, and 1 gigabit or 10 gigabit backhauls rather than high availability and is a respected and the 100 baseT backhauls in use with today’s frequently requested speaker at industry CMTSs. events.

John has 18 patents issued and 27 patents REFERENCES pending in a variety of technologies including telephony, VoIP, wide area networking, and 1. Chapman, John T., “The Wideband broadband access for HFC cable networks. In Protocol for a DOCSIS Network”, his spare time, John enjoys spending time Proceedings of the SCTE Emerging with his wife and two daughters. John is a 6th Technologies Conference, January 2005 Degree Black Belt Master in Tae Kwon Do 2. Townsley, Mark et. al, RFC 3931, “Layer and enjoys white water canoeing and skiing. Two Tunneling Protocol - Version 3 (L2TPv3)”, IETF, February 2005 Previous papers by John may be found at http://www.johntchapman.com

TRANSPORT, CONTENT, AND SERVICE IMPLICATIONS ON VOD NETWORK TOPOLOGY

George Kajos, Vice President of Engineering Conrad Clemson, Sr. VP, Technical Operations Broadbus Technologies

Abstract deployments. Some architectures propose decentralized VOD server deployments; Video on Demand (VOD) is evolving and others propose centralized server growing rapidly. As a result, transport, deployments, finally some compromise with content, and service offerings are changing a hybrid approach. the fundamental economics and operational efficiency of Video on Demand networks. On Initially, content usage data from a small Demand services of a few premium movies VOD installation is examined. Then, this and HBO On Demand are giving way to the paper evaluates the effect of content NFL on Demand, Nickelodeon on Demand, placement on the transport network and the Everything on Demand. Content distribution economics of a complete VOD solution as a networks comprised of a small array of function of centralizing vs. decentralizing catchers directly connected to the back end servers. In the face of dramatically of a small video server farm are evolving into decreasing transport costs, the paper full fledged propagation services and identifies the few scenarios in which edge hierarchical storage models. The streaming caching may be an effective approach to network has evolved from a network with certain VOD applications. The paper also streaming servers directly connected to ASI examines content propagation and ports to a GbE based transport network. replication. Finally, this paper examines how Services are expanding from traditional the evolution of new services, some of which transactional, free, and subscription services may be personalized to a per subscriber basis, to include a new variety of On Demand should dictate the placement of both the service offerings. Everything from reality TV, servers and the content. to advertising, to personal ads is becoming available on line. With each of the elements SERVICES of the On Demand system dissected into its pieces, the paper will put these elements The last half decade has validated Video together in a single cohesive view of optimal on Demand. Hundreds of VOD deployments On Demand network topologies based on the have occurred featuring MOD, SVOD, and evolution of transport, content and service FOD [CED]. A consistent server design type. point had previously been 500 – 1,000 streams and 1,000 – 2,000 hours of content. Server clustering allowed installations to OVERVIEW support 10,000 – 20,000 streams.

The VOD environment is clearly evolving Accordingly, these successes have in multiple simultaneous dimensions. Within unleashed the potential demand for a wide this environment, there are several competing range of new services. Examples of new architectures for appropriate VOD server services being trialed or conceived include: was a one-to-one relationship with the • Music on Demand bandwidth available from hard disks. • Non-linear Live Broadcast Extremely sophisticated striping and • Network PVR scheduling techniques were employed to • Customized content drive up stream counts. Moreover, custom • HD Content and Widescreen format trick files were prepared for fast forward and for all the of above rewind in order to remove the variance from disk access. In general, this meant a trick file It is assumed that Music on Demand will for every fast forward/rewind rate or a single not have a major impact on streaming and fast forward/rewind rate. storage capacity. A hard drive based server also posed HD content is still evolving. The impact limitations on the amount of content which of the current format is a 4X multiplier on could be ingested. Updating hard drives with streaming bandwidth and storage new content reduced streaming bandwidth requirements as MPEG 2 HD content is while disk writes were scheduled. In an being transmitted at 15 Mbps. The demand environment with just MOD and SVOD for HD is steadily increasing. services, content propagation could be scheduled at off hours with little adverse MOD Wide Screen versions appear as just affect. However, as we explore live another content. They are typically broadcasts and real time propagation, the equivalent in size. required inbound content loading bandwidth goes up considerably. Network PVR and Live Broadcast services have the potential to greatly impact This generation of video servers is capacity requirements. Quantifying the designed around two principles: required bandwidth is straightforward. Service offerings and business rules • Independent scaling of streaming determine temporary storage requirements. bandwidth and storage capacity For example, a service offering of 100 SD • Real time content ingest and and 10 HD channels has an ingest turnaround performance requirement in excess of 500 Mbps and a content storage requirement, temporary or Server architecture designers must permanent, of 250 Gbytes/hour. The carefully consider the tradeoffs between the selection of content to retain and the duration following: it is to be made available could vary widely based on the business rules of the offering. • Processor performance • Disk I/O bandwidth SERVER DESIGN • Network bandwidth • Memory bandwidth The continuing advance of technology • Backplane/Interconnect I/O allows this generation of video servers to bandwidth break the dependency of streaming capacity on disk bandwidth. First generation video servers relied on streaming from hard drives. Consequently, the number of streams served SERVER PLACEMENT AND • Transport network – the network NETWORK DESIGN supporting streaming to On Demand clients. Extensive research has been conducted • Propagation network – the network and numerous papers have been written on supporting On Demand and scheduled the topic of video server network topology. propagation to servers. Most authors describe the approaches as • Content Storage Server – a centralized, decentralized, and hybrids of the generalized library and central two. Looking at the next-generation of server repository for the content made capacity and content library sizes, and taking available for On Demand services. into account both cable HFC and IP video, it • Preparation Server – receivers of live is instructive to generalize the following broadcasts which then encoded for components: propagation to video servers to On Demand clients for play. • Video Server Complex – server technology capable of streaming, • On demand clients – the media local content storage, On Demand and decode and display point at the scheduled ingest, and session/stream subscriber. Figures 1 and 2 depict management. centralized and decentralized VOD environments respectively.

Figure 1: Centralized Server Placement

Figure 2: Decentralized Server Placement

The distinctions between these 5. The decentralized server environment environments follow: receives both scheduled and On Demand content over the propagation network. 1. The centralized server environment has complete connectivity to the entire The differences in the two environments population of supported On Demand are chosen to emphasize the trade-off clients over the centralized transport between the cost of transport network network. bandwidth and content replication. 2. The centralized server environment receives only scheduled content and little The network bandwidth/storage trade-off or no On Demand content over the is not the only consideration between propagation network. centralized and decentralized approaches. 3. In the centralized environment, it is Other considerations include: possible to load balance across the entire population of clients. In the fully • Replication of control components at decentralized environment, load decentralized sites. balancing is restricted to the partitioned • Operational costs of additional subset of clients. decentralized sites.

4. The decentralized server environment has In reality, most VOD system designs are connectivity with a partitioned subset of hybrid approaches. For example, even in a the population of supported On Demand centralized environment, it is unlikely that clients over regional transport networks. every server need to have connectivity to every client. Acceptable load balancing is This paper uses the term “On Demand possible with reduced connectivity. In a propagation” to refer to the requirement to decentralized environment, for added move content to a server which has received reliability it would be advantageous to have a purchase request and does not have the more than one server capable of reaching required content. The case arises in server each On Demand client. The system could environments where not all content is located operate in a degraded mode until system on every server. In a centralized server repair completes. environment, a session manager could direct the request to an appropriate server. CONTENT PROPAGATION However, in a decentralized server As discussed above, in first generation environment, clients are partitioned by video servers, content propagation servers. It is unreasonable to replicate all requirements corresponded to the gradual content at every site in a decentralized server refresh of new MOD, SVOD, and FOD environment. Consequently, this paper offerings. These could be loaded onto video defines the case when content must be server complexes with little affect on server transferred to a server to grant a client performance during low usage periods. request. Figure 3 demonstrates how On Demand propagation works. The client This generation video server must be requests content not available on the local designed for two new sources for content server or server complex. A request is made propagation: to a regional propagation server for the required content. A “filler” content is • Live Broadcast transmitted to the client at the start of the • On Demand propagation. upload. The filler could consist of previews or advertising. When the server buffers enough content, play out begins.

1 2 3 4 Content Server Content Server Content Request Propagation

Local content 1 2 3 4 Streaming Server Content Streaming tim e

Client Content Request Figure 3: On Demand Content Fulfillment

Live broadcasts and On Demand alternatives in the WAN, such as Packet propagation, move guaranteed quality of over SONET and ATM, as GbE moves to service from a server memory or disk 10Ge, it appears that 10 GbE will become subsystem problem to the propagation the most cost effective, high bandwidth network. Unlike the dedicated transport solution. network which is most often provisioned for the maximum stream capacity dictated by the Consequently, this paper examines ways HFC QAM capacity or as a percentage to interconnect geographically separated (provisioned take rate) of the total pool of 10Ge pipes. The most straightforward are clients, the design of the propagation network dark fibers and long reach optics to create should include a policy on how to allocate point to point links. However, this is an between scheduled, live broadcast and On inefficient use of the bandwidth available in Demand bandwidth. As discussed earlier, the fiber and would only make sense if only Live Broadcast input bandwidth requirements a single trunk of 10 Gee is required. could range in the 500 Mbps. Figure 4 depicts an example of how the allocation Another approach is wave division policy could vary during a 24-hour cycle. multiplexing (WDM). WDM technology allows data from multiple sources to share a single fiber by transmitting on individual wavelengths. WDM interfaces have been incorporated in switches and multiplexers.

Two types of WDM are in use today: Dense WDM and Coarse WDM. DWDM is ideal for high bandwidth, long haul applications. Current DWDM technology can squeeze over 30 channels in C and L optical bands.

CDWM uses lower cost optics and is

characterized by wide channel spacing over Figure 4: Network Bandwidth Allocation a wide spectrum. CWDM technology can Policy supply 18 channels from 1270 to 1610 nm.

TRANSPORT AND PROPAGATION WDM technology is ideal for NETWORK TECHNOLOGY accommodating 10 Gee streaming pipes to remote QAM or DSLAM locations The streaming network has evolved from a network with streaming servers directly OBSERVED DATA connected to ASI ports to a GbE based transport network. Ethernet price points In this section, VOD and SVOD are have continually fallen due to the ever examined as a starting point for planning for increasing reach of Internet protocols from new services. VOD today is dominated by traditional LANs to geographically SD (standard definition) content and is dispersed WANs. While there are transmitted as MPEG2, 3.8Mbps/stream. It is used for Movies on Demand and across 703 contents. As one would expect on “Subscription on Demand” Services. a Saturday, the maximum number of sessions peaks around 9:30 PM as shown in Figure 6. Figure 5 and 6 depict VOD usage for one The distribution shown in Figure 2 plots day, January 29, 2005, at a relatively small content usage from most used content (117 site. The days’ totals are 5,133 streams plays) to least viewed content (1 play).

5133 streams vs. 703 contents

Content #1 117 streams

Figure 5: January 29, 2005 – Streams versus Time of Day

Figure 6: January 25, 2005 – Streams vs. Time of Day

These observations are typical and fact, the popularity distribution can be fit to reported in a number of other works [5]. In Zipf’s law, which states that the probability of requesting a program m, where m = 1, 2, 3… out of N movies is : C/m where C=(1+1/2+1/3+…+1/N)

Figure 7: Simultaneous Content Viewing at Peak Usage

Figure 8: January 29, 2005 – Most Popular 3 Contents

Figures 7 and 8 examine the relationship contents are being viewed by two or more set between simultaneous sessions and content. top boxes. Figure 8, depicts that at the peak In Figure 7, it can be seen that 58 of the 133 of viewing, the top three contents total 57 of the 322 sessions. As video servers are scaled, edge to conserve transport costs. While this these relationships will be used to assist with approach has academic appeal, it does not the tradeoff between memory, network, and hold up to real world scrutiny. The problem I/O bandwidth. with this approach is that it assumes a uniform distribution of concurrency in each Finally, additionally data obtained from service group. The data suggests that each discussions with a number of MSO’s with service group has its own peak concurrency. mature VOD deployments suggests that These server designers would propose to actual VOD usage data varies widely from either provision the entire system to the service group to service group. While it is not average concurrency or, worse, the peak possible at this time to publicly share concurrency. Provisioning to the average specifics of these results, it is generally concurrency will result in denial of service to understood in the industry, that within a the peak service groups and over provision system, peak usage may vary from 2% to the low concurrency groups. Provisioning to 14% on a service group by service group the peak concurrency will result in massively basis. over provisioning the entire system.

This data proposes some interesting When transport costs are equal to or less conclusions. expensive than the streaming server costs, as they are today, the only logical way to First, real world data validates a Zipf provision a VOD system is to centralize the distribution model across multiple architecture. This provides the operator with deployments. The Zipf curves favor caching tremendous economies of scale in the architectures, in general, and specifically streaming subsystems. At the same time it making caching architectures more effective allows the operator to provision across all as the stream count grows. This is the case, service groups without stranding streams at because in a Zipf curve, the tail is relatively the edge of the network. constant, while the peak of the curve grows dramatically with stream count. As NEXT GENERATION VOD centralization takes place, VOD servers that ARCHITECTURE accommodate extremely high concurrency, such as caching servers, will serve more In this section, hypothetical VOD system streams at a lower cost. is created for the purpose of further exploring centralized approaches and decentralized Additionally, the tail stays relatively approaches. Assume that an environment is constant. In hybrid architectures, it is provisioned for a take rate of 300,000 active generally assumed that less popular titles are clients. There are two server capacities streamed from the core to conserve the available – 3,000 and 15,000 streams. In this replication of storage. This data supports that exercise, the paper examines the number of model, but it also points out that a library servers and the equivalent number of gigabit server which has a caching capability, can Ethernet links required at each extreme. One perform both tasks. additional consideration is the possibility that a decentralized server will need double the Some server designers would argue that it ingest bandwidth of a centralized server to makes sense to cache the popular titles at the accommodate On Demand propagation.

Table 1: Servers and Transport Links

300,000 Streams Servers GbE 10GbE Propagation streams per Transport Transport GbEs server Links Equivalents Centralized 15,000 20 60 6 20 Decentralized 3,000 100 12 100/(200)

Some general observations can be made: • Higher cost transport in centralized model • More servers, storage, control • Greater potential to share storage in systems, in decentralized model centralized model • Ability to collapse ten 1GbEs into one 10GbEs in the centralized model Figure 9 depicts an architecture which • Potential additional load on the collapses the transport network connectivity. propagation network in distributed model

Broadcast Preparation QAMsQAMs Streaming Server Server Server DSLAMDSLAM L2 L2/L3

10 Ge DWDM/CWDM L2 Propagation L2 Network Network

Content Storage L2 Server

Streaming QAMs Server

Figure 9: Next-Generation Collapsed VOD Network Architecture

The transport network is represented as a Conrad Clemson WDM optical network connected through Sr. VP, Technical Operations layer 2 switches. Because 10Ge pipes are Broadbus Technologies, Inc. steered to specific Lambdas and partition the 80 Central Street client space, the architecture is not a fully Boxborough, MA 01719 centralized environment. However, most of [email protected] the advantages of the centralized Office: 978.264.7905 environment are realized. REFERENCES SUMMARY 1. Skarica, Christopher, and Joe During the previous half decade, many Selvage, “Environmentally Hardened VOD installations were monolithic and self optical CWDM Solution for MSO contained for ingest and streaming. Advances Access Network Capacity in server and transport technology allow new Expansion,” Cable-Tec Expo, June services to be considered for VOD 15-18, 2004. deployments. In this paper, data regarding 2. Kajos, George, “VOD Storage content popularity during a single day and at Architecture for Hierarchical peak load was presented from one VOD Network Content Libraries,” Cable- installation. The next-generation VOD Tec Expo, June 15-18, 2004. architecture presented in the paper is well 3. Jancola, Mark, “Considerations in suited to meet the scale to the requirements Scaling Video-on-Demand Systems,” dictated by new services. Cable-Tec Expo, June 15-18, 2004. 4. “North American Cable ACKNOWLEDGEMENT Deployments,” CED Magazine, December 2004. The authors wish to acknowledge Michael 5. Vassiliakis, Constantine, Michael Kahn for extracting and processing the Peterakis, Peter Triantafillou, “ Video January 29, 2005 data. Placement and Configuration of Distributed Video Servers on Cable AUTHOR’S CONTACT INFORMATION TV Networks,” Multimedia Systems, 8:92-104, 2000. George Kajos VP, Engineering Broadbus Technologies, Inc. 80 Central Street Boxborough, MA 01719 [email protected] Office: 978.264.7906 UNIFIED DATA AND VIDEO CMTS: ONE SYSTEM FOR ALL SERVICES

Scott Cummings and Victor T. Hou Broadcom Corporation

Abstract be quickly examined. The term HE is used a bit loosely here. There are distribution hubs This paper opens with a quick history of and other types of nodal locations that may the traditional Head End (HE) and current provide the functions described in this paper CMTS platforms in the HE. The “loosely as a HE. coupled, tightly bolted” architecture of the traditional HE is exposed. The issues in Cable Modem Termination System (CMTS) adding new services and technologies will become apparent. The current DOCSIS 1.x, and 2.0 CMTS platforms predominantly deliver data service Then an overview of several current and for cable modems. These systems operate new technologies entering the next with a downstream modulation type of either generations of HEs is listed. Each technology 64-QAM or 256-QAM in the downstream. is described. Then the Modular CMTS (M- The downstream channels support either 6 CMTS) architecture is applied to the Next MHz (North American plants) or 8 MHz Generation Network Architecture (NGNA) for (European plants). The CMTS also provides the HEs. The traditional CMTS is subdivided some networking services for the cable into three modules: CMTS core, Edge QAM, modems (DHCP, ToD, Gateway). Some and Upstream Receiver. A new component to CMTSs incorporate these servers, others do the HE is defined: a System Resource not. DOCSIS 1.1 introduced Quality of Manager (SRM). Service features to the CMTS such as rate shaping and per user rate limiting. The incoming technologies are applied to The CMTS also provides upstream data the SRM and M-CMTS. As these technologies services. DOCSIS 1.1 is limited to a are incrementally added to the SRM and M- maximum bandwidth for a single channel of CMTS, the changes to support them are 10 Mbps. To provide symmetric bandwidth simple. Adding servers, SRM interfaces, and with the downstream, the DOCSIS 1.1 CMTS possibly CMTS cores and the system is ready platform are designed for a 1:N downstream for the new technology or service. The to upstream ratio, where N would be 4, 6, or combination of the SRM and M- 8. DOCSIS 2.0 introduces many upstream CMTS convert this HE architecture into the modulation types which now allows the unified CMTS, a system for all services. CMTS to offer QPSK, 8-QAM, 16-QAM, 32- QAM, 64-QAM, or 128-QAM (SCDMA Traditional Head End (HE) only). However, the downstream to upstream ratio has not changed. The traditional HE is composed of many devices that offer several services. Video, The current CMTS platforms are Data, and Voice services are all supported by composed of a rack of blades. Typically the the HE. Each of these services and the CMTS contains a network blade, a processor equipment used to deliver these services will blade and a physical layer blade. Additional blades can be added to the rack to meet the combination of both analog and digital signals are transmitted on the same HFC plant. needs of the node or head end supported. The physical layer blade maintains the 1:N ratio. These are standalone systems that merely The second type of video is narrowcast share the Hybrid Fiber Coaxial (HFC) cable video. The narrowcast video is composed of plant. Video on Demand (VoD) and other selectable video services. These services are not The current CMTS platforms also delivers streaming video all the time. A given video Voice over Internet Protocol (VoIP) service. stream may be streamed on multiple RF The VoIP services are bidirectional services channels. Billing systems for the individual that have restrict latency and jitter purchase of video streams are also required requirements relative to the general data for these services. The narrowcast video services. Due to these requirements, the VoIP services often also require a return path for services are often segregated from the data ordering and billing confirmation.

CMTS Ethernet Digital Rack(s) IP gateway Video Racks VOD rack (data/voice) Analog Digital Video with with video with Video Rack Rack(s) incoming servers, DHCP, ToD, with with up/ demods, billing Call agents, up/down down video systems, servers, convertors convertors servers, modulators mods, Combining QAM mods, and upconv external Splitting and upconv. upconv Diplexed Coax network

Traditional HE system level diagram

services. This segregation may be as simple Narrowcast video may also be used to send as dedicated channels for VoIP, or as rigorous programs only when they are viewed. Rather as a dedicated CMTS for the VoIP services. than broadcast all the channels available to the subscriber, a future model for narrowcast video would send only the channel requested The CMTS systems make up only a small in either multicast or unicast streams. part of the Traditional HE.

Ignoring the equipment just to receive Video Distribution video (multiple satellite dishes, local

antennas, high speed fiber connections, The majority of equipment in a traditional downconverters), the HE is composed of HE is for video distribution. There are two multiple QAMs, upconverters and MPEG distinct types of video distribution in a HE. multiplexing stations. The MPEG multiplexing stations may be simple The first is broadcast video. The broadcast multiplexing units or very sophisticated and video may be delivered with traditional NTSC expensive units that are capable of analog signals, or with digital video MPEG transcoding and transrating MPEG streams in transmissions. In most HE today, a real time.

The distribution of broadcast video and with the certification process to guarantee narrowcast video both use the same type of compatibility across multiple vendor devices. equipment. The billing, switching, and return The certification and qualification process path requirements of the narrowcast video also delivers a very robust and reliable generally create implementations for these system. The Consumer Premise Equipment systems that are stand alone systems relative (CPE) connected to the HFC plant can now all to the broadcast video systems. be controlled using this DSG tunnel.

The CMTS systems providing data and VoIP services also tend to be independent The DSG technology is available today. systems from the video distribution systems. The DSG effort has been active at CableLabs This creates a collage of equipment in the HE. and the specifications are complete. This also creates operational issues for maintaining a HE. Each system requires IP video individual expertise to be maintained and operated. As new technologies and/or new solutions for existing services become The concept of delivering video over IP on available, these systems tend to contain an HFC plant has been around for quite some requirements that lead to stand-alone time. IP video dominates video delivery to implementations as well. Personal Computers (PCs). QoS, buffering issues, and overhead issues have all hampered New Technologies the acceptance of IP video. IP version 6 (IPv6) has addressed some of the QoS issues. There are several new technologies that are Memory and disk space technology continue currently available or are soon to be available to follow Moore’s law. The cost for a bit of to the next generation of HEs. These RAM or disk space continues to drop. This technologies include DOCSIS Set-top has greatly reduced the buffering issues with Gateway (DSG), Internet Protocol (IP) video, IP video, and will continually reduce Channel Bonding, the Personal Video buffering issues well into the future. The Recorder (PVR), MPEG-4, part 10 (AVC) additional overhead for the IP headers can be digital video compression and 1024-QAM addressed with Payload Header Suppression downstream modulation. Each of these (PHS) in DOCSIS CMTS platforms. innovations has targeted applications and benefits. The power of IP video comes from a couple of sources: economies of scale, and the DSG robust feature set of the IP protocol. The IP technologies are ubiquitous. These solutions The DSG technology provides a DOCSIS are extremely cost effective from an tunnel to any device connected to the HFC equipment standpoint. IP video also plant. This tunnel is a 1 way (downstream automatically provides several avenues for only). However if the CPE device contains advanced video services. These services some form of an embedded CM (eCM), a 2 include stopping, pausing, forwarding, way communication path can be implemented. rewinding, book marking, and other video (both downstream and upstream). This opens control features. The IP video can be served a development path for the HE to control to a wide range of devices. PCs, STBs, and devices other than Cable Modems (CMs) with even Hand Held Devices (HHDs) are all DOCSIS. DOCSIS has provided the MSO capable of processing IP video.

PVR The PVR has the opportunity to take a The PVR has been available to the similar migration from a dedicated video consumer for over a year now. The original device to at general purpose storage devices PVRs provided storage to hold recorded video that can play and record video, audio, still programming and a simple software Graphical photos, and even display web pages. The User Interface (GUI) to help program and PVR actually could be the link between the manage stored recordings. The early PVRs STB and the PC. The STB has had a had limited storage and therefore did not offer processor, memory, an Operating System many features beyond a mechanism to record (OS), and a GUI for several years. Many of and store video. the later models of STB have Ethernet and/or USB ports on them. Keyboard style remote This is similar to the MP3 player. The control units exist for both STB and high end MP3 player started as a simple device to store TVs. The lone missing piece of a PC in an and record audio. Over time, the MP3 player STB is the disk drive. The convergence of the has expanded to be a general purpose storage Web and Video, the STB and the PC, could device that has the capability of playing and all be in the PVR. recording audio.

Bonding Combining Technology 120 Mbps Receive, Reorder Segment, Recreate original packet

30 30 30 30

Mbps Mbps Mbps Mbps

6 MHz 6 MHz 6 MHz 6 MHz

30 30 30 30 Mbps Mbps Mbps Mbps

Bonding Dispersion Technology 120 Mbps Segment, Sequence, and Transmit

Diagram of channel bonding a 120 Mbps load across 4 downstream channels

Channel Bonding simple. Channel Bonding bonds N number of existing channels into 1 larger virtual channel. Channel Bonding is a new technology that This provides several features. First, the is part of the discussions for DOCSIS 3.0. maximum throughput to a single CPE will be The concept of channel bonding is fairly increased by a factor on N. An MSO could offer a high speed service of well over the the channel bonding system can use the other 38.8 Mbps that a single 256-QAM channels to help get critical packets downstream channel can provide today. downstream with minimal jitter and latency effects. Another feature of Channel Bonding is the statistical multiplexing gain of having more 1024-QAM modulation users share a wider channel. Statistical multiplexing benefits occur when a channel Many new technologies have been applied gets oversubscribed. Service on that channel to the HFC plant. No advancement in has to be reduced or stopped for some users. DOCSIS from its inception has provided If another channel was undersubscribed, the improved bandwidth in the downstream. PHS excess load could be taken by the second can be applied in the downstream to save channel. Today, instantaneous load balancing some bandwidth. The only improvements in is not available in CTMS platforms. If both bandwidth in the downstream over the past 5+ channels could be virtually combined, the new years has been the expansion of the HFC bonded channel would not be oversubscribed plants frequency spectrum. He original HFC and no reduction or loss of service would be plants started with an upper frequency limit of observed by the users. 550 MHz. This has expanded to 650, 750, 850, and now there are discussions of opening The granularity of the Channel Bonding up to more than 1 GHz. The frequency solution may also provide other benefits. For expansion has hardly been a new technology. Channel Bonding to work, the CPE must have Replacing cable, connectors and amplifiers the ability to simultaneously receive multiple with cleaner, equipment specified up to the channels. With more channels to select where higher frequencies is not an architectural and when to place downstream packets, technology improvement. latency and jitter could be better controlled. This is similar to the statistical multiplexing A more efficient modulation scheme can gain with the expectation that the system has be introduced to the HFC plant. Currently, not been loaded to the point of over the highest modulation scheme available to an subscription. As the system is approaching HFC plant is 256-QAM. This provides 8 bits the maximum subscription rate, this level of per symbol. 1024-QAM provides 10 bits per the loading or congestion begins to create an symbol. Upgrading the modulators and increase in packets that request the demodulators to 1024-QAM would improve transmission at the same transmission time. the raw bandwidth by 25%. Clearly in a single channel, two packets cannot be transmitted at the same time. One Beyond the straight bandwidth gain of the packet will be transmitted, and the other modulation order, a statistical multiplexing packet will either be latent or get advanced. gain will also be accrued. To get the 1024- In this case, jitter has been introduced. As the QAM modulation to operate, an HFC plant system approaches the maximum subscription with no margin will have to supply 6 dB more rate, jitter and latency begin to increase. Signal to Noise Ratio (SNR). This begs the question: Is it easier to extend the frequency A Channel Bonding solution that has in the HFC plant by 25% or cleanup up the MPEG layer granularity and the freedom to HFC plant for 1024-QAM? Actually, it select the channel at transmission time can would be synergistic to do both and gain (1.25 send multiple packets at the same time. If a * 1.25 = 1.5625) 56.25 % more bandwidth. channel is at the maximum subscription rate, MPEG 4 Part 10 (AVC) video compression screen resolution would use all of the streams. The sum of the three streams would likely be The MPEG-4 Part 10, advanced video larger than the size of a single stream targeted compression (AVC) standard provides for an HDTV application. However, the sum superior compression versus MPEG-2 with of the three streams is likely to be smaller the same resulting video quality. The than the sum of the three streams required to improved suppression will allow more AVC support video across the range of three streams per RF channel versus MPEG-2. devices in this example. AVC, also known as MPEG-4 Part 10, MPEG-4 AVC or ITU-T H.264, is an The Multiple Description feature of AVC extension to MPEG-4. AVC offers many new is still under academic exploration. The compression techniques to provide equivalent technology of this technique is not available quality to MPEG-2 at half or less than half the today. bandwidth. The most effective tool for better compression is the new mechanisms provided Digital TV and HDTV for advanced prediction techniques. Clearly, for digital video, both MPEG Transport The large screen digital TV phenomenon Streams (TS) and IP video, the additional is here. High Definition Television (HDTV) compression saves bandwidth. is the consumer buzzword attached to these monster TVs. HDTV signal transportation AVC offers another bandwidth saving will load the existing HFC plants. technique. The compression of video can be Simultaneous, Standard Definition Television coded with a Multiple Description coding (SDTV) and HDTV transmission will burden technique. Rather than code the video into a the HFC plants. single stream, the video can be coded into multiple streams. Each stream is then sent to In all of the excitement of these new the decompression device. The bigger and better TVs, the catalyst to this decompression device can then use all the technology has been overlooked. The Federal threads to create the resulting video. Communications Commission (FCC) has However, the decompression device may not mandated by 2007 that all over-the-air use all of the streams to decompress the video. broadcast TV transmissions be digital. This The resulting video will have lesser quality requires the TV manufacturers to be relative to the video using all the streams. delivering digital ready TVs well before 2007. The Multiple Description technique could be The one-way communications specification used to compress the video at different pixel has already been agreed upon, and the two- resolutions. The first stream would produce way communications specification to these video for a device with 180 lines of video digital TVs is therefore implied to arrive by (PDA, HHD, …). The second stream could 2007. then be decompressed to generate video with 360 lines of video. The next stream could The digital TV mandate provides some then generate a resulting stream of 720 lines relief and some consternation for the MSO. of video. An opportunity to migrate to an all digital Each stream then is sent downstream. A HFC plant is clearly opened with this device with limited screen resolution would mandate. This converts the bandwidth decode the only the first stream. A device inefficient analog channels into bandwidth with moderate screen resolution would use the efficient digital channels. This bandwidth first two streams, and a device with HDTV relief is short lived. The digital TVs will be HDTV capable and the consumer will be To efficiently create access to every CPE demanding more HDTV programming. An device, the CMTS must have access to the HDTV video stream ranges from 4 to 5 times entire HFC plant. The traditional CMTS larger than an SDTV video stream. contained QAM modulators and upconvertors and receivers embedded into the CMTS in The two-way digital TVs will have the hardware driven ratios. For the unified opportunity to become more than just video service providing CMTS, embedding an entire playing devices. The advanced digital TV HFC plant worth of QAMs, upconvertors, and may offer program guides, web page receivers would not be cost efficient. Two capability, or text streaming services. These systems sharing the same downstream channel are low bandwidth services. The digital TV must be managed. Routing all traffic through could evolve into a device with specialized a CMTS regardless of the need for the CMTS gaming electronics that allow internet gaming. in the downstream path is not cost efficient. As the digital TV gets more complex, the bandwidth requirements of the HFC plant will Ethernet DEPI continue to rise. Network Edge QAM Fabric Next Generation Network Architectures Ethernet/ DTI To begin to deal with all of the new IP technologies that either are destined to be implemented in a HE or simply loading the MAC Upstream requirements for the HE, the architecture of Core Receiver the HE itself must be reassessed. As noted TBD earlier in this paper, the HE has grown into a collection of systems sharing the HFC plant. Diagram of M-CMTS To reduce both operating expenses and the expense of turning over equipment in the HE, To deal with the cost and manageability of new architectures have been proposed. the downstream of the HFC plant the modular CMTS is required. The modular CMTS takes The Unified CMTS via the modular CMTS all of the functions of the traditional CMTS and subdivides these functions into three The beginning of this new architecture is modules: CMTS core, Edge QAM, and being formed with the CableLabs Modular upstream receiver. The CMTS Core provides CMTS (M-CMTS) specification. This all of the MAC functions required by specification is vital in the development of DOCSIS. Classification, rate shaping, future CMTSs. The intent of future CMTS security, header suppression, routing, architectures is to allow it to become a central bridging, and other packet level functions. entity in more than just data. The next The Edge QAM performs all of the generation CMTS must be able to control, downstream PHY requirements of DOCSIS. maintain, and organize all of the streams The Edge QAM may also be accessed by transmitted and received over the HFC plant, other streaming devices other than a CMTS not just voice and data. The unification of core. The upstream receiver performs the service control is the target of the next upstream PHY requirements for DOCSIS. generation CMTS.

individual QAM or QAMs. The Edge QAM The Edge QAM must have low latency Queues for video transport. The edge QAM must also have The key to this architecture is the Edge prioritized queues for data and management QAM. It is arguable whether the Edge QAM message transport. The QoS capabilities is part of the CMTS or really should be provided by DOCSIS must not be hindered by considered the stream-to-RF bridge to the the Edge QAM. The M-CMTS specification HFC plant. Assuming an all digital HFC provides two separate constructs for plant, the plant may have as many as 160+ transmission. QAMs on that plant. These QAMs may be transferring video, user data, voice, The first construct is an MPEG Transport application information, and/or plant (MPT) mode. This mode is expecting MPEG management information. All of this traffic elementary streams wrapped in a UDP does not have to flow through a traditional datagram. Some video processing to de-jitter CMTS. However, the HFC plant may have the incoming video and correct the PCR is CPE devices on any QAM that are in need of required for these streams. This processing is DOCSIS support. The CMTS must have required by the Edge QAM. The interface access to the Edge QAMs. also provides a transparent mode that does not get processed by the video processor or The Edge QAM must be accessible by DOCSIS timestamp function. The interface multiple devices that are providing a wide also provides an MPT operation for DOCSIS array of services over the QAM channel. The packets. The DOCSIS MPT operation would ideal Edge QAM solution would be to have an have some resyncronizing circuitry to Ethernet port for receiving digital traffic and maintain timing. Behind all the MPT an RF connector to transmit the traffic onto to processing is a MPEG multiplexer to merge the HFC plant. An Edge QAM device may the streams onto the QAM channel. support a single QAM or multiple QAMs. The second construct is a Packet Stream The Edge QAM must be able to place the Protocol (PSP) mode. This mode preserves QAM signals on the appropriate frequencies, the DOCSIS QoS features. DOCSIS MAC and there require some upconverting management message are typically prioritized capabilities. ahead of user data on the downstream channel. To preserve this prioritization The Edge QAM device is really a multiple queues that can be prioritized must sophisticated Ethernet Network Interface Card exist in the Edge QAM. The PSP mode of (NIC) for the HFC plant. This should allow operation also differs from the MPT mode in the vendor community to increase the density that the PSP flows are terminated at the Edge of QAMs per RF connector, and reduce the QAM where the MPT flows are transparent cost per QAM. The MSO now can purchase and not terminated. The combination of the number of Edge QAM blades required to modes provide the Edge QAM the versatility provide the necessary number of channels. If it needs for different applications. the plant extends the frequency range, then more Edge QAMs can be plugged in at that The CMTS Core time. The CMTS Core provides all the CMTS The key to the Edge QAM versatility is core functions in the traditional CMTS. The the interface. The Edge QAM must provide CMTS core also is burdened with the channel some MPEG multiplexing capabilities. This bonding process. The M-CMTS has placed will allow multiple devices to share an the channel bonding in the CMTS core versus in the Edge QAM or a combination of both. QAM. The current Edge QAM definition The CMTS core will have an Ethernet does not seem to take advantage of the interface to the HE network and another capability of multiple Edge QAMs on a single Ethernet interface to the Edge QAM network. blade. Concepts like channel bonding require the CMTS core to packet process the bonded The CMTS core will bridge or route stream and send the dispersed streams to packets onto the HFC plant (via an Edge individual Edge QAM devices. This prevents QAM). The CMTS core will classify, rate channel bonding, at the Edge QAM, from shape, police, bond, encrypt, and header dynamically selecting the appropriate channel suppress packets. The intent of the M-CMTS to send packet segments. architecture is to have CMTS core implementations that are scalable. As more The Edge QAM device may actually DOCSIS processing is require for a given HE, support multiple QAMs per Ethernet port. more CMTS cores could be added to the HE. The system would be far simpler if the CMTS core managed and routed packets, and the The Upstream Receiver Edge QAM prepared them for transmission. A channel bonding mechanism in PSP mode The definition of the interface between the would allow the Edge QAM to pull packets CMTS core and the upstream receiver is will from a multi-QAM queue and put the channel be considered in the future. The downstream bonded segments onto the bonded QAMs. division was fairly simple. Leave DOCSIS in This would allow the Edge QAM to optimally the MAC, so the Edge QAM was simple and balance the QAMs versus the CMTS core could be used by non DOCSIS entities. making some crude estimation of where to Applying the same philosophy to the send the channel bonded segments. upstream, the interface would lend itself to be some sort of FEC block wrapped in an Putting the M-CMTS pieces together Ethernet packet. Then the CMTS core could rebuild the packet. However, reusable The modularization of the M-CMTS upstream receivers with non DOCSIS provides architectural advantages. Another upstream devices may not be preferred. hidden advantage to this architecture is the connectivity of the parts. The CMTS Core, The intent of modularizing the CMTS is to Edge QAMs, and Upstream receivers are all allow it to have access to the entire plant and connected together with an Ethernet network become the standard control mechanism fabric. Just as the Edge QAMs can be scaled inside the HFC plant. With this in mind, it and shared with other services in the HE, this may be preferred to have the upstream network fabric can as well. By leveraging the receiver return Ethernet encapsulated packets commodity priced technology of the Ethernet with DOCSIS information wrapped around network fabric, the MSO gets an inexpensive the completed DOCSIS packet. This would solution with excellent versatility. The allow the CMTS core to be a management technological and business success of the tool versus a packet processor. Ethernet network fabric clearly speaks for itself. The concept of having the Edge QAM provide DOCSIS packet processing has been System Resource Manager dismissed for the current revision of the M- CMTS specification. However, there is merit The M-CMTS is perfectly tuned to to revisiting a different definition of the Edge leverage shared Edge QAMs. In the traditional CMTS, the CMTS managed the access for DSG messages to every QAM on load on the CTMS QAM or QAMs. Now, the the HFC plant, that HE would require a M-CMTS cannot really manage all of the traditional CMTS for every QAM. The M- Edge QAMs. The load on the Edge QAMs CMTS has several techniques to handle this may not be known by the M-CMTS. A new issue without a traditional CMTS per QAM. device must be put into place to manage the loading of the Edge QAMs. This device will If the DSG CPE device is a one-way be called a System Resource Manager (SRM). device, then the M-CMTS is required to replicate all DSG messages destined for that The SRM will manage the video servers, device on all the QAMs. This requires a audio servers, IP server, or any server or single CMTS core receiving and forwarding device streaming content over the Edge the DSG message or messages. The messages QAMs. The SRM must have the capability to get tunneled through the CMTS core and on dynamically start and stop services, and to the existing Edge QAMS, with the message dynamically route services to the Edge going to each Edge QAM. The load on the QAMs. The SRM may also need to monitor CMTS core would be minimal. congested links in the Ethernet network fabric to prevent over subscription of that fabric. If the DSG CPE device is a two-way Again the SRM will have to perform the rate device that allowed the SRM and M-CMTS to shaping, load balancing and manage the track the QAM the device is receiving, then routing of the streams in the same manner that the DSG server could send the message to the the Traditional CMTS provides these appropriate QAM. A point to point message functions for data traffic over DOCSIS. sent through a single CMTS core and routed to the proper Edge QAM is highly efficient. New Technologies and the M-CMTS PVR The intent of modularizing the CMTS was to create a HE architecture that would A simple example of leveraging the PVR leverage all of the current and new with DSG over the M-CMTS would be to technologies seamlessly. To demonstrate the download program guides to CPE devices. value of this M-CMTS architecture, the afore This could be as simple as the show listings mentioned technologies will be applied to the per channel for video, or music listings for SRM and M-CMTS architecture. As each audio devices. technology is added, the SRM and M-CMTS will leverage the combination of the Using the techniques listed with DSG this technologies. The result will be that the SRM would be extremely simple. This would allow and M-CMTS will have unified all of the operational content to be stored on the PVR. services and technologies under one This content could be updated at the MSO architecture. convenience.

DSG Channel Bonding

In an ever competitive world, the cost Channel Bonding provide extremely high reduction of the CPE device is inevitable. speed service to a single device. Leveraging This will create CPE devices on the HFC DSG, PVR and the M-CMTS, channel plant with only one tuner. This tuner may be bonding will be used to download a movie as tuned to any QAM. For a HE to provide an impulse purchase.

A customer orders a movie from a service only real difference now is that the DSG provided on the CPE device. The application tunnel is used to communicate multiple and the data required for the application can applications that in turn will make requests to be downloaded and stored using DSG and the System Resource Manager (SRM). The PVR. Now leveraging the DSG tunnel to SRM then configures the multiple video make the transaction, the movie could be servers to stream the video to the appropriate downloaded in around 3 minutes bonding 10 groups of bonded channels for each service QAMs. The downloaded movie would be required. The number of MPEG-2 and AVC stored on the PVR on the CPE device. The servers can dynamically change over time. trailer for the movie could be downloaded in The HE architecture and the CMTS tunneling seconds and playing while the movie itself commands to the CPE devices do not need to downloaded. The customer would have the change. This leads to a cost efficient option of watching the movie right after the migration path from MPEG-2 to AVC in the trailer or viewing the movie at a later date. HE. The same strategy will apply to MPEG- N in the future. The movie could be stored in the HE as an MPEG stream on a MPEG server. The DSG A second example of HDTV, AVC and tunnel is used to inform the movie ordering the M-CMTS will mix HDTV and SDTV application of the address of the appropriate broadcast programming using AVC over the servers to complete the movie ordering HFC. The assumption will be that the transaction. Once the transaction has been program that is being streamed is a broadcast completed, the SRM can get the appropriate program that can be viewed by several types MPEG server to stream video to the of CPE devices. The Multiple Description appropriate Edge QAMs. The SRM can then feature of AVC can generate streams for prevent any further activity from HHD, SDTV STB and HDTV digital TVs. oversubscribing any given Edge QAM. This The SRM then directs the streams to the example again takes very little processing appropriate QAMs. If the CPE devices from a DOCSIS CMTS core and simply requiring more than 1 stream are channel leverages the network connectivity in the M- bonding devices, then the streams can be CMTS HE to get the video to the appropriate dispersed across multiple QAMs. The HHD Edge QAM. devices that require only a single stream could be instructed to find that stream on the 1024-QAM appropriate QAM. The SDTV STB and HDTV digital TVs can be instructed to pull There is no need to glorify this, just take off the streams of interest off of the the last example and go 10/8 times faster. appropriate bonded QAMs. This is using Now instead of 10 channels in the prior bonding as a mechanism for picking up bonding example, only 8 are required. multiple streams as opposed to picking up a single stream striped across multiple channels. HDTV and AVC This system is using AVC for its compression and Multiple Description streaming benefits. The example now grows into having The channel bonding can now be used as a multiple CPE devices on the HFC plant. load balancing tool rather than a burst Some devices require MPEG-2 video, some performance tool. The DSG tunnel is used to devices can receive MPEG-2 or AVC, and direct the CPE devices to the desired streams. some devices are AVC only devices. The All of this is accomplished with the SRM and M-CMTS architecture.

IP Video up, then the system replaces the voice traffic with gaming servers and gaming traffic. To make a long story short, integrating IP Again, add the server for the service that is video takes nothing more than replacing required and the problem is solved. MPEG video servers with IP video servers and providing the appropriate number of Boring, scalable solutions may not make CMTS cores to support the IP Video through for a dramatic ending of a technology paper, DOCSIS requirements. IP video requires but they would be a dream to the current cable more buffering, but the PVR or the never operator. Dynamically changing services are ending improvements in memory densities an operator nightmare. New proprietary can both be implemented to deal with any IP equipment means lengthy bring up time, video buffering issues. expensive installing and maintenance costs, and lengthy Return on Investments (ROI). The Big Ending Installing a server for a service that is in demand today is both time and cost efficient. The big ending has already gotten to be a The ROI could be measured in weeks not bit boring. As the examples continued to pile years. Somehow the boring M-CMTS has on new technologies and services, the system become the cable operators’ Holy Grail. quite frankly easily handled them. The key to providing more services became a matter of Wow, all of this supplied by an M-CMTS having the source servers. The control and architecture. When the M-CMTS architecture distribution mechanisms did not change. is viewed from the entire HE, it becomes the VoIP could be added to the mix. The boring unifying piece to the multiple service puzzle. solution is to add the required CMTS cores This architecture allows the unification of and upstream receivers to handle the load. If services through the M-CMTS. Adding a you already have idle CMTS cores or SRM to an M-CMTS puts the finishing upstream receivers, then just use those. If touches on the Unified CMTS, a system for VoIP declines and an IP gaming service picks all services.

Ethernet MAC Cores Network SRM, DSG Fabric servers (Switches, SNMP Legacy Legacy Edge QAM Routers) services Racks Analog Digital Video Video Rack Rack(s) Upstream with with up/ Server Racks Combining Receiver up/down down Audio Splitting Racks convertors convertors Video (PPV, VOD, IP) Diplexed Gaming, Billing, Coax Web Hosting, Network Future Killer Application

Diagram of the Unified CMTS, the system for all services USING NEW ADI 2.0 STRUCTURES TO EVOLVE VIDEO-ON-DEMAND MANAGEMENT & SERVICES

Yasser F. Syed, Ph.D. Cable Television Laboratories

Abstract between cable services from other media service types. The new ADI 2.0 specifications allows for the separate handling and combining of the Bandwidth used for Pay-per-View (PPV) offer & title metadata, content, and its services in time will be recaptured by VoD metadata. This becomes a valuable set of services while still retaining the value offered managing tools as catalogues get larger, new by PPV – current top twenty movies. But the VOD marketing techniques are envisioned, model for VOD service can extend beyond and content becomes more long lasting and this to carry TV shows, news, music videos, less traditional. This paper will first describe and other events found on broadcast while what are the new ADI 2.0 structures and including possible commercial support and compare this against older ADI 1.1 channel identification. This can be further structures. It will also describe how VOD extended to offer content with a continuity services are changing and what does it imply not offered on broadcast because of niche for new management and service needs and demands and limited bandwidth: viewing how this will be easier to handle using the past show episodes, taking educational new structure & backend distribution. Then it classes, seeing telescoping ads, touring local will talk about possible ways to transition homes for sale, etc. The initial success of from a 1.1 ADI structure to a 2.0 structure. VoD bring challenges of its own that need to Finally the paper will describe how the new handle these new possibilities while scaling ADI 2.0 structures can help meet these new to increased customer usage demands. types of demands on cable systems. The Cablelabs 1.1 ADI/VOD metadata documents is the current defacto INTRODUCTION specification used for delivering content with metadata to cable systems via satellite data Traditional VoD evolved as a replacement transport or by tape delivery. Its intended of a Pay-Per-View (PPV) model comprising purpose is to deliver a movie, its preview, of a selection of top 20 movies that are and its boxcover along with metadata for a replaced every 2-3 weeks. Initial trial single video title offering in a one-way deployments as far back as 1993 were delivery system (see figure 1). This is the created to see if it could replace the PPV main delivery mechanism from a content model. But really it was the establishment of provider or distributor to a cable headend or digital cable networks that gave VoD a regional cable distribution center. With platform to grow on. Since then, VOD constant requests for new ways to offer deployments among cable operators have Video content, this one logical structure has moved beyond the trial stages towards been refitted constantly to meet new VOD becoming a viable digital cable service service, technology, and process demands; product. It is one of the differentiators but at a considerable operational cost. This

approach becomes harder to maintain as there Assets

is more content, more ways to offer content, and shorter lead times for distributing this The basic building blocks to represent material. logical VOD structures are called assets. Each asset at a minimum has metadata to provide a universally unique identification 1.1 Package (Provider ID/ Asset ID), versioning within the asset (updateNum), and management (Asset Lifetime window -- which determines Title how long the asset can remain with that cable operator). From an asset management perspective, the metadata is necessary to be able to track, manage, and ultimately purge Box-Cover Movie Preview assets from the cable system. There are three general types of assets: content assets, group assets, and metadata assets.

The content asset (CA) contains the Figure 1: VOD 1.1 Structure content file (or link to the file) and the metadata that belongs with the content. The new ADI 2.0 specifications create an Examples of this could be a video (movie, approach to building a VoD structure more preview), still-image (box-cover), or even flexible to these new demands. The ADI 2.0 audio. This content asset by itself has no specifications provide a way to create more context, meaning that it is not yet associated than one type of logical structure while with a particular VoD offering. This has creating a common approach to managing the advantages because the content asset can then delivery of these structures. The delivery be possibly shared with several VOD mechanism has expanded to include multi- offerings at the same time. For example a distribution approaches, more automation, preview could come in with a set of other and increased validation. The additional previews in a barker offering and still be improvements in ADI 2.0 specification allow reused as the main preview for its movie. for more handling of large-sized content files Furthermore it allows the option of separate and the ability to utilize this content in more transport and ingestion of large Giga Byte than one VOD offering. (GB) content files into the cable system on a time schedule different than the rest of the ADI 2.0 SPECIFICATIONS offer metadata. The metadata that goes along with this asset can indicate the size of the The ADI 2.0 structures specification content, if it is corrupted, and whether it is defines the basic building blocks and completely built. Other metadata that is connectors that can be used to build logical contained within the content asset are structures for any type of VOD offering. It shareable metadata to describe the content also describes the message document but non-specific to the offer. An example of envelope needed for delivery of this this would be Screen Format as opposed to information. BillingID.

Actions are done on the content assets standard definition format release. In this through operations. This replaces the verb case, a movie metadata asset could be created structure described in the 1.1 documents. for each version containing the licensing “Accepting” a content asset brings the asset window as well as other metadata specific for into the cable system as an unattached asset. each version. The licensing window metadata “Destroying” the asset removes the asset cannot reside at the content asset level from the cable system by effectively ending because it is metadata pertinent to that the content asset’s lifetime. “Replacing” the particular offer. An interesting aspect of content asset can update the content metadata suborganizing metadata into different assets or even swap the content file though this is is that a different provider can potentially rd not often done except to recorrect existing create each asset. This can allow for 3 party corrupted files. contracted contributors authorized by the provider of the owning group asset. Contrary The group asset (GA) defines the context to this, the 1.1 specifications restrict all to use a collection of assets which could be a metadata and content to originate from the VOD movie, a movie on SVOD, a barker same provider. loop for new action movies, or whatever people possibly define. It is the organizing The metadata operations are in reference asset that can group and identify everything to the particular owner group asset. The including other assets collectively specific to metadata asset can be ‘Added’ to that group that context. This type of group asset has a asset (once the group asset is ‘Opened’ of flag to indicate itself so asset management course). The metadata asset can be systems can identify this as an organizing ‘removed’ from the group asset. Lastly the asset. This group asset can also contain its set of metadata in that asset can be own metadata that is universally applicable ‘Replaced’ as a whole for updates to the for other assets it organizes. metadata asset. The metadata asset is always subjugated to the existence of the owning Since the group asset is the main group asset which means once a group asset organizing asset indicating context, it needs is ‘dropped’ all its member metadata assets to be initially ‘Opened’. Other operations for are removed as well. the group asset is ‘Dropped’ – the GA and its member assets are removed from system, Connectors ‘Replaced’— information in the group asset is replaced, and ‘Closed’—a hint that no In the 1.1 specifications, an implicit more metadata or assets could be added to structural definition is created that restricts the GA context. the logical structure to be only one general type: namely a title with a movie, box-cover, The last building block is the metadata and preview (plus add-ons). This is fine for asset (MA). It needs to belong to only one delivering a top-twenty movie, but was a specific instance of a group asset and cannot rough fit for delivering other types of content exist on its own. It basically suborganizes a or title offers. The ADI 2.0 specifications set of metadata that may be specific to only a create an explicit structural definition where subset of the assets organized. An example of the relationships between the assets under its possible use could be releasing a title in a certain rules can be redefined to suit the high definition format a week ahead of its particular offer type. (see figure 2) The group membership connection was Logical Structure 1 Logical Structure 2 previously talked about in the paper when discussing metadata assets. This is the only Group Asset 1 Group Asset 5 type of membership relationship allowed. It limits the metadata asset to be a member of 1 Reference Member group asset instance and prevents the MA to Metadata exist outside of that context. Reasons for this Group Asset 2 Asset 5 is to tightly connect all elements of the offer Reference together such that it can be managed as a Members Group Asset 4 single entity. This allows for example a Metadata Metadata complete deletion of an offer by simply Asset 1 Asset 2

dropping the organizing group asset without Reference Reference References needing to see how the changes affect other Shareable Shareable Shareable Content Content Content similar offers. Asset 1 Asset 2 Asset 5

The second logical connection tool is the reference pointer. Because of the acceptance Figure 2: Explicit Structural Connectors of the provider ID / asset ID as a universally unique identifier, assets can be tracked and There also exists another connection tool be referenced by this identifier. The one- (associate content signal) that is not required way pointer reference can be embedded in for creating the logical structure but does the metadata of the parent asset and can be consider transportation timing issues for visible to the asset manager for validation or retrieving large sized content assets. As noted be kept hidden until the particular application before, the content assets do not come at the processes it. This is the main method to same time a group asset is opened or when a logically connect the shareable content asset metadata asset is added. Often they already to the offer with a reasonable amount of exist in the system, but a copy needs to be assurance due to the near static nature of a retrieved as the offer is created. This requires content asset. This same reference connector preplanning to transport the file across could also be used between any two assets sometimes very large regional networks. This including outside group assets though the connection tool does not create additional integrity risks would need to be minimized logical structures but ensures that the content by proper logical structure development for assets are physically available to be the specific application document. By this connected with the offer. mechanism, it is possible to apply second level of offers, create playlists from multiple Message Document titles, or allow for other ways to innovate in the future. The message document is the delivery ‘parcel’ that arrives at the cable asset management system. The message document can contain operations and their assets or operations for existing assets in the cable system. Contrary to a 1.1 behavior, these operations do not have to be related to the same offer. This can allow an encoding house to pitch a batch of previews from one or even a set of content providers to cable systems. validation on operations and assets enclosed Also contrary to a 1.1 behavior, the offer has within the document. This is an advantageous the option of being created from one message point to do this type of checking because it is document or multiple message documents. the first point of ingest into a cable system. (see figure 3) The basic level checking of the operations and assets are through message level xml Metadata for the message document are parsing and validations mechanisms as well used for transport types of issues. It allows as content file integrity checks. Note this is rd for “sender” to identify 3 parties pitching not a basic validation of the organizing the message document. It also provides context but a basic validation of the metadata for a unique identifier for the components sent in the message documents. message document (docNumber), creation Validation of the offer happens at the time, relative ingest priority, and contact info application level where the logical structure for transport link issues. An exception to the is fully visible. metadata being used just for the message document is an acknowledgement field. The This separation of the message level and acknowledgement field provides an http application level metadata allows for the connection to report back basic level message layer to be versioned independently from the application level. The asset manager can manage and route assets intended for GA Title applications that it need not be fully MSG Document #1 cognizant of. By allowing for this separation, Accept (CA1) the same message document layer could be Accept (CA2) Accept (CA3) used to deliver components for a VOD Doc Ack Open (GA) MA1 MA2 MA3 application or for an advertising server. Add (MA1) Box-Cover Movie Preview Add (MA2) Metadata Metadata Metadata Furthermore the use of versioning on Add (MA3) namespaces also allows xml support of a mix

CA1 CA2 CA3 of older/newer versions of application and Still- Image Video Video Content Content Content delivery systems. This allows for many on- Asset Asset Asset demand applications to be developed and supported using the common platform Alternative Message Document Distribution Scenario described in this specification.

GA MSG Document #1 Title The message document format can also be Accept (CA1) Doc Ack used to deliver auxiliary information to the Accept (CA2) Accept (CA3) cable systems or back to the provider. An example of this is the return ACK message MSG Document #2 MA1 MA2 MA3 Open (GA) Box-Cover Movie Preview (call it an ACK document) sent back upon Add (MA2) Metadata Metadata Metadata receipt of the message document containing MSG Document #3 operations and assets (call it the ADI2 Add (MA3) CA1 CA2 CA3 document). Add (MA1) Still- Image Video Video Content Content Content Asset Asset Asset This concept is extensible and has created Figure 3: Single and Distributed Approaches to Create a new specification called the ADI2.0 AIM a Structure (asset inventory messages) specification. This specification contains two types of message documents related to VoD, but do specification, file integrity information not create any new logical structures. (filename, filesize, checksum, metadata flag), has been moved to ADI level metadata to The first message document described is allow non-application routers, servers, and the Notification to Deliver Content (NIDC) ingestion points to check for content file which is a message sent from the provider to integrity during transfers without need to the cable operator informing the operator on understand the application context. A a periodic basis what content assets (movies) message-level ACK can indicate as soon as are being planned to be sent. The cable possible in the process if the content files operator can then use this information for needs to be resent. Another content-level asset management and resource management ACK is used to monitor the progress of file of their servers & systems. This is especially ingestion as it propagates through the cable useful as content catalogues increase and network. For more distributed systems, the become more long lasting in cable systems. associate content signal and the content URL is useful to assemble the content components The second message document described with the logical structure in time for it to be is the Provider Notification Message (PMN) operational. In terms of shelf space which is basically another acknowledgement management, the NIDC function gives prior but this time on a content asset level instead knowledge of upcoming content storage of a message document level. This can be demands. Lastly the content files themselves used as an event mechanism to indicate the are shareable to allow them to be repurposed condition of a content asset as it is further for other offerings and services instead of processed in the cable system. requiring the content to be resent for each new purpose. ENABLING NEW MANAGEMENT & SERVICES Asset management functions benefits from the ADI 2.0 specification as well. The By creating a common platform to build management of the asset lifecycle process is and deliver logical structures, the ADI 2.0 improved by separating out an asset lifetime specifications can assist addressing many of window from the licensing window. This the concerns in transport of content and allows for situations such as having an asset management as Video-on-Demand scales up exist in the system but maybe not “in- in volume as well as new services. service” at that time. The specification also creates a clearer demarcation between Delivering content files in a large cable message document functions to deliver asset environment often takes time because of the components and higher-level application transfer of large video files (in GB) across an responsibilities. This aids in the health and increasingly distributed cable network. With fault monitoring of the delivery to headend the advent of High Definition Video, transport network separately from logical increasing the size of content catalogues, structural issues. Also reference links can be shorter delivery lead times, more demand for defined that need to be recognized by the content storage & management, and AMS to do things like validate that the reuse/repackaging of content; it is becoming referenced asset pointed to exists in the increasing critical to minimize the number of system. Alternatively, a reference link can be content file deliveries and increase the defined that only is processed at the robustness of the transfer process. In this new application level and in the context of that specific application processing it. Lastly, possible ways to offer and organize VOD shifting to a schema-based format allows for content material enables more variety of VoD more specific XML parsers to be used to types of services. validate elements and specific value formats of them. The acknowledgement function in TRANSITIONAL CONSIDERATIONS the ADI 2.0 specification improves the initial FOR 1.1 STRUCTURES 1-way pitch by indirectly providing a first level success/fail response for the pitched The ADI 2.0 structure does not create a message document. Video-on-Demand application. It provides the asset structure tools, and delivery The new structural tools developed in the transport mechanism to create an organizing ADI 2.0 specifications can enhance at the asset structure for this. To create an actual application level the value in how assets can VoD application based on the ADI 2.0 be offered. In 1.1 initially, the offer, title, and platform, one or more group assets, metadata content were pitched and identified as a assets, and content assets and their monolithic group. With the break-up of this connections need to be defined. The metadata group and new types of structural concepts elements residing in each of these assets also introduced, it becomes easier to offer content need to be defined. in new ways. For instance dynamic offers can be created fairly quickly from existing In creating the VOD 2.0 application for content assets in the system. Examples of this real world purposes, the existence of 1.1 are: a weekend choice of discounted films, an structure and compatibility needs to be updated barker list for an action movie considered at least for a while. The VoD 2.0 folder, a broadcasting TV show that can be structure may initially be restricted in its offered on VoD just as it starts, coming instance to accommodate a 1.1 transitional attractions that can be updated on all VoD mechanism, but the structure itself should be movies, or VoD that can be offered with or flexible to allow for more this. An example without commercials. of additional features could be support of a playlist to automatically update coming A new concept that can be utilized is the attractions. Another example of this is to standardized use of a playlist which can support alternate versions of the movie determine the continuous play order of a list within the same offer. Since 1.1 VoD systems of assets. This can be used for things like and 2.0 VOD systems would both need to be sports highlights, newscasts, double features, supported at least for some time, this barkers, and other things. With the creation restricted case needs to be considered in the of the metadata asset, there is an ability to structural design of the 2.0 application during have 3rd party contracted contributions as a the transitional phase. part of the group while still retaining an overall owner for the group. For instance a It is important to know that VoD 2.0 music soundtrack for a movie supplied by the systems could still at the same time receive record company could be part of the VoD VoD 2.0 group asset structures that do not offering. Also encrypted and trick files could consider 1.1 backwards compatibility. This be added to the group asset as additional would allow for more creative development content assets that can vary depending on the of services to enhance the VoD experience. It VoD system that it is sent to. With these would also give incentive for older systems additional tools and flexibility, the variety of to transition faster from a 1.1 environment. An example of this could delivery of a scalability issues from a structural collection of TV episodes that can be offered perspective as VOD increases in both volume for one price for a set or an individual price and variety. Many different applications can for a la carte. use these same ADI2.0 specifications to create logical structures for there specific Some of the immediate new features type of VoD offering. The most immediate demanded in VoD services could still be application to be developed on the ADI2.0 deployed in this transitional phase. With the platform will address the single title VoD mechanism of reference pointers, new logical offering that allows for a transitional phase structures can refer to other logical with existing 1.1 VoD platforms. structures. Since this reference mechanism points to the unique identifier of the asset, REFERENCES this same reference could point to a 1.1 VOD title as well. This would allow for some new 1. Cablelabs Asset Distribution Interface 2.0 features to incorporate existing 1.1 titles. Specification Version 1.1, For instance, dynamic category folders like www.cablelabs.com an action folder can refer to a set of existing 2. Cablelabs VoD Content Specification titles including 1.1 titles. Some of these titles Version 1.1, www.cablelabs.com can also belong in other folders simply 3. Cablelabs ADI2.0 Structures through the reference mechanism. These Specification, www.cablelabs.com types of strategies can ease the transitional 4. Cablelabs ADI2.0 Asset Inventory pains of switching cable systems to the new Messages Specification, ADI 2.0 platform. www.cablelabs.com 5. “VoD Victories: Be Careful What You CONCLUSION Wish For”, Carl Weinschenk, Comm. Tecnology, Feb. 2004. This paper describes the ADI2.0 6. “Supporting Large Scale VoD Through specifications. It defines general asset types Common Metadata”, Yasser F. Syed, and their relationships to create many Proceedings of NCTA 2004, New different types of logical structures as Orleans, May 2004. opposed to a single logical structure defined 7. “Game Planning For Bigger, Better for VoD 1.1. It also describes how a structure VoD”, Karen Brown, CED, May 2004. can be created through one or more message 8. “Ingest & Metadata Partitioning: document deliveries into the cable system. Requirements For Television on These new mechanisms create flexibility to Demand”, Robert G. Scheffler, address some of the existing issues with Proceeding of NCTA 2003, Chicago, delivery of large content files, management May 2003. of VOD titles, and creations of new types of 9. “Video-on-Demand:Quality Shows”, The VOD offerings. It also addresses some of the Hollywood Reporter, December 1st, 2003.

VOD SERVERS – EQUATIONS AND SOLUTIONS

Glen Hardin1 and W. Paul Sherer2 1Time Warner Cable, 2Arroyo Video Solutions

Abstract others with complex elegance. At first glance, it is seemingly a basic problem to Video-On-Demand (VOD) is now a widely solve, but, as the multipliers in front of each deployed product with a ready audience. No of these variables scale independently and longer a “trial” product, it is a cornerstone infinitely, the solution quickly becomes offering for the cable industry - generating complex. The physical solution to the revenues, reducing churn and setting MSOs equation can be based in either proprietary or solution apart from satellite. commodity hardware and bonded together with plenty of custom software Yet the technology underpinning VOD services is still in its infancy, and, as new Content: VOD services are developed, the VOD infrastructure must continue to evolve if the The old real-estate adage “location, potential of these new services is to be location, location” has its analogy in VOD realized to its fullest. and it is “content, content, content”. Content is the main driver for the success of VOD. This paper seeks to provide context for Add additional compelling content and the VOD server technology - where it has been, stream use rates will increase. where it is and where it might be going. This discussion is presented in context of the The amount (and type – High Def content changing VOD server equation. is 4 times as resource consuming as Standard Understanding this equation is paramount to Def) of content drives the total amount of understanding the solution going forward. storage the system requires. In the original VOD services, the content variable was limited to the top 100 “hit” titles – requiring THE ORIGINAL EQUATION: CONTENT perhaps 250 content hours of storage. As + STREAM = VOD VOD technologies proved themselves, new services such as subscription video–on- Historically there have been two basic demand were added and the total number of variables to the VOD equation: the content storage hours grew to support them. The variable and the streaming variable. Each hours of storage grew from a few hundred VOD server solution has attempted to hours to 800 hours. With the increase in the understand and resolve the relationship number of subscription services and recent between content and streaming. All vendors new services such as Free-On-Demand, in the marketplace work to optimize Music-On-Demand and High-Definition-On- performance and price as they tackle the Demand (HDVOD), the storage requirement basic problem of how to access the stored quickly has quickly grown to thousands of content, transfer it across the bus architecture hours. and pump it out of the video server without interruptions. Some do it with brute force and Depending on a server’s architecture, everything and it must be accomplished scaling the content storage may be as easy as flawlessly without interruptions. adding more drives or an additional disk array to the system or as complicated as Scaling streaming is a very complicated replacing all the drives in the system. The proposition and different vendors have one thing that is for sure, if the VOD service approached the problem in different ways. is to be successful, the multiplier to content Historically, VOD server vendors relied on variable can go in only one direction, ever core disk Input/Output (I/O) subsystem increasing. performance to attain their stream

VOD Server Streaming Paradigm Streams Versus Content

4000

Trend Future Problem 3500

3000 Historical Performance 2500 Validation Points

2000 Current Problem Trend 1500 Stream Counts Performance At The 1000 Original Problem Extremes Is A Very Expensive 500 Proposition

100 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 Number of Titles

Streaming: performance. Some vendors chose to implement complex interconnect and RAID Streaming needs to access the content architectures to gain the efficiencies of stored on disk and route it across a bus or parallelism and thereby increase streaming interconnect and pump it out of the server. performance, while others scaled through Since the sole purpose of streaming is to simple server replication. In either case, deliver content, all other functions may need validating a server’s streaming performance to operate at a lower priority including the was accomplished by taking a single piece of reception of new content. Content delivery is content and streaming it out at the server’s max stream capacity (the easy way) and into the VOD server platform, and taking a unique piece of content per unique immediately (within seconds) allow all stream to the server’s max capacity (a much customers to stream them. This rapid increase harder problem to solve). Testing at both in ingest requirements is a natural outgrowth extremes guaranteed that the server could of the increase in content offered in VOD deliver the content in any way the customer form, but it also seems to be a universal in could ever order it. This performance at both the various next generation On-Demand extremes came at a relatively high price, but services under development - including

DVB-ASI Architectures- Hardwired Solution VOD Servers ASI QAMs VOD Service Group

VOD Servers ASI QAMs VOD Service Group

VOD Servers ASI QAMs VOD Service Group

GigE VOD Architectures - Stream Anywhere To Anywhere VOD Servers GigE to ASI ASI QAMs VOD Service Group

VOD Servers GigE to ASI ASI QAMs VOD Service Group

GigE QAMs VOD Service Group VOD Servers GigE Switch

was reasonable with a relatively small Network Personal Video Recording (NPVR), amount of content. broadcast “Start-Over” and client applications like Weather-On-Demand. THE NEW EQUATION: INGEST + CONTENT + STREAM = ON DEMAND These real-time acquisition-based services SERVICES greatly impact VOD servers and in multiple ways. Content storage requirements are As discussed above, server architectures growing tremendously as the number of have historically focused on optimizing the networks offering On Demand content output capabilities of their servers at the grows. Instead of supporting 1200 titles, the expense of their input capabilities. However, VOD servers increasingly need to support increasingly a new factor is changing the multiples of that number. Streaming is also original server performance equation. The impacted, both because the wealth of new new factor is ingest. content must be written to non-volatile storage (i.e. disk), and because of the Ingest: increase in the quantity of streams as subscribers access the new content. Servers are increasingly required to Additionally, since the quality of service receive MPEG files in real-time, ingest them must be maintained both for content ingest and for streaming, VOD servers will have to devices. This shift to GigE also reduced the work within even tighter performance barriers to entry, allowing new vendors and tolerances as both these variables scale. This innovation into this market. equation is far more complicated equation than what was originally required in the early The Edge days of VOD. As a result of the shift to GigE within the Architecturally, real-time acquisition- video server, the requirement and costs for based services favor more centralized content DVB-ASI moved further out into the storage solutions that allow single ingest network. To maintain compatibility with the points to serve all customers. The ingest existing QAM devices, new devices were server must have interconnectivity to all developed to translate the GigE back to ASI service groups. Supporting such features in to interface to the existing QAMs. Now, highly distributed server architectures is native GigE interfaces are available from overly complicated and almost impossible. every QAM manufacturer, negating the requirement to convert the GigE signals back With the broader width of content into ASI prior to the QAM. This will further offering and the advances in parallel reduce the cost and complexity of the VOD technologies, the current VOD server solution. architecture paradigm needs to be re- examined. Furthermore, the market now has Transport the historical experience to evaluate the necessary performance requirements against Advancements within the transport the usage patterns of the On-Demand- technologies have greatly facilitated the shift Services offered. from highly distributed VOD architectures to

more centralized architectures. Transport REMOVING THE COMPLEXITY FROM technologies have gone from inefficient ASI THE VIDEO SERVER transports to single GigE pipe on a pair of

Advancement in other technologies, fibers to 40 times 1G, 40 times 2.5G and including software technologies, has allowed finally 40 times10G on a single pair of fibers. the complexity of the VOD server to be Highly distributed architectures also required simplified. multiple instances of storage arrays and copies of the content. Centralizing storage ASI to GigE and/or the servers has the added benefit of allowing for greater efficiencies through The most important shift in complexity of sharing of the storage arrays across many the VOD server was the removal of the streaming devices. As a result, fewer storage DVB-ASI interface and replacement with the arrays are required as fewer copies of the GigE interface. As a result, the VOD vendor content are needed. no longer had to develop and support custom DVB-ASI cards within the server, which was Core Switch a huge cost reduction for the server companies. VOD servers with DVB-ASI also The most desirable and advantageous require the video server’s streaming capacity method of connecting the GigE video server to be in parity with the edge capacity as the into the cable plant is through a GigE core video servers are physically tied to the edge switch. Advancements in switching technologies now allow for fully meshed customer base can now be served by a 2000 non-blocking delivery of video. stream GigE server with the same blocking factor. The integration of a core switch enables video servers to stream from anywhere to Furthermore, the addition of the core anywhere. That is to say, any video server switch between the VOD server platform and streaming port can service every VOD the plant also minimize the requirement to service group. Since content is no longer tied develop methods of interconnecting various directly to storage at the edge, but is now discrete storage arrays though some backend centrally available to all streaming devices, back-end switching fabric. the content is no longer bound to a particular video server component. The core switch also Software Infrastructure has the added benefit of reducing the overall streaming requirement of the server. The Finally, advancements in open software server’s streaming capacity no longer needs standards that allow interoperability between to be in parity with the edge QAM capacity, VOD vendors have greatly influenced the but only with the max stream utilization. This marketplace. MSO’s are no longer held allows the video server to scale captive to a particular vendor once the initial independently from the edge QAM capacity. purchase is made. Each time the system For example, a DVB-ASI server with the expands or major features are added and new capacity of 3000 streams serving a given server capacity is required the MSO can

Commodity Hardware Performance Curve (Based on Moore's Law)

Proprietary Performance Deficit

Proprietary Market Hardware Lag Solutions

Performance Commodity R&D Hardware Cycle Solutions

Market Disadvantage

Market Time Advantage choose the best of breed among the vendors. only control its server costs. If, for a given set This ensures that VOD vendors remain of hardware, a video server can sustain n competitive in terms of price and number of streams, then the minimum per performance. stream cost equals $ / # streams per unit of hardware (not including the cost of PROPIETARY HARDWARE SOLUTIONS development for the necessary software and VS. COMMODITY other associated costs). Regardless of the hardware solution, the software is the The graphic attempts to illustrate the valuable component of intellectual property relationship between video server of any vendor. performance and commodity hardware performance based on Moore’s Law, an industry-accepted concept that hardware It is prophesized that this curve cannot performance will double every 18 to 24 sustain its exponential growth forever but in months. In general, the hardware commodity the near-term it provides guidance and performance curve increases due to parallel insight into the future capabilities of the advancements in all the technologies within market. the PC market: faster and multiple processor machines and bus infrastructures, faster and Proprietary Hardware Solutions: denser DRAM, drive technology, and network interfaces. Many VOD server vendors have developed a proprietary solution by creating The increased performance described and integrating custom hardware components above results in several secondary benefits and/or custom interconnection technology. If for the constrained environment of the cable accomplished effectively, the resultant headend: less space, power, cooling, and solution should outperform what is available wiring are required. The newer solutions are in the commodity market using a similar much more dense and efficient in terms of generation of technology. Mbps per rack unit and the number of Mbps per watt of power consumed. Less power The difference between the performance consumed infers lower cooling requirements. of proprietary solution and the commodity Additionally, as the outputs of the video curve determines the performance advantage server become denser, fewer wires are of the proprietary solution. The performance required to integrate these servers into the advantage translates into a market advantage plant. It has been demonstrated that a server for a period of time until the commodity of 5000 streams @ 3.75 Mbps can now be curve catches up with the proprietary wired into the plant with just a couple of 10 performance. Server companies offering G interconnects. Historically, this proprietary solutions must exploit this finite interconnect would require upwards of 31 time of market advantage through sales to DVB-ASI wires or even 21 GigE recoup their investment in hardware R&D. At connections. Less wiring substantially the same time they must also continue to simplifies the integration work. invest in the next generation server solution lest they fall below the commodity Both proprietary and commodity server performance curve. It is a never-ending race vendors try to optimize their server costs to stay ahead of the commodity curve and a because the stream price is determined by risky business proposition. It is easy for competition within the market. A vendor can vendors with proprietary solutions to fall below the commodity performance curve if Commodity Hardware Solutions: they do not carefully time their adoption of the newer higher performing hardware. There In the emerging VOD industry back in the is a high cost to develop performance gains early 90’s, the raw commodity server market above the commodity performance curve barely was able to eke out enough leading to expensive R&D cycles. Those performance from a given platform to justify R&D costs must be re-cooped before the costs of VOD. The market price for commodity performance catches up, streams was magnitudes higher than it is otherwise, sales opportunities will evaporate today. as the performance advantage disappears. In sum, it is possible to develop a proprietary There are several enabling factors that solution that exceeds the commodity curve, allow for commodity hardware solutions to and the more is invested the longer this be competitive today. First the content advantage will remain. However, the equation has changed dramatically as commodity market has proven time and time described above, i.e., 10,000 hours of content again that the commodity curve will versus the historical 100 hours of content. eventually catch up regardless of the Second, the base hardware available in the technology. commodity market has the necessary off-the- shelf performance required to deliver dense Since MSOs cannot be expected to VOD streaming. Video servers supporting perform forklift upgrades enthusiastically or multiple GigE and 10 GigE pipes per two or frequently, proprietary hardware solutions three rack units. And finally, the MSO also face the challenge of integrating newer market has accepted the premise of caching higher performing hardware into an existing based on its historical content use patterns lower performing solution. Typically, and the cost/performance trade-offs proprietary solutions rely on symmetric associated with cache-based servers. server performance with all machines within a server complex operating at the same VOD vendors with architectures based on performance level. But it does not make commodity off-the-shelf servers abstract the sense to integrate new high performing hardware solution from the software solution hardware and operate it only at the existing and, at a minimum, develop loosely coupled performance levels. Therefore in addition to systems. The VOD delivery solution is constant efforts to keep up with the software–based, which makes the hardware commodity hardware curve, vendors of choice an independent decision. As such, the proprietary solutions must undertake the vendor is able to choose the best-of-breed development of many lines of custom code in within the commodity market. order to have older and newer generation hardware interoperate (if at all possible) at A potential drawback to working solely their respective levels of performance as one with commodity hardware is that the integrated seamless solution. performance of commodity platforms must lag slightly the commodity performance Hence, even in proprietary hardware curve, due to the need to re-qualify new solutions the software is as important as the hardware platforms as they become available hardware and is, in fact, the key intellectual in the market. Best-in-class solutions abstract property within the VOD server platform. out the software from the hardware allowing performance to more closely follow the size of the working set so the minimal commodity performance curve. amount of expensive components may be utilized to achieve the desired level of Vendors using commodity servers must, performance. like those with proprietary architectures, address the problem of integrating CACHING SERVERS advancements in hardware into their solution. To address this issue, a commodity based Overall caching server vendors try to solution needs to be developed in a manner optimize the right mix of components and that supports asymmetric server performance costs to get the greatest return on within the solution. performance

By achieving solution independence from Caching components vs. Costs the underlying hardware, commodity vendors RAM allow MSO’s to utilize existing procurement ~$350 / GB and maintenance contracts for the underlying High Performance Drives hardware. This allows the MSO to leverage ~$5 / GB its volume purchase agreements with Standard Performance Drives commodity hardware vendors. Further, ~$1 / GB internal expertise can be more readily leveraged across hardware platforms which When a caching architecture is part of any are used for multiple service solutions. system be it a microprocessor or a VOD server one of the first questions which must Another reason to focus on commodity be asked is how to determine what to keep in hardware is enhancing the utility of the the cache and for how long. All such systems intellectual property which must be created use a mixture of predictive and reactive by the vendor. Simply put software algorithms to decide what to cache. intellectual property is readily reapplied across multiple generations of underlying The most common predictive algorithm is commodity hardware with little to no the “next obvious thing”. That is based on redevelopment or retraining across whatever is happening now the next obvious generations. This allows the commodity thing will most likely happen next. In a vendor to focus on continual enhancement of microprocessor, this usually means the next system robustness and functionality without instruction after the current one – in a VOD the need for continual investment in long lead server this usually means the frame after this time hardware development cycles in order to one. Some seek to predict events at a much keep pace with the performance created by higher level. In a microprocessor this might the overall computing market. be to predict which program will be run or in a VOD server which title will be played. The All servers, even the historical servers, problem with this approach is the decision at stream from RAM. The difference with the this level depends on factors which are new caching servers is that the RAM is used beyond reasonable prediction a priori. Johnny to capture the “working set” of the cache. That is, the set of content which is active at Carson can die and the “Best of Carson” this given moment and likely to be active in titles can suddenly become very popular. the near future. The goal is to minimize the Janet Jackson can suffer a wardrobe malfunction and a particular sequence from Traditional VOD servers tended to treat a the Super Bowl can see extremely high piece of content as whole but the “wardrobe utilization. malfunction” example illustrates a portion of a piece of content may have radically Most effective caching systems do not different usage patterns associated with it rely on accurate prediction at this level but than other portions. A more efficient cache rather rely on reactive algorithms. The can recognize that this portion has radically approach is to make observations at a slightly different usage and treat it differently than broader level than the low level predictions the rest of that piece of content. These usage described earlier. But to then assume that patterns can happen for many reasons in these higher level decisions will be tend to be addition to an event in the content. New self-similar. That is, for example, if 50 of the forms of navigation such as chaptering can last 100 plays have been for a certain allow entry into a piece on content at a set of sequence of a Super Bowl then is it is likely locations. The chapters in effect become that many of the upcoming plays will be for mini-pieces of content with a larger whole. the same sequence. In this way a cache can Another trend is the creation of virtual assets. adjust quickly to decisions which it cannot These are logical assets composed of predict accurately beforehand but it can components from other assets which are observe accurately and react to. perceived as single asset by someone viewing

Simultaneous Access Content Curve Simultaneous Streams on a Single Title Cache Tier 1 Tier Cache 2 Tier Cache 3 Tier Cache 4 Tier Cache

Content - Number of Titles

them. This could be an edited topical news many MSO’s are now considering update or a play list of music videos. geographic resiliency in their system planning. Centralized but in at least two Another reason to no longer view a piece locations and interconnected through of content as whole is responsiveness. Even switching and transport. The idea is that the if a user has held a bookmark for resume for content storage must survive a natural a long enough time that the local cache has disaster such as a hurricane or tornado or a flushed the content, it is desirable that the manmade problem such as a portion of the VOD system should be able to “resume” very power grid going offline. By having content quickly and easily. To do this, the caching stored in geographically diverse location the system should retrieve content starting from odds that such an event would take two or the point where the resume occurred, rather more facilities offline is greatly reduced than from the opening scenes of the piece of compared the odds of a single facility going content. offline.

As has been noted earlier, VOD is now a For many years the resiliency of content cornerstone revenue generating service. As was assured via RAID 5 technology. With such it must be robust and available 24 X 7. costs of modern disks becoming so low, in This means MSO’s should look to vendors to many situations it is simply more cost provide automatic resiliency to system faults effective to keep multiple copies. The issue and to allow for maintenance and upgrade with RAID 5 resiliency structures is that without service outage. there is an assumed extremely high bandwidth path among the components of a One important consideration is the unit of RAID 5 structure. This is reasonably easy to failure for which the system is resilient. achieve among components in single system When considering the failure modes which but becomes increasingly onerous when must be compensated for, most would think resiliency is spread across many systems. In a some hardware fault such a network interface RAID 5 system of n components when a failure. In reality, for all types of video failure occurs, all n-1 remaining components server, whether based on proprietary must participate in recovering the lost hardware or commodity hardware, the most information - which must be regenerated common failure mode is a failure in the through computation. The bandwidth impact software not a failure in the hardware. So in of this process will often make it impractical this sense the most common unit of failure particularly across geographically diverse must be considered to be the server itself. content storage facilities. This means the entire server function must be recoverable automatically. That is, the All of the above is leading to the creation current workload must be recovered intact by of caching tiers – each with a role to fulfill. other systems without the need for human The exact boundaries of these tiers will to intervention. This level of resilience has been large degree be determined by the cost and applied to telephony and data applications reliability of transport between the tiers. but is just now being designed into VOD servers. Level 1 – The Edge

In terms of content availability, while This is the tier closest to the service there is a definite trend to centralized storage, groups. The content kept here will be fairly active and will have fairly high reuse. The Because of the large amount of storage lower two more expensive caching components of cost storage components are used here. It RAM and high performance disk drives will would be expected for every stream play be used in this tier. The key here is to capture request which accessed the regional or the “working set” with the minimum amount national library that thousands or tens of storage capacity. The goal of this tier will thousands of stream play requests would normally be to satisfy 90 to 95% of the have been seen by the edge systems. stream requests within this tier to provide sub-second responsiveness. However, while Level 4 – Regional or National Library handling the bulk of the stream, this tier would have very little of the overall cache This tier is the ultimate source of content storage capacity - typically only a few available to any edge system. All content percent. This creates a very efficient usage of available to any edge system is resiliently the high cost caching components in this tier. stored somewhere in the library. This tier will have geographic resiliency and multiple Level 2 – Local Storage points of ingest. The regional or national library tier has the greatest storage capacity The tier behind the edge serves to and the greatest resiliency of all the tiers. decouple the higher instantaneous performance of the edge from the much more NEW TESTING PARADIGMS modest performance of the local library. This The advent and adoption of new caching tier can be seen as a performance matching server architectures requires a re-examining tier which uses greater cache storage capacity of how servers are tested and qualified. The to only allow a small number of the total old method of validating a server’s streaming stream requests to have impact on the local performance by taking single piece of content library. At this level the storage is still and streaming it out at the server’s max viewed primarily as cache with the implicit stream capacity and by taking a unique piece assumption that if need be content can always of content per unique stream up to the retrieved from the local library and resiliency server’s max will not result in the desired and of content is less important. practical price and performance point. The historical usage patterns must be applied to Level 3 – Local Library the testing and validation of the new caching servers. This tier is the demarcation of the relatively inexpensive and readily available A new term needs to be defined to help local transport to the relatively expensive and normalize the validation of servers. The term scarce long haul transport. This content has Cache Gain represents the additional high resiliency so that single failures of steaming capacity above what is available devices or servers can be handled without through the core disk I/O subsystem requiring retrieval from the regional or performance. For example, if a given server national library. If the area served by the has a disk I/O performance of 1000 unique local library is large or prone to problems streams but can deliver not only the 1000 such as hurricanes the storage may be unique content streams from disk but an implemented with geographic resiliency. The additional 500 duplicate content streams from percentage of all content accessible from the cache, the server would demonstrate a Cache associated edge systems is very high. Gain of 50 percent. In that same example, if a server were able to deliver 1000 duplicate What is a believable cache gain? Is it content streams from cache then the Cache 10%, 100% or 1000%? Only careful Gain would be 100 percent. monitoring of live systems in the field will Since caching servers vary greatly in the prove out the actual achievable gains. number of cache tiers and performance within the tiers, setting simple easy standards However, this architectural advance of performance is difficult. The process of clearly represents the next step in the normalizing the system performance to core evolution of the infrastructure for on demand disk I/O performance provides a baseline services. With the advent of this architecture from which to work. the stage is set for a plethora of new services reliant on much greater breadth of content CONCLUSION and much more dynamic usage. “Start Over” and network PVR fit well to this architecture. The jury is still out. Although all the best More applications will come. and brightest within the VOD server community agree that there are cache gains to One can now see the infrastructure be made as of now, there is not enough coming into being which will enable the empirical data of cache effectiveness to efficient delivery of a fully personalized unequivocally say what exactly what the entertainment to every MSO customer on cache gains are for a given type of content every television. and service.

ISBN 0‐940272‐01‐6; 0‐940272‐08‐3; 0‐940272‐10‐5; 0‐940272‐11‐3; 0‐940272‐12‐1; 0‐940272‐14‐8; 0‐ 940272‐15‐6; 0‐940272‐16‐4; 0‐940272‐18‐0; 0‐940272‐19‐9; 0‐940272‐20‐2; 0‐940272‐21‐0; 0‐940272‐ 22‐9; 0‐940272‐23‐7; 0‐940272‐24‐5; 0‐940272‐25‐3; 0‐940272‐26‐1; 0‐940272‐27‐X; 0‐940272‐28‐8; 0‐ 940272‐29‐6; 0‐940272‐32‐6; 0‐940272‐33‐4; 0‐940272‐34‐2; 0‐940272‐35‐0; 0‐940272‐36‐9; 0‐940272‐ 37‐7; 0‐940272‐38‐5; 0‐940272‐39‐3; 0‐940272‐40‐7; 0‐940272‐41‐5; 0‐940272‐42‐3; 0‐940272‐43‐1; 0‐ 940272‐44‐X; 0‐940272‐45‐8; 0‐940272‐46‐6; 0‐940272‐47‐4; 0‐940272‐48‐2; 0‐940272‐49‐0; 0‐940272‐ 50‐4; 0‐940272‐51‐2; 0‐940272‐52‐0; 0‐940272‐53‐9; 0‐940272‐54‐7

© 2015 National Cable and Telecommunications Association. All Rights Reserved.