EFFENBERGER LAYOUT 2/20/07 3:28 PM Page 51

TOPICS IN OPTICAL COMMUNICATIONS An Introduction to PON Technologies

Frank Effenberger, Technologies US; David Cleary, Calix, Inc.; Onn Haran, PMC Sierra Glen Kramer, Teknovus, Inc.; Ruo Ding Li, Motorola, Inc.; Moshe Oron, Tellabs, Inc. Thomas Pfeiffer, Alcatel-Lucent Germany

ABSTRACT adopted IEEE standard that was designed for fixed and mobile access networks. It has a useful Passive optical networks are the most impor- range of about 5 km at a data rate of 70 Mb/s. tant class of fiber access systems in the world WiFi is more mature than WiMAX, but it has a today. This article first reviews the reasons why range of only 100 m and a bit rate of 10–50 the PON as a general architecture is so impor- Mb/s. In spite of this limitation, WiFi is more tant. We then outline in some depth the tech- widely used for access today than WiMAX due nologies used to implement this architecture, to its maturity. including the G-PON and E-PON systems being Although both WiFi and WiMAX are rela- deployed today, and the advanced PON systems tively low cost to deploy, they lack sufficient that provide the evolution path to ever higher to support applications. These bandwidths. technologies use a point-to-multipoint architecture. This means that bandwidth is INTRODUCTION: shared by multiple users — in some cases hun- dreds of users. Consequently, WiFi and WiMAX THE MOTIVATION FOR PON are useful for Web surfing applications, but impractical for higher-bandwidth and higher-rev- One of the most critical decisions for any busi- enue applications such as IPTV. ness involves the purchase of capital equipment. Another access technology option available to Of the many factors that influence this decision, service providers is copper — more specifically, equipment cost and the resulting revenue poten- (DSL) over copper. Unlike tial are two of the most important. Service pro- wireless, DSL uses a point-to-point architecture. viders face this decision when upgrading existing So instead of sharing 50 Mb/s over all sub- access networks or expanding into new areas. scribers, DSL can provide 50 Mb/s to each sub- They want to minimize the cost of deploying scriber. Unfortunately, DSL shares a access equipment while maximizing revenue shortcoming with wireless: it is a noise-limited from the service offerings. Of these two parame- access technology. In other words, the effective ters, the cost of deployment is easier to deter- bandwidth DSL provides to a subscriber depends mine than revenue potential because future on the level of noise, which in turn depends on revenue involves considerable speculation. As a the length of the copper loop. DSL is capable of result, the raw bandwidth capabilities of an 50 Mb/s for loop lengths less than 300 ft, but can access technology are often used as a proxy for only provide 10 Mb/s at 10,000 ft. If operators revenue potential. Thus, the most important want to offer a compelling video service with 30 decision a service provider makes when purchas- Mb/s, they need to shorten loop lengths to ing network equipment is how to strike a bal- roughly 3000 ft or less. This is a viable approach, ance between minimizing the equipment cost but the cost is only slightly lower than an all- and maximizing the bandwidth. fiber approach. The passive optical network (PON) is just The final option to consider for access tech- one of several access technologies used by ser- nology is fiber. An can be archi- vice providers, but it enjoys a dominant posi- tected using either dedicated or shared fibers. A tion in the access market. Before discussing dedicated fiber plant, often referred to as a the specific details of the PON, it is worth- point-to-point network, provides a dedicated while to survey the alternate access technolo- fiber strand between each subscriber and the gies in order to understand the reasons for the central office (CO). PON’s success. In a shared fiber architecture, a single fiber Access networks fall into three categories: from the CO serves several dozen subscribers. wireless, copper, and fiber. Wireless has the low- This fiber is brought to a neighborhood where est deployment cost because it has the lowest the signals are broken out onto separate fibers costs. WiFi (802.11) and WiMAX that run to the individual subscribers. (802.16) are the standards for wireless access Point-to-point fiber networks have a low mar- and access. WiMAX is a recently ket penetration mainly due to the additional cost

IEEE Communications Magazine • March 2007 0163-6804/07/$20.00 © 2007 IEEE S17

Authorized licensed use limited to: IEEE Xplore. Downloaded on December 3, 2008 at 10:42 from IEEE Xplore. Restrictions apply. EFFENBERGER LAYOUT 2/20/07 3:28 PM Page 52

GIGABIT PASSIVE Up to 60 km OPTICAL NETWORKS Up to 20 km ONU STANDARDIZATION HISTORY The gigabit-capable PON (G-PON) is speci- fied by International WDM ONU OLT Union — Telecommunication Standardization ONU Up to Sector (ITU-T) G.984 series [1-4]. G-PON 128 split definition began in the Full Service Access Network (FSAN) consortium in 2001. In Jan- Video ONU uary 2003 the first two standards were Tx approved by the ITU-T, covering the require- Downstream: 1490 nm, 28 db, 2.488 Gb/s ONU ments and basic architecture (G.984.1), and Upstream: 1310 nm, 28 db, 1.244 Gb/s the physical-medium-dependent (PMD) layer Downstream video: 1550 nm (G.984.2). In February 2004 G.984.3 specifying the G-PON transmission convergence (TC) layer was ratified, followed by G.984.4, which I Figure 1. G-PON physical . standardizes the G-PON management require- ments. Since then, a few amendments have reached consent by the ITU-T on most of the it adds over a shared fiber infrastructure. documents in the series. Depending on the average loop length, the con- struction costs of outside plant based on dedicat- PMD LAYER ed fiber exceed those of outside plant based on The G-PON network architecture supports a shared fiber by anywhere from 20 percent to 100 two-wavelength WDM scheme for downstream percent. and upstream digital services (Fig. 1). Addition- In shared fiber architectures, there are two ally, another downstream wavelength is allocated ways the signals are broken out. One method is for distribution of analog video service. The net- called active (AE), and the other is the work supports up to 60 km reach, with 20 km PON. With AE the individual signals are split differential reach between optical network units out using electronic equipment near the sub- (ONUs). The split ratio supported by the stan- scriber. In the PON the signals are replicated dard is up to 128. Practical deployments typically passively by the splitter. would have lower reach and split ratio, limited A shared network based on a PON has sev- by the optical budget. eral advantages over one based on AE. The ITU-T G.984.2 specifies the PMD layer for outside plant of a PON incurs lower capital G-PON, covering the range of G-PON upstream expenditures as it has no electronic compo- and downstream bit rates, and the optical param- nents in the field. The PON also lowers the eters for the various rate combinations. operational expenditures, since there is no As network operators requirements evolved, need for the operators to provide and monitor the preferred G-PON bit rate was selected to be electrical power in the field or maintain back- 2.488 Gb/s downstream, 1.244 Gb/s upstream. up batteries. A PON has a higher reliability This focus has then allowed the definition of than AE because in the PON outside plant best practice optical parameters for G-PON, there are no electronic components, which are which was documented as an amendment to prone to failure. Lastly, perhaps one of the G.984.2. The parameters, known as Class B+, most crucial features of a PON-based access apply to a network with or without a video over- network is its signal rate and format trans- lay and to ONUs based on either APD or PIN parency. Upgrading to higher bit rates is sim- technology. pler for a PON than for AE. Both require upgraded electronics in the CO and customer GTC LAYER premises, but, unlike AE, there is nothing that The G-PON TC (GTC) layer specified by [3] needs upgrading in the outside plant for a performs the adaptation of user data onto the PON, as the passive splitters are agnostic to PMD layer. Additionally, the GTC layer pro- PON speed. For all of the reasons cited above, vides basic management of the G-PON network. the PON is by far the most widely deployed The GTC layer defines two adaptation access technology. The rate and signal format methods for data transport: asynchronous transparency became a sort of insurance poli- transfer mode (ATM) and G-PON-encapsula- cy that eased carriers into deploying PON tion-method (GEM). However, as GEM has outside plants with the understanding that an become the preferred method, ATM is not dis- access network could flexibly be upgraded as cussed hereafter. GTC with GEM allows low new technologies mature or new standards overhead adaptation of various protocols, evolve. including Ethernet and time-division multiplex- Not surprisingly, this article is dedicated to ing (TDM). GTC also provides the medium various flavors of PONs that all use the same or access control (MAC) function, coordinating very similar outside plants, but differ significant- the interleaving of upstream transmissions ly in signaling rates, data formats, or protocols from multiple ONUs. they employ. These PON technologies include The control functions of GTC consist of a Gigabit PONs, Ethernet PONs, and wavelength- protocol and procedures for registering ONUs to division (WDM) PONs. the G-PON network, and monitoring their health

S18 IEEE Communications Magazine • March 2007

Authorized licensed use limited to: IEEE Xplore. Downloaded on December 3, 2008 at 10:42 from IEEE Xplore. Restrictions apply. EFFENBERGER LAYOUT 2/20/07 3:28 PM Page 53

Ethernet TDM OMCI Service management GTC Equipment TC adaptation sublayer management

ATM GEM adaptation GTC OAM

Framing sublayer GTC/PMD config, Embedded OAM fault, performance management Frame Encryption FEC MAC PLOAM sync ONU activation Security DBA Physical media dependent (PMD) layer

User plane C/M plane

I Figure 2. G-PON functional relationships and layering.

and performance. GTC also allows configuration of transport features such as forward error cor- rection (FEC), encryption, and bandwidth allo- ONU cation. Figure 2 illustrates GTC layering, and the main functions of the user and control planes. T-cont A PortID 22 The GTC is divided into two sublayers. G-PON interface The lower framing sublayer defines the GTC PortID 6 T-cont B frame structure, which is asymmetrical, carry- PortID 12 ing different overhead information down- stream vs. upstream. The GTC uses a 125 µs downstream frame, and also transports an 8 T-conts Service kHz signal that provides a reference clock to Bundles of GEM ports GEM ports adaptation May be per CoS; serve as blocks; Service Contain flows from ports the ONUs. The upstream frame comprises a bandwidth allocation unit; logical/physical ports; map payload sequence of transmissions from ONUs as dic- identified by Alloc-ID identified by Port-ID over GEM tated by the optical line terminal (OLT). GTC framing in both directions is rate agnostic; I Figure 3. The hierarchy of G-PON multiplexing: ports, T-Conts, and PONs. that is, different G-PON rates maintain the same frame structure and vary only in the amount of payload. The downstream GTC comprises the physi- protocol-independent connection-oriented cal control block (PCBd), a header containing encapsulation for variable-size packets. GEM’s all overhead fields, followed by the payload virtual connection unit is called a GEM port, part. The PCBd includes framing related fields, and may contain a flow to/from a physical or and the operations, administra- logical port of an ONU. GEM frames include a tion, and maintenance (OAM) (PLOAM) field. 5-byte header indicating the port ID and length The PLOAM carries a message-based protocol of the frame. GEM frames may be fragmented; for PMD and GTC layer management. Finally, hence, a client packet may span multiple GEM the PCBd includes the bandwidth map field frames. G.984.3 includes appendices specifying specifying the ONUs’ upstream transmission transport of Ethernet and native TDM over allocation. GEM. On the GTC upstream, each ONU transmis- Figure 3 illustrates GEM ports’ flow in the sion is headed by a physical layer overhead field context of the G-PON multiplexing hierarchy. As (PLOu), including a preamble and delimiter, shown, GEM ports are bundled onto transmis- which are configurable by the OLT. To assist sion containers (T-conts). A T-cont is the unit of with dynamic bandwidth allocation (DBA), the upstream bandwidth allocation by the OLT. The PLOu may include the dynamic bandwidth T-cont arrangement is configurable by the OLT; report field (DBRu), which carries traffic queu- however, popular schemes are a single T-cont ing reports from ONUs. The PLOu may also per ONU, or multiple T-conts, one per service include a PLOAM field of identical format to class, per ONU. the downstream PLOAM. The PLOAM and The OLT bandwidth allocation method for DBRu are optional and present in a frame only ONU upstream transmission may be static or upon OLT request. dynamic (DBA). Two methods of DBA are The higher sublayer of GTC is the TC adap- defined for G-PON: status-reporting DBA, tation sublayer based on GEM. GEM defines a which is based on ONU reports via the DBRu

IEEE Communications Magazine • March 2007 S19

Authorized licensed use limited to: IEEE Xplore. Downloaded on December 3, 2008 at 10:42 from IEEE Xplore. Restrictions apply. EFFENBERGER LAYOUT 2/20/07 3:28 PM Page 54

ONUs. The MIB comprises a set of managed Open systems entities, each containing a set of attributes. Cre- IEEE 802.3 ation of managed entities and their attributes is interconnection (OSI) layering diagram reference model designated to either the OLT or ONU. Since G-PON ONUs may support a broad variety of interfaces and services, OMCI model- Application MAC control ing is very rich in content. However, each MIB instance, representing a specific ONU, contains Presentation Media access control (MAC) a small subset of objects. OMCI models physical aspects of the ONUs, such as their equipment Reconciliation configuration, power, and various port types, Session Gigabit media such as plain old telephone service (POTS), Eth- independent ernet, xDSL, T1/E1, radio frequency (RF) video, Transport interface (GMII) and MoCA. At the service layer, OMCI covers Physical coding sublayer (PCS) high-speed access using various flow Network classifications and quality of service (QoS) Physical medium attachment (PMA) schemes, TDM-voice, voice over IP (VoIP), cir- Physical medium dependent (PMD) cuit emulation service (CES), IPTV, and more. Data link For each of those objects OMCI supports con- Medium figuration, fault, and performance management. Physical dependent interface (MDI) Additionally, OMCI standardizes the software download for ONUs and the housekeeping of Medium the MIB itself.

FUTURE G-PON EXTENSIONS I Figure 4. Relationship of IEEE 802.3 layering model to OSI reference model. A few G-PON enhancements are currently in the works. They include the following: • Definition of wavelength blocking filters. field, and non-status-reporting DBA, which is The filters would be supported at G-PON based on OLT monitoring per T-cont utilization. ONUs to ensure that next-generation Refer to [5] for a more detailed description of ONUs using additional wavelengths could SR-DBA. in the future be installed on currently The GTC layer control plane is mainly oper- deployed G-PON optical data networks ated via the PLOAM message protocol and (ODNs) side by side with G-PON ONUs. some overhead fields referred to as embedded • Extension of a G-PON’s optical budget to OAM. It includes the following management allow deployment of longer reach and high- functions: er split ratio. This may require an active • PMD layer management — Configuration extender box to be deployed at the ODN. of upstream overhead; monitoring health of • Inclusion of higher data rates. The down- physical layer, and generation of alarms or stream rate would likely be 10 Gb/s, but the statistics accordingly. upstream rate is still an open question of • GTC layer management — Configuring 2.5, 5, or 10 Gb/s. GTC framing options, such as usage of upstream/downstream FEC, requesting PLOAM, DBRu, and so on. ETHERNET PASSIVE • ONU activation — The GTC layer defines OPTICAL NETWORKS the process to activate an ONU on the G- PON network, including a ranging proce- EPON HISTORY dure to measure the ONU distance and set In November 2000 IEEE 802.3 announced a call its equalization delay. The optical power for interest for a new study group called Ether- level of the ONU may also be tuned. net in the First Mile (EFM). The group was to • Encryption management — GTC mandates extend Ethernet into the subscriber access area. Advanced Encryption Standard (AES) as Ethernet over point-to-multipoint (P2MP) its downstream encryption mechanism, with fiber (also known as EPON) became one of the a per-ONU encryption key. Encryption may focus areas of this group, along with Ethernet be selectively applied on a per GEM port over copper, Ethernet over point-to-point (P2P) ID basis. A procedure is defined for key fiber, and OAM tracks. In September 2001 the exchange. IEEE Standards Board approved the EFM Pro- ject Authorization Request, resulting in the for- G-PON MANAGEMENT mation of the P802.3ah task force. Network operators require full management of The EFM task force completed its charter in G-PON systems’ equipment and services, while June 2004, culminating in ratification of IEEE supporting interoperability between ONUs and 802.3ah [6]. OLTs of different vendors. G.984.4 specifies the ONT management and control interface (OMCI) SCOPE OF WORK to address those requirements. IEEE 802.3 focuses on two lower layers of the OMCI comprises a full ONU management open systems interconnection (OSI) reference information base (MIB), and the ONT manage- model [OSI94]: the physical and data link layers. ment control channel protocol (OMCC) that Each of these layers is further divided into sub- conveys MIB information between the OLT and layers and interfaces. Figure 4 shows the sublay-

S20 IEEE Communications Magazine • March 2007

Authorized licensed use limited to: IEEE Xplore. Downloaded on December 3, 2008 at 10:42 from IEEE Xplore. Restrictions apply. EFFENBERGER LAYOUT 2/20/07 3:28 PM Page 55

All services use the EPON’s Ethernet same LLID (LLID 1) roots are 2 ONU 1 4 unmistakable. EPON 2 1 1 traffic uses the same 3 1 Ethernet packet 1 3 1 1 2 4 2 Physical ONU is registered OLT 1 format, with 3 as a single device 1 1 standard IPG, 2 4 Virtual as found in any 2 ONU 2 LLID Header Payload FCS enterprise switch. Preamble Ethernet frame Different services Virtual use different LLIDs ONU 3 For that matter, (LLIDs 2, 3, 4) Virtual EPON uses the same ONU 4 MAC found in any (a) Physical ONU is registered as multiple virtual ONUs IEEE 802.3-compliant device.

ONU 1 All services use the same LLID (LLID 1)

1 1 1 Physical ONU is registered OLT as a single device 1 1 1 2 2 3 Virtual Upstream burst ONU 2 2 1 1 1 2 Virtual 3 ONU 3 Different services use different LLIDs Virtual (LLIDs 2, 3, 4) ONU 4 LLID Header Payload FCS Preamble Ethernet frame Physical ONU is registered as multiple virtual ONUs

(b)

I Figure 5. EPON a) downstream operation; b) upstream operation.

ers and interfaces defined for Ethernet devices While in the IEEE 802.3ah standard EPON operating at 1 Gb/s data rates. link budget was conservatively specified as 24 dB with minimum 1:16 split ratio, in practice the EPON TECHNOLOGY transceiver technology has matured enough so EPON technology provides bidirectional 1 Gb/s that components providing 29 dB power budget links using 1490 nm wavelength for downstream became commercially available, resulting in most and 1310 nm for upstream, with 1550 nm EPON-based networks being deployed with a reserved for future extensions or additional ser- 1:32 split ratio, with some as high as 1:64. vices, such as analog video broadcast. EPON’s Ethernet roots are unmistakable. EPON’s rapid adoption was driven by the EPON traffic uses the same Ethernet packet early decision to define the physical layer specifi- format, with standard IPG, as found in any cation using relatively minor modifications to enterprise switch. For that matter, EPON uses inexpensive high-volume 1 Gb/s optical compo- the same MAC found in any IEEE 802.3-com- nents. This has greatly reduced optics cost to pliant device. The new P2MP connectivity is levels comparable to those of continuous mode supported by a protocol called Multipoint Con- optics. trol Protocol (MPCP), which uses standard Using the same philosophy of “define the Ethernet packets generated in the MAC control specification for rapid high-volume deployment,” sublayer. the EPON upstream burst lock timing was EPON does not use encapsulating framing in relaxed to use available continuous mode mixed either the upstream or downstream direction; signal components. The downside is somewhat instead, the content of the Ethernet preamble is lower upstream utilization, but since other access modified. An upstream burst is simply a technologies are far more asymmetric, this slight sequence of Ethernet packets with regular IPG difference was deemed minor. between them, preceded by a longer sequence of

IEEE Communications Magazine • March 2007 S21

Authorized licensed use limited to: IEEE Xplore. Downloaded on December 3, 2008 at 10:42 from IEEE Xplore. Restrictions apply. EFFENBERGER LAYOUT 2/20/07 3:28 PM Page 56

IDLE codes used for receiver synchronization. the exception of a special algorithm defined by Expecting large-scale Any management or control information is deliv- the major carrier in China for its network. ered in normal Ethernet frames. deployments of An ONU in customer premises equipment EPON MANAGEMENT LAYER G-PON systems to (CPE) is enumerated by the OLT in the CO OAM functionality is another important EPON start soon, network equipment using an MPCP handshake. The pro- breakthrough. Ethernet now includes link layer cess is: management that enables OLTs to remotely operators and • Using the discovery GATE message, the manage attached ONUs. system vendors are OLT sends a request to all unregistered OAM is established after the discovery pro- ONUs to transmit. cess and is maintained by periodic message seeking NG-PON • An unregistered ONU answers by using a transmission. Information about remote failures solutions that can REGISTER REQ registration request mes- is conveyed using flags in OAM messages to coexist with GPON sage. indicate failure status. The remote ONU can be • When received and approved, the OLT reg- instructed to return incoming packets as part of on the same fiber isters the ONU using the REGISTER mes- the remote loopback functionality. plant and enable sage. Link monitoring, where any Ethernet variable • The handshake ends with the ONU of the remote port can be retrieved by the OLT, gradual network acknowledgment REGISTER_ACK. is arguably the most useful EPON OAM func- capacity upgrades. During steady state operation, the OLT con- tion. trols the ONU’s transmission window with OAM link information can be extended GATE messages. The ONU reports its queue beyond the OLT by placing a Simple Network status using REPORT messages. The OLT then Management Protocol (SNMP) agent at the calculates the ONU transmission window length OLT. A soon to be finalized RFC, “Managed using DBA. Objects of EPON,” details the EPON MIBs [7]. All time-driven events are synchronized to An EPON CPE device contains much more the PON clock, a 16 ns resolution counter that is than a MAC. OAM includes vendor extension carried in all MPCP messages. The ONU uses mechanisms to provide a convenient and the received timestamp to lock to the OLT time lightweight method to manage the additional base. The OLT uses returned timestamps to functionality. This can lead to differing OAM measure ONU round-trip delay and schedule variants as carriers customize their products. collision-free upstream transmissions. EPON’s packet preamble contains addition- FUTURE EPON EXTENSIONS al fields not found in packets sent over P2P A significant EPON enhancement to run at Ethernet links. In downstream transmission the higher speed has begun. The IEEE has formed logical link ID (LLID) field defines the desti- the P802.3av task force to consider the definition nation ONU. An ONU filters the received of an EPON PHY that operates at 10 Gb/s frames based on the LLID in the frame’s downstream and 1 or 10 Gb/s upstream. This preamble and its own unique LLID value enhancement would provide a significant capaci- assigned by the OLT (Fig. 5a). A special value ty increase for TDM PON systems. is reserved for broadcast messages sent to all ONUs. In upstream transmissions the LLID EXT ENERATION OLUTIONS field marks the source ONU (Fig. 5b). A cyclic N -G PON S redundancy check (CRC) field validates pream- Historically, data rates associated with broad- ble integrity. Most ONU equipment registers band consumer service offerings have increased as a single ONU and uses a single LLID for at a rate of approximately 1.3 times/year. This data transport. However, some equipment reg- growth has been driven by services such as con- isters as multiple virtual ONUs, thereby estab- vergent subscription television and the Internet, lishing multiple LLIDs. This allows EPON to high-definition television, digital photography access the same traffic granularity on the PON and video, new models for content production, as G-PON. distribution, and marketing, possible re-emer- When a physical ONU registers as multiple gence of thin client computing, and so on. Pro- virtual ONUs, the OLT treats each virtual jecting this trend into the future, in the long ONU as a separate ONU. Correspondingly, the term we will face bandwidth demands beyond OLT grants each virtual ONU separately, current G-PON capabilities, requiring R&D in including repeated allocation of the optical this field already. overhead. The OLT also maintains a separate Different groups around the world have management channel to each virtual ONU, and recently started to address this topic. Both has to identify the SLA allocated to each virtu- FSAN and IEEE are now discussing ways how al ONU. to extend their standards to 10 Gb/s line rates. EPON uses a frame-based FEC mechanism Several research projects around next-genera- based on the RS(255,239) algorithm. Each frame tion PON (NG-PON) are investigating the is encoded separately, and all per-frame parity topic on a wider scope, for example, the Euro- bytes are added at the end of the frame. This pean PIEMAN and MUSE II projects in which approach allows ONUs without FEC capabilities different hybrid network solutions are evaluat- to receive FEC-encoded frames, ignoring the ed that combine the classical TDM/time-divi- appended parity data. FEC can be selectively sion multiple access (TDMA) PON with WDM activated per ONU. channel allocations as well as with optical Although not defined in the IEEE 802.3ah amplification and transparent long-haul feeder specification, all EPON implementations incor- transport. porate encryption. Encryption is AES-based with Expecting large-scale deployments of G-

S22 IEEE Communications Magazine • March 2007

Authorized licensed use limited to: IEEE Xplore. Downloaded on December 3, 2008 at 10:42 from IEEE Xplore. Restrictions apply. EFFENBERGER LAYOUT 2/20/07 3:28 PM Page 57

A major drawback of OLT WDM-PON as λ compared to Tx, 1 ONT 1 TDM-PON such as λ W Tx, 33 EPON or GPON, is M D u M λ the requirement to x λ Rx, 1 1 provide multiple λ λ ... λ1 2 λ 32 33 optical ports at the λ

Tx, router

32 AWG W central office. D To gain acceptance it M λ λ ... λ33 32 is therefore required Rx, λ 64 λ 33 32 to use highly D λ ONT 32 integrated multiple e 64 λ channel transmitter m W Tx, 64 u D x and receiver arrays. M λ Rx, 32 λ Rx, 64

I Figure 6. A typical logically point-to-point WDM-PON architecture.

PON systems to start soon, network operators B+) to 21 dB, potentially allowing low-cost and system vendors are seeking NG-PON solu- WDM sources to be used. tions that can coexist with G-PON on the same Upgrading an existing PON to the above P2P fiber plant and enable gradual network capaci- WDM-PON requires replacing the existing ty upgrades. At the same time, it is highly power splitter with an AWG router. However, required to keep the fiber plant as transparent this upgrade is not particularly desirable, as it as possible while moving to NG-PON in order requires work on the outside plant and disrupts not to block further evolution paths. The time existing customers. A centralized split PON (star consuming and costly deployment of optical topology) can relatively easily be upgraded to fibers, especially in the distribution plant and such a kind of WDM-PON. For distributed split drop sections, must remain in place for PONs (tree and branch), it is impractical to take decades without needing modifications or this upgrade path. replacements. Another disadvantage of this approach relates to the loss of transparency of the outside plant. WDM PON An alternative architecture that avoids this issue WDM PONs have been actively researched as a reuses the existing PON and keeps the power potential technology for NG-PON. This PON splitters in place. Wavelength selection at the uses multiple wavelengths in a single fiber to ONU is performed using an additional bandpass multiply the capacity without increasing the data filter (1 dB loss), and at the OLT by an AWG or rate. Different realizations have been proposed, a set of TFFs. A class B+ link budget is thus of which a majority focus on the network archi- increased from 28 to 34 dB. This is compensated tecture in which a passive wavelength router is for by the fact that the line rate can now be used to replace the passive splitter in the PON reduced by a factor of four while still offering fiber plant. As a result of this, each OLT-ONU eight times the bandwidth per user of the origi- pair has a dedicated and permanent wavelength nal G-PON. assignment, and requires two transmitter/receiv- In the first scenario above (with the wave- er pairs to form a point-to-point link (Fig. 6). length router at a remote ) the gained 7 dB A passive wavelength router located at the link budget might as well be spent for an addi- remote node is realized by arrayed waveguide tional 1:4 power splitter after the router, thus grating (AWG) or a set of thin film filters offering, say, one (WDM adapted) G-PON per (TFFs). An AWG can operate over multiple free WDM channel instead of providing simple P2P spectral ranges, permitting use of the same connections. With the number of users per G- device for both downstream and upstream trans- PON now being reduced to four, the bandwidth mission. To allow for outside environments, an per user is again increased by a factor of eight, AWG needs an athermal design. AWGs have an but now for 128 users. optical loss of around 5 dB, which is about 12 dB less than that of a typical 1 × 32 power split- LOW-COST WDM SOURCES ter. Taking into account the second AWG in the Unlike in dense WDM (DWDM) transport sys- CO, a WDM-PON based on AWG architecture tems used in long-haul and metro networks, it is would reduce the link budget from 28 dB (class too expensive and impractical to implement

IEEE Communications Magazine • March 2007 S23

Authorized licensed use limited to: IEEE Xplore. Downloaded on December 3, 2008 at 10:42 from IEEE Xplore. Restrictions apply. EFFENBERGER LAYOUT 2/20/07 3:28 PM Page 58

WDM PONs using DWDM lasers that emit More near-term targets relate to extending As the deployment unique or tunable wavelengths, particularly at the G-PON physical layer to its logical reach of of PONs grows into the ONUs. Past and current research has been 60 km and 128 ONUs. The goal is to maintain focused on finding low-cost solutions to provide the G-PON current wavelength plan of the many millions of so-called colorless ONUs so that a single type of 1480–1500 nm downstream and 1260–1360 nm homes served, it can ONU can be used everywhere in the WDM upstream bands and to use, for example, semi- be seen that a new PON. A single type of ONU eliminates the issues conductor optical amplifiers (SOAs) or optical- related to inventory, maintenance, and installa- electronics-optical (o/e/o) converters in both era of access tion associated with individual DWDM transmit- directions to extend the reach and split ratio networks is upon us. ters. from today’s 20 km to 60 km and from 1:32 or One early suggestion for realizing low-cost 1:64 to 1:128, respectively. To reduce the ASE The 100 year-long WDM sources is spectral slicing. With a broad induced signal-to-noise ratio degradations in case history of the copper optical spectrum source such as an LED, the of SOAs, the upstream ONU wavelengths might network is finally WDM router or filter will automatically generate be restricted to a smaller wavelength range than the “correct” wavelength channel from the input defined today (e.g., to 1300–1320 nm). Special coming to an end, spectrum. Unfortunately, very high slicing losses care has to be taken to cope with the bursty and the age of PON of up to 18 dB make this approach less attractive nature of the upstream transmission in order to than it might appear at first sight. avoid varying optical gains originating from dif- has begun. More recently, a WDM PON system based ferent power levels and guard times between on wavelength-locked Fabry-Perot (FP) lasers, bursts. Also, the optimum positioning of optical applying injection of spectrally sliced amplified or o/e/o repeaters has to be evaluated for differ- spontaneous emission (ASE), was proposed ent network layouts. and commercialized. The system consists of modified FP lasers as transmitters for both ONCLUSION OLT and ONU. ASE generated from an C Erbium doped fiber amplifier (EDFA) is sent This article has outlined the current and next from the CO through the transmission fiber. generations of PON technologies. While there After passing through the AWG, it is spectrally are considerable differences between these sys- sliced into multiple narrow bands, each of tems, there are also striking similarities. This which is injected into the identical FP lasers at should be no surprise, as they share the same different ONUs, forcing them to operate on a fiber medium and physical topology. Fundamen- single wavelength mode, different for each tally, the differences amount to an issue of ONU, instead of emitting multiple modes. The design style and base technology choice, rather most recent version of the product supports 16 than anything profound. Also, as experience has WDM channels at 200 GHz spacing, each oper- shown, all technologies have found their applica- ating at 1.25 Gb/s and supporting about 21 dB tions, and all are likely to coexist for the foresee- ODN link budget. able future. In another approach WDM lasers located at Most important, all of these systems have a the OLT send their unmodulated emission to similar set of requirements on the access cable the ONUs for modulation and then reflect this facilities. Since the cost of deploying cables is by modulated light back to the OLT. A reflective far the largest expense in any wireline network, optical amplifier (RSOA) is it is critical to get it right the first time. And used at the ONU to perform the modulation, because all PON systems readily support the amplification, and reflection. However, recent same outside plant, it means that network opera- studies have shown that such schemes, even tors can deploy PONs today with one technolo- with optimized design, suffer from various gy, knowing that someday they could migrate to reflection and backscattering issues, thus limit- another system. ing the supported link budget to 16 dB, regard- As the deployment of PONs grows into the less of data rate. many millions of homes served, it can be seen A major drawback of WDM PON compared that a new era of access networks is upon us. to TDM PON such as EPON or G-PON is the The 100-year history of the copper network is requirement to provide multiple optical ports finally coming to an end, and the age of the at the CO. To gain acceptance, it is therefore PON has begun. necessary to use highly integrated multiple channel transmitter and receiver arrays. Opti- REFERENCES mized array designs also offer the potential to [1] ITU-T G.984.1, SG 15, “Gigabit-Capable Passive Optical reduce electrical power consumption and heat Networks (G-PON): General Characteristics,” Mar. 2003. dissipation. [2] ITU-T G.984.2, SG 15, “Gigabit-Capable Passive Optical Networks (G-PON): Physical Media Dependent (PMD) IMPROVING THE OPTICAL POWER BUDGET Layer Specification,” Mar. 2003. [3] ITU-T G.984.3, SG 15, “Gigabit-Capable Passive Optical There is interest in exploring cost saving through Networks (G-PON): Transmission Convergence Layer service node consolidation by using a high split- Specification,” July 2005. ting long-reach optically amplified PON. One [4] ITU-T G.984.4, SG 15, “Gigabit-Capable Passive Optical Networks (G-PON): ONT Management and Control early prototype version of such a system was Interface Specification,” June 2005. called SuperPON, targeting 100 km and 2048 [5] A. Cauvin et al., “Common Technical Specification of the G- ONUs, and was developed by the European PON System among Major Worldwide Access Carriers” IEEE PLANET project in the mid-1990s. The present Commun. Mag., vol. 39, Oct. 2006, pp. 134–41. [6] IEEE 802.3ah, “Ethernet in the First Mile,” June 2004. PIEMAN and MUSE II projects are heading [7] IETF draft-ietf-hubmib-rfc3636bis, “Managed Objects of toward similar figures, but additionally include Ethernet Passive Optical Networks (EPON),” to be pub- the WDM dimension. lished.

S24 IEEE Communications Magazine • March 2007

Authorized licensed use limited to: IEEE Xplore. Downloaded on December 3, 2008 at 10:42 from IEEE Xplore. Restrictions apply. EFFENBERGER LAYOUT 2/20/07 3:28 PM Page 59

BIOGRAPHIES (GPON) product lines. Prior to this, he worked at Advanced Fiber Communications as a broadband DLC architect, and RUO DING LI received B.A. and M.S. degrees in physics from between 1982–1998 at Telrad, Israel, as a software manag- Beijing Normal University, China, in 1984 and 1987, respec- er for switching/SONET products. He received a B.Sc. in tively, and a Ph.D. degree from Free University of Brussels, computer science and mathematics from Hebrew Universi- Belgium, in 1991. He has been with Texscan, Uniphase, ty, Jerusalem, Israel, in 1982. and Sycamore Network, working in the area of fiber optical communications system engineering and architecture. Cur- THOMAS PFEIFFER joined Alcatel (now Alcatel-Lucent) in 1986 rently, he is with Motorola Laboratory, focusing on optical doing basic research on linear and nonlinear transmission access technology. in optical fibers, including the development of fiber ampli- fiers and lasers for up to 40 Gb/s and for HFC networks. He GLEN KRAMER is chief scientist at Teknovus, Inc. He received then started investigating low-cost optical system solutions his Ph.D. in computer science from the University of Cali- for access. Since 2002 he is in charge of Alcatel’s research fornia at Davis. He chairs the 10 Gb/s EPON Task Force in activities on optical technologies and systems for next-gen- IEEE 802.3. The author of Ethernet Passive Optical Net- eration access networks. works (McGraw Hill, 2005), he has done extensive research in areas of traffic management, QoS, and fairness in access DAVID CLEARY is director of advanced technology at Calix. networks. He is the founder of the EPON Forum, and He is responsible for identifying new and emerging tech- teaches EPON tutorials and workshops at conferences nologies that will lower cost and improve performance of around the world. access networks. He represents Calix in international stan- dards bodies, and serves as the company’s liaison to media FRANK EFFENBERGER ([email protected]) is the direc- and industry analysts. He has a Ph.D. from the University tor of FTTx in the Advanced Technology Department of of Colorado and has been working in the field of electro- Huawei Technologies USA. He received his Ph.D. in electri- optics for over 20 years. cal engineering from the University of Central Florida. Pre- vious positions include the analysis of all types of access ONN HARAN is a PMC-Sierra Fellow. He joined PMC-Sierra networks at Bellcore, and the research and design of B- following Passave acquisition. At Passave, he served as PON and G-PON standards and systems at Quantum Bridge chief technology officer and was instrumental in develop- Communications. He is the editor of the major PON stan- ing the IEEE802.3ah specification. He was previously with dards in the ITU. Texas Instruments, where he contributed to the standard. Previously, he served in a top R&D unit of the MOSHE ORON is a systems engineer at Tellabs, Petaluma, Israel Defense Forces (IDF). He holds a B.Sc. (cum laude) California, where he focuses on defining system architec- from the Technion, Israel Institute of Technology, and an tures for Tellabs’ Broadband-PON (BPON) and Gigabit-PON M.Sc. from Tel-Aviv University.

IEEE Communications Magazine • March 2007 S25

Authorized licensed use limited to: IEEE Xplore. Downloaded on December 3, 2008 at 10:42 from IEEE Xplore. Restrictions apply.