Marketing V. Reality in Multi-Antenna Wi-Fi

Marketing v. Reality in Multi-Antenna Wi-Fi

Ruckus Wireless | White Paper A Back-to-the-Basics Guide to 802.11 Beamforming for Discerning Buyers

Executive Summary A number of Wi-Fi vendors have recently begun to market this firm foundation of the basic principles, we then compare beamforming features, attempting to follow our lead in the and contrast the main varieties of multi-antenna techniques field. As rapidly growing numbers of mobile devices and employed in current Wi-Fi systems, concluding with a look at increasing bandwidth demands hit wireless LANs everywhere, real-world performance comparisons. it’s no surprise that customers and suppliers alike are focusing We don’t want to spoil the ending, in part because understand- more attention on the fundamental radio performance of their ing the results we show in the wrap-up comparison below networks. requires you to go through all the building blocks to see why Most new vendor marketeer converts to beamforming are this is the case. But we do want to emphasize up front that the unfortunately glossing over (as marketeers will do) a number network performance differences between the WLAN technol- of critical limitations in the chip-based beamforming their APs ogy choices you face can be very large, and they can have an offer, including the absence of current client support for the equally large impact, for better or for worse, on the experience standard, incompatibility with the higher data rates in 802.11n of your WLAN users. So please read on, and let us help you that require spatial multiplexing, and at best only small perfor- become a better-informed buyer. mance gains. WLAN lifetime purchase-decision impact of BeamFlex adaptive The problem is that this marketing gloss is beginning to con- antennas v. conventional 3x3 AP with generic beamforming: fuse those responsible for WLAN purchase decisions, who are 70% difference in client service capacity. typically not radio-frequency engineering specialists and there- 4-Year Aggregate Client Throughput per AP, Terabytes fore unable to see past the gloss to make adequately-informed choices. As we’ll show, not all multi-antenna techniques in Wi-Fi Ruckus 3x3:2 AP 440 are equally capable of solving the basic WLAN challenge — de- with BeamFlex livering fast, reliable connectivity to many clients in challenging environments. Generic 3x3:3 AP with Explicit 260 Beamforming This paper goes back to the basics of how radio signals work in Wi-Fi systems, explaining the functions multiple antennas can more capacity 70% for client service serve, using plain English, lots of pictures, and only as much with Ruckus trade jargon as necessary, defined carefully and clearly. Using Page 2 Marketing v. Reality in Multi-Antenna Wi-Fi

1. Introduction really understand this stuff, either), and certainly with little It’s often said that imitation is the sincerest form of flattery. At chance of savvy customers calling the technical foul. Ruckus, we are indeed flattered that so many in the WLAN indus- Since all this stuff matters so much to the performance of your try have embraced the idea of what is commonly called beam- network, and the experience of your users, we’re ready to take forming, since we have for several years now been pioneering the up the challenge of getting you enough knowledge to make application of multi-antenna technology in Wi-Fi. In fact, we have good Wi-Fi network design decisions, however. To build the built our entire business on the performance, ease of use, and foundation required to accurately assess claims and likely cost-effectiveness gains this technology can provide, when done performance benefits for multi-antenna systems, we’re going right. Our rapidly growing community of customers is enjoying to go back to the basics here, using lots of pictures and defin- the substantial benefits our approach to shaping radio energy in ing carefully the necessary jargon along the way to try to help their networks brings every day, in terms of higher client through- make things very clear. put rates, better coverage, more resistance to interference, and much more reliable Wi-Fi performance. We start with an old-fashioned single-antenna access point, shown in Figure 1 (below), with a common omni-directional an- For organizations currently shopping for new WLAN gear, tenna, or an “omni”. When this device transmits, as the anten- on the other hand, many vendors’ approaches to the whole na’s name suggests, it sends the same signal in all directions. subject of beamforming are rather less than helpful. As is all too While this approach has a certain satisfying design simplicity, it common in the wireless industry, marketing departments have has substantial performance disadvantages. The vast majority gotten a little out of control and are promoting “beamforming” of this radio energy is completely wasted, since an access point capabilities that have little or no connection to the real world of can only talk to one client at a time. This excess energy creates Wi-Fi radios, planting the seeds for significant customer confu- self-interference in the WLAN, stepping on neighboring APs sion and disappointment in the process. As we’ll show in the and their clients and reducing the possibility of channel reuse following material, it’s easy to imitate beamforming messages nearby. Meanwhile, the tiny fraction of transmit energy that in pretty brochures (with several vendors stretching the truth actually reaches the client yields a lower throughput rate, as while they’re at it), but another thing entirely to deliver genuine we’ll show shortly, than would the case if the energy could be and material performance gains in day to day operation. focused more tightly (since client throughput is directly related We’ve put together this brief paper to help clear up some con- to available signal strength). fusion about how different versions of multi-antenna systems FIGURE 1: Radio signal distribution pattern from an access work, what kind of performance gains have been validated for point with one omni-directional antenna. them in the real world, and when they will be commercially relevant given chipset adoption and interop certification timelines. Omni 2. What’s this “beamforming” thing again? Transmit (Tx) Pattern This will take some explanation. One of the reasons for vendor marketing excesses in this area is the simple fact that the concepts in multi-antenna technology can get complicated, especially for Wi-Fi buyers and users who may have strong technical backgrounds in other fields but who have little experi- Next we introduce another omni antenna to begin to explore ence in the arcana of advanced radio-frequency engineering the tools this might provide us for better control of the radio — in other words, the majority of you. We certainly don’t mean signal. As shown in Figure 2 (see next page), the combination that observation as a slight to any of you, since it’s just fine with of two copies of the same signal transmitted from two neigh- us that you leave the gnarly technical bits of this area to the ex- boring omni antennas creates a set of intersecting troughs perts. No one would expect you to engineer your own compact and peaks, much like the wave rings that might be achieved by fluorescent light bulbs or LCD televisions, either. The downside, tossing two separate rocks into a still pond at the same time. though, is that it’s easy for vendors to spew nonsense about RF In some locations, the peaks of the signal from transmit behaviors while sounding vaguely credible, often without even realizing themselves that they’re completely disconnected from the real physics of wireless (since very few marketing people Page 3 Marketing v. Reality in Multi-Antenna Wi-Fi

antenna 1 (“Tx 1”, in the jargon) line up in space and time with One way in which the phenomena of constructive and destruc- the peaks from Tx 2 — this is referred to as constructive com- tive combinations of multiple transmit sources can be put to bination. In other locations, the peaks of Tx 1’s signal are lined productive use is through the addition of active control of indi- up with the troughs of signal from Tx 2, which yields destruc- vidual transmit signal phases. This is the narrow (and most tech- tive combination. If a receive (Rx) antenna is placed in the zone nically correct) definition of the term beamforming, and the of perfect constructive combination, it would pick up roughly type of multi-antenna processing that is coming into fashion in twice the signal strength of a single Tx antenna’s output, with- a growing number of vendors’ Wi-Fi products. (This is not what out doing any intelligent work on its own — its analog receive Ruckus products do today, which we’ll explain in section 4.) In electronics simply sum the signals received automatically. In beamforming systematic manipulation of the phase of signals contrast, a zone of complete destructive combination would transmitted from multiple antennas is used to place zones of yield zero signal, a phenomenon useful in reducing intra-AP in- constructive combination that fall ideally at the location of the terference at the network level (more on this later). The repeat- client of interest. We illustrate this in Figure 3. The depiction of ing patterns in radio communication signals allow us to use the transmission pattern has been cleaned up here in order to the concept of phase to describe the peak or trough match-up show only the areas of constructive combination, which is the relationship between two different signals. common convention for showing “antenna patterns” — essen-

FIGURE 2: Fundamental concepts in multi-antenna processing for increased signal strength (technology often broadly categorized as “beamforming”)

Signal strength Tx Antenna 1 2x signal Time Receive Antenna (Rx)

Constructive In Phase Combination

Tx Antenna 2

No signal Destructive Tx 1 Combination 180º Out Rx of Phase

Tx 2

FIGURE 3: How signal phase control works in beamforming.

Tx 1 Constructive Combination

Destructive Combination Tx 2

Phase adjustment per antenna to direct location of constructive combination on client Client antenna Page 4 Marketing v. Reality in Multi-Antenna Wi-Fi

FIGURE 4: Phase-based beamforming can only be used with copies of the same signal.

Same waveform Different waveforms

2x signal strength Garbage Tx 1 Tx 1

In Phase Rx In Phase Rx

Tx 2 Tx 2

tially the equivalent of geographic contour maps, where height We put this into practice first in Figure 4. Remember that the in this case is signal strength. You can see where the term receive processing expected of the client in a phase-based beamforming arises, since the resulting patterns tend to have beamforming system is just simple summation of the signals re- lobes of constructive combination areas that look somewhat ceived at any given time. Figure 4 illustrates visually a point we like “beams” of energy shining out from the antenna array, can also make through a music analogy: if you had two audio much like the beam of a flashlight, that are “formed” by the speakers side by side blasting two different tunes (say, Bach’s system controlling the individual phases of the antennas. “Con- Brandenburg Concerto #1 and Black Sabbath’s Iron Man) at the trolling phase” in this context means essentially “changing same time, what you’d hear would be essentially just noise. If when you start transmitting.” Outside the lobes of the pattern they were both playing the same tune at exactly the same time, as drawn are areas of destructive combination. you’d hear a louder version of the tune.

Phase is adjusted by the system to compensate for different The other key underlying assumption in beamforming is that travel times between each antenna and the client of interest, the system knows where the client is, in an RF-signal-path so that the signals from Tx 1 and Tx 2 arrive at Rx with their sense of the term “where”, in order to choose phase adjust- peaks aligned in time, maximizing signal strength at the client. ments to point or “steer” one of its beams in the right direc- So far, so simple. Things need to get a little more interesting to tion. In the Wi-Fi world there are two different approaches assess with clarity what’s being used in Wi-Fi today. being used for educating the AP about client “location”:

A key underlying assumption in beamforming is that both Tx 1 Implicit Beamforming and Tx 2 are transmitting the same signal. To understand why In the normal course of communication from client to AP, the that is important, we need a short word on the nature of the AP can detect from its multiple antennas the different phases signals themselves. So far we’ve been using simple sinusoidal of arrival of a signal from the client on each of the AP’s anten- curves to depict our wireless signals, so they may appear to be nas. This is roughly analogous to the way human ears process just generic energy levels, and it might not be obvious why one sounds that arrive at each ear at different times and therefore segment of the orange curve couldn’t be combined at Rx with give an indication of the direction from which the sound came. any other. The signal-shape reality of today’s encoded wireless It’s worth noting that for the same reason our ears can be de- transmissions is much more complex. In order to achieve high ceived by sound bouncing off acoustically reflective surfaces, throughput, many bits are transmitted at the same time on a the AP’s impression of the client achieved by measurement of single signal “wave”, in a format called a constellation, where at a signal arrival phase differences is not a terribly reliable indica- single snapshot in time each bit holds a particular place in a ma- tion of actual physical location — because of signal reflection trix in the real and imaginary number space. Fortunately for the off surfaces in the environment in which the AP and client are many of you we’ve just lost with that last sentence, this particular operating. In implicit beamforming the phase differences are batch of complexity is not important to understand in the con- used as just that — phase differences that should be applied to text of evaluating multi-antenna processing technology. For sake the AP’s transmit antennas to achieve maximum constructive of keeping things as simple as possible, we’ll show “real” wireless combination on the next transmit to that client. signals through squiggles we borrowed from the audio electronics world, in order to illustrate some key concepts. Page 5 Marketing v. Reality in Multi-Antenna Wi-Fi

The flaw in using the radio-space characteristics of the uplink 3. OK, got beamforming. Now what’s from client to AP as a model for what should be used to ma- “spatial multiplexing” ? nipulate signals in the downlink is that signal behavior can dif- Phase-based beamforming was actually the simpler story. The fer substantially between the uplink and downlink paths. We’ll more complex topic that is more essential to the high data rates show quantitatively in our subsequent section on performance built into the 802.11n Wi-Fi protocol is spatial multiplexing, or SM. assessment that implicit beamforming really just doesn’t work In the less technically-minded segments of the Wi-Fi community very well, for primarily this reason. the term MIMO is often used as a synonym for SM, which it’s not, really. Since we’re trying to sort things out clearly here, it’s worth Explicit Beamforming cleaning this one up at the outset with a couple definitions: To improve on the poor performance of implicit beamforming, the alternative that is just now coming into infrastructure MIMO = an acronym for “multiple input, multiple output”. De- products in Wi-Fi involves communication from the client to fined from the vantage point of the air between an AP and a cli- the AP of what would work best for the client (in terms of the ent, the term refers simply to a system design where more than AP’s transmit phase and other settings), given the client’s cur- one transmit antenna put signals into the air (multiple input), rent vantage point in the radio space. No Wi-Fi clients on the and where more than one receive antenna extracts these sig- market today support this feature, however, so this approach nals from the air (multiple output). MIMO systems can be used faces some substantial commercial challenges for broad adop- to do a variety of different forms of multi-antenna processing, tion. More technically, while it certainly improves the quality of but not necessarily all at the same time, as we’ll discuss below. the AP’s understanding of the characteristics of the best radio SM = spatial multiplexing. A system in which a single flow of path to the client, it remains subject to a significant handicaps bits is broken up into two or more streams that are transmit- endemic to small-antenna-count beamforming in the context ted into the air from Tx antennas separated in space (with one of Wi-Fi networks. We elaborate on the nature of these chal- stream per antenna). These multiple streams are then extracted lenges in the performance section. from the air by the same number of Rx antennas, also sepa- One final note on this category: since the manipulation of rated in space, and then recombined into the original single bit signal phase must be done at the PHY layer (at the lowest level flow. SM requires a MIMO antenna architecture, i.e. multiple of hardware), both implicit and explicit beamforming function- antennas on both client and AP. The point of the exercise is to ality must be built into the Wi-Fi chipset. For this reason, these get bits over the air interface faster by adding additional paral- techniques are often referred to as “chip-based beamforming” lel “lanes” for traffic on the same spectrum. in the industry. Figure 5 illustrates how this works from a radio signal perspective. It may appear at first glance to break the rules for signal

FIGURE 5: How spatial multiplexing works.

Different coded waveforms Matrix decode, (on same frequency) not simple sum Data stream split ...1011 Tx 1 Rx 1 ...1011

...1101001101 and cross-encoded for Tx Data stream recombined ...1101001101

...0010 Tx 2 Air, Rx 2 ...0010 with spatial diversity

or Page 6 Marketing v. Reality in Multi-Antenna Wi-Fi

combination we introduced in section 2, since the blue and can do spatial multiplexing or phase-based beamforming, but orange signals received on each client antenna look like our not both. The purpose of a third transmit antenna introduced metaphorical Bach and Black Sabbath arriving at the same ear- on a conventional AP is unclear in a world filled with at best drum, just creating noise. The details of exactly how SM signals two-antenna clients: what are you going to do with it, exactly? are coded on transmit and decoded on receive go beyond the In Figure 6 we illustrate the scenario that gives us pause. If you scope of this paper (and would require reviving cool matrix- use two of your three Tx antennas for SM in order to achieve math tricks you learned in the undergrad linear algebra class the substantial data rate gain it can provide, you have the you’ve certainly forgotten by now). For our purposes here we choice of either leaving the third Tx antenna quiet, or doing can just say that the combination of special pre-encoding and something active with it. You can’t do more SM, since you only spatial diversity (signals differing based on where in physical have two antennas on the client (a third stream would required space they are received) are leveraged in receive processing to a third client Rx antenna). What happens if you try to do phase- disentangle the streams that got combined in the air between based beamforming with it? Let’s say, for sake of argument, Tx and Rx. Note that this technology can be used by clients to that you replicate the second of the two encoded streams send multiple streams to APs as well. Also note that without and transmit it on the third antenna. Let’s also assume (and spatial diversity between the streams, decoding fails. this is a generous assumption that would be nearly impossible One Wi-Fi vendor has recently introduced APs with three anten- to achieve in practice) that you can adjust phase on the third nas capable of both transmit and receive (a “3x3:3” architecture antenna so that the simple sum of signals from Tx 2 and Tx 3 in Wi-Fi parlance, which means 3 transmit, 3 receive, and capabil- result in a higher-amplitude version of stream 2 arriving at both ity for 3 spatial streams). This vendor also claims performance Rx 1 and Rx 2 with the same diversity that would have been the gains from the explicit beamforming processing that has just case if Tx 3 wasn’t involved at all. The SM decoding process on been implemented in the new generation of Wi-Fi chipsets. the client would then yield a normal version of stream 1 and a We’ll address those alleged performance gains quantitatively in stronger-signal version of stream 2. So what? If you anticipated a moment, but first this product form presents a logical problem this result and tried to use a higher modulation class (i.e. higher related to spatial multiplexing that we need to explore. bit rate) on stream 2’s signal, you would break the process of multiplexing and demultiplexing the original bit stream, since As we’ve shown, spatial multiplexing requires that Tx antennas it’s essentially a simple round-robin algorithm that assumes the produce different signal waveforms in order for the system to bit rates of the two multiplexed streams are the same. If, on code and decode the multiple streams, while phase-based the other hand, you leave the modulation class alone, then you beamforming requires the transmission of multiple copies of have a lot of extra signal strength on stream 2 that services no the same signal waveform. With two transmit antennas (and useful purpose for the target client, but that does spew more two receive antennas on the other end of the air link, to com- co-channel interference into the surrounding network. plete the MIMO requirement for SM), it’s obvious that a system

FIGURE 6: How SM fails in combination with phase-based beamforming with 3 Tx antennas.

Weaker signal strength on stream 1 than ...1011 Tx 1 on stream 2, so lower data rate?

...1101001101 Rx 1 ... 1 1

...0010 Tx 2 ...?0?0?1?01 Duplicated for Data stream recombination Rx 2 beamforming? ...0010 out of sync

...0010 Tx 3 or, alternatively, extra signal strength in stream 2 goes unused for higher throughput, causing unnecessary and harmful self-interference in the network Page 7 Marketing v. Reality in Multi-Antenna Wi-Fi

FIGURE 7: Importance of spatial multiplexing to higher 802.11n bit rates.

802.11 PHY Rates Overview 802.11b 802.11a/g 802.11n Peak Bit Rate, Mbps 40 MHz channel 20 MHz channel Modulation Bit Rate Modulation Coding Bit Rate MCS Spatial Modulation Coding 800nsGI 400ns GI 800ns 400ns Mbps Mbps Index Streams GI GI DBPSK 1 BPSK 1/2 6 0 1 BPSK 1 7 7 14 15 DQPSK 2 BPSK 3/4 9 1 1 QPSK 1/2 13 14 27 30 CCK 5.5 QPSK 1/2 12 2 1 QPSK 3/4 20 22 41 45 CCK 11 QPSK 3/4 18 3 1 16-QAM 1/2 26 29 54 60 16-QAM 1/2 24 4 1 16-QAM 3/4 39 43 81 90 16-QAM 3/4 36 5 1 64-QAM 2/3 52 58 108 120 64-QAM 2/3 48 6 1 64-QAM 3/4 59 65 72 135 64-QAM 3/4 54 7 1 64-QAM 5/6 65 72 135 150 8 2 BPSK 1/2 13 14 27 30 GI=800 ns 9 2 QPSK 1/2 26 29 54 60 10 2 QPSK 3/4 39 43 81 90 11 2 16-QAM 1/2 52 58 108 120 12 2 16-QAM 3/4 78 87 162 180 13 2 64-QAM 2/3 104 116 216 240 14 2 64-QAM 3/4 117 130 243 270 Note: 802.11a/b/g are all single stream 15 2 64-QAM 5/6 130 144 270 300 ... 3 ...... Abbreviations 23 3 64-QAM 5/6 195 217 405 450 MCS modulation and coding scheme ... 4 ...... GI inter-symbol guard interval 31 4 64-QAM 5/6 260 289 540 600

and a graphical view:

Peak bit rate, Mbps 1,000

100

10 802.11n >1stream, 40 MHz, 400ns GI 802.11n 1stream, 40 MHz, 400ns GI 802.11n >1stream, 20 MHz, 800ns GI 802.11n 1stream, 20 MHz, 800ns GI 802.11a/g 802.11b 1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 23 31 MCS number Page 8 Marketing v. Reality in Multi-Antenna Wi-Fi

We’ll return to the question of the merits of a 3x3:3 architecture As with most true innovation, we’ve moved beyond existing in today’s market in section 7. categorization, so we tend to just stick with BeamFlex.

To put this question and our performance assessment section There are two primary advantages in our ability to use a com- 6 in context, we repeat for convenience in Figure 7 the well- bination of multiple antennas on individual Wi Fi radio chains: known tabulation of bit rates defined for the various modula- better antenna patterns, and compatibility with spatial multi- tion and coding schemes in the IEEE’s Wi-Fi standards series. plexing. We’ll look at each in turn. The highlighted part of the 802.11n table emphasizes that all the more interesting rates require two or more spatial streams. FIGURE 8: High-level view of Ruckus BeamFlex architecture. In other words, sensible 802.11 systems will bias their designs toward maximizing the use of spatial multiplexing for clients Off-the-Shelf RF that can support it. 802.11 Silicon RF

4. What does Ruckus BeamFlex™ do, then? Host MAC PHY RF n antenna elements While we often use the term “beamforming” for the multi-antenna processing we do in our APs, as a matter of convenience Better antenna patterns when we don’t want to have to get into a long discussion of Better than what? The baseline is the kinds of beam patterns distinctions between different multi-antenna architectures, that can be created with phase-based beamforming — for that’s not an entirely accurate term for what we do, given the which, see Figure 9 on the next page. This approach is limited strict definition we introduced in section 2. As we show in to using only as many antennas as it has radio chains. With Figure 8 below, our approach involves digitally switching a only two or three antennas to work with, the shapes that can selection from a large number of antenna elements to connect be created are fairly limited in structure, and they are marked with the individual radio chains in the RF front end of off-the- consistently by a couple of substantial issues: they are symmet- shelf Wi-Fi silicon. A “radio chain” is the RF engineering term ric, and their lobes or beams tend to be relatively narrow. The for the analog radio part between the chip doing the digital Wi symmetry means that from the perspective of a target client, Fi protocol processing and the antennas. In BeamFlex there is half the energy transmitted is wasted. From the vantage point no analog adjustment of phase on each radio chain. Instead, of neighboring APs in the WLAN, this energy is worse than an optimal combination of antennas is selected on a packet- wasted — it means louder co-channel interference. The narrow by-packet basis to focus patterns of radio energy in the right beams mean they are pretty unforgiving about inaccurately direction based on the inherent characteristics of the antenna pointed beams. If an AP’s estimate of the right phase combina- elements themselves. The selection for a given client is based tion to use for its antennas is a little off (either because it was on the throughput last achieved with that client, confirmed using imperfect implicit feedback, or because the higher-qual- through the ACK packet that is a standard part of the 802.11a, ity explicit feedback forthcoming in Wi-Fi systems has gotten b, g, and n Wi-Fi protocols and that is supported by all clients out of date because of delays in its use, high client mobility, today. [Note that some vendors like to make up for their lack of or rapid changes in the environment like a door closing), the intellectual property contributions in Wi-Fi by de-positioning beam formed will fall where the client isn’t, and an area of BeamFlex as “non-standard” or (gasp) “proprietary” — which destructive combination will fall where the client is, making the couldn’t be further from the truth. Our APs are absolutely 100% whole exercise worse than useless. compliant with the 802.11 protocols and require absolutely zero special behavior on the part of clients.] Ruckus BeamFlex, in contrast, creates highly asymmetric patterns that have much more forgiving lobe shapes, and with Of the many terms used in the general area of multi-antenna huge variety across physical as well as polarization space — see processing techniques (such as smart antennas, beam switch- Figure 10. Because the n elements of a Ruckus antenna matrix ing, beamforming, and so on), the most accurate classification can be switched combinatorially to the radio chains of a Wi-Fi of BeamFlex would probably be in the category of adaptive chipset, the number of possible patterns on each chain is 2n. antennas. The statistical optimization engine that powers its superior performance is also managing a number of other vari- ables at the system level, including MCS selection and power control, so it is about more than just antenna adaptation itself. Page 9 Marketing v. Reality in Multi-Antenna Wi-Fi

FIGURE 9: Full range of patterns of constructive signal combination that can be achieved through phase-based beamforming, with two or three transmit antennas in a typical Wi-Fi AP’s configuration (other patterns not shown are simple mirror images of these across a vertical axis).

Two Antennas

Three Antennas

FIGURE 10: Sample of patterns of constructive signal combination that can be achieved through Ruckus BeamFlex (total varia- tions available = 2n, where n = the number of elements in the AP’s antenna matrix) 1 2 3 4 5 6 2n

• • •

The asymmetry of these patterns provides very significant FIGURE 11: Typical BeamFlex antenna system pattern and benefits when you look at a WLAN with many access points. inherent interference reduction. As Figure 11 shows, a typical BeamFlex pattern has as much dBi Client as 10 to 15 dB of inherent self-interference suppression over 10 5 Typical more than half of the total coverage area. As a result, Ruckus 0 -5 BeamFlex APs tend to be better neighbors of each other in a network -10 Pattern than is the case for conventional approaches. -15 -20 -25 Compatibility with spatial multiplexing -30 BeamFlex is the industry’s only multi-antenna approach that can support both spatial multiplexing and constructive signal combi- 10–15 dB nation at the same time using only two transmit radio chains. An interference mitigation over more example of how this works is shown in Figure 12 below. than 180° Page 10 Marketing v. Reality in Multi-Antenna Wi-Fi

FIGURE 12: The unique ability in BeamFlex to use spatial multiplexing and coherent signal combination gains at the same time.

1 Two-stream encoding 2 as in Figure 5. 3 4 Rx 1 Tx RF 5 6 7 Decoding as in Figure 5. Tx RF 8 9 Rx 2 10 11 BeamFlex engine 12 Better, and balanced, Higher signal strength assigns streams to coherent combination enables higher modulation antennas selected of both signals on both class (bit rate) on both for best signal Rx antennas streams, for higher 802.11n propagation MCS

5. A proper multi-antenna processing taxonomy 6. On to what matters most: performance To summarize the multi-antenna technologies we’ve reviewed As we’ve developed a baseline understanding of how these here, and to set the stage for our final assessment sections 6 and things work, we’ve noted a few characteristics of the various 7, we present here a thorough taxonomy of current approaches. multi-antenna processing approaches that affect their performance. Now we pull these observations together, along with Attributes Implicit Explicit BeamFlex Beamforming Beamforming external validation, to quantify these technologies’ typical per- 802.11 protocols a, b, g, n n a, b, g, n formance gains in real networks. We’ll discuss the key metrics supported and validation for each of the three categories in turn, and then adaptation effectively closed loop closed loop assemble a depiction of their relative performance gains in a open loop (uses guesswork complete rate v. range comparison. based on uplink) client behavior none must send AP none requirement transmit charac- Implicit Beamforming teristics ‘recom- mendation’ A small number of vendors have embraced the chip-based ap- source of measurement client’s recom- client ACK proaches to beamforming over the past year, largely just to add feedback of uplink signal mendation packet on from client previous “beamforming” to their marketing materials, since we’ve done transmission so much to popularize the idea. Given that the explicit version supports 802.11n NO* NO* YES requires client functionality that has yet to reach the market, spatial multiplexing all but one recent entrant into the beamforming race are using typical perfor- none 2 dB 6 dB implicit beamforming. As we saw in section 2, implicit beamform- mance gain (see section 6) ing suffers from fundamental flaws: the absence of any corrective network self- NO* NO* YES feedback about whether or not a set of antenna phase decisions interference reduction has been effective at all, the reliance on uplink characteristics to vendor none 2 dB 6 dB estimate the downlink (an unreliable metric), incompatibility with examples 802.11n spatial multiplexing rates (without adding impractical network self- none none 10–15 dB interference numbers of RF chains), and patterns of coherent combination that reduction vendor Cisco, Meraki HP Ruckus examples

*would require 2 or more radio chains and antennas per spatial stream, a configuration no commercial AP supports today Page 11 Marketing v. Reality in Multi-Antenna Wi-Fi

are both sensitive to phasing inaccuracies and very symmetric expectations for the technical value of implementing the explicit (generating more concentrated co-channel interference to neigh- beamforming part of the 802.11n standard. boring APs). 2. In fact, their own lab testing has shown that the technique is As a result, it’s reasonable to expect at best modest performance only marginally more effective than implicit beamforming — gains. In fact, results we’ve seen from 3rd-party testing (see yielding gains that range from a fraction of a dB to at most 2 dB. Figure 13) suggest that the gains can be closely approximated 3. It follows naturally that commercial implementation has always by the number 0. We exclude implicit beamforming from further been a low priority for the chip vendors, and its entry into chips analysis here for this reason. now was motivated almost exclusively by pressure from their larger AP vendor customers who wanted to add the capability to FIGURE 13: Example third-party comparative test results for their sales story. implicit beamforming

For the purposes of our quantitative comparison in the balance of Zap UDP Mb/sec, 2.4 GHz Location 2 this paper, we will give our esteemed competitors a little benefit

Ruckus 7962 107.63 of the doubt and assume the chipset vendors’ figure of 2 dB in performance gain is a reasonable scenario. Cisco Aironet LAP1142N w/ BF 72.9

Cisco Aironet LAP1142N w/o BF 69.85 BeamFlex

Aruba AP125 43.4 We have a large number of external validation points from cus-

0 20 40 60 80 100 120 tomers and other 3rd parties that indicate our APs perform about Minimum @ 50% (average throughout) twice as well as conventional APs from any of the others. This can

Zap take the form of 2x the throughput for a given client distribution, UDP Mb/sec, 2.4 GHz Location 5 or 2x the coverage. This kind of performance improvement can Ruckus 7962 35.775 most easily be summarized as a 6 dB gain in link budget, averag-

Cisco Aironet LAP1142N w/ BF 23.7 ing across many different situations.

Cisco Aironet LAP1142N w/o BF 25.875 Putting it all together Aruba AP125 0.85 We summarize these performance perspectives in Figure 14 on 0 5 10 15 20 25 30 35 40 Minimum @ 50% (average throughout) the next page. To address a common first reaction to this kind of chart: for those of you wondering what happened to the 450 Mbps rate that 3-stream spatial multiplexing is advertised to Explicit Beamforming deliver, please note the conditions assumed for this comparison. At this writing, just one vendor has brought a new AP to market The peak 3-stream rate of 450 is raw bits in a 40 MHz wide 5 GHz based on the new generation of Wi-Fi chipsets that include ex- channel at 400 ns guard interval. We prefer to frame the analysis plicit beamforming. Since there are no client devices that support in a domain relevant to a real world still well populated with 2.4 this technology yet, we have not seen any performance testing GHz-only clients, and one in which actual usable packet through- from third parties. put (net of 802.11 protocol overhead) is of more interest. So our We do have close working relationships with the chipset vendors analysis depicts UDP downstream traffic in 2.4 GHz with an 800 who are enabling this next step in phase-based beamforming. In ns guard interval, which cuts the 450 down to 195 because of the conversations far from the limelight of AP vendor marketing ef- shift to a 20 MHz channel. Recall Figure 7. The final adjustment forts, the engineering teams at the Wi-Fi chipset suppliers (many to the peak rate of about 170 on the chart is the shift from gross of whom have long histories in the area of multi-antenna process- bits (195) to UDP packets — reflecting some “tax” paid for 802.11 ing techniques and therefore credible views on the subject) have overhead. The other dimension of realism is client to AP range — provided us the following information: we’ve assumed here ETSI requirements for EIRP (100 mW) and a healthy dose of interference and fading margin for busy indoor 1. For the same reasons we’ve outlined above (incompatibility with spatial multiplexing, limited beam-shaping degrees of free- dom when you have at most three antennas, and the inaccuracy- sensitivity of the narrow beams thus formed), they have low Page 12 Marketing v. Reality in Multi-Antenna Wi-Fi

conditions (15 dB). As most everyone has experienced, through- decode the two streams effectively. The performance of an AP put declines, and quite quickly, as a function of distance from the with implicit or explicit beamforming capabilities would look A P. exactly the same as this curve, because of the mutual exclusiv- ity of spatial multiplexing and phase-based beamforming. FIGURE 14: Rate and range comparison of various multi- antenna technologies. See text for additional explanatory [4] 802.11n 3x3:3 omni. This adds a 3rd transmit antenna to notes and sources. curve [3]. Note the rapid convergence of the 3x3 system’s RF Technology Comparison Peak Throughput, Mbps performance with that of the 2x2 system. The higher modula- 180 tion classes (based on 64 QAM, a very complex constellation)

160 in combination with 3 spatial streams are only effective, in practice, at very short ranges. 140 [1] .11n 1x1:1 omni [2] .11n + Explicit Beamforming (1 stream) 120 [3] .11n 2x2:2 omni [5] 802.11n 1x1:1 + BeamFlex. This adds the 6 dB BeamFlex [4] .11n 3x3:3 omni 100 link budget gain to curve [1], showing a roughly 2x increase in [5] .11n 1x1:1 + BeamFlex either throughput or range. 80 [6] .11n 2x2:2 + BeamFlex

60 [6] 802.11n 2x2:2 + BeamFlex. This adds the 6 dB BeamFlex link budget gain to curve [3], yielding a similar 2x rate or range 40 increase. 20 To net this all out in terms of the impact of a vendor selection 0 0 5 10 15 20 25 30 35 40 45 50 made today over the life of an organization’s WLAN, we need Range, m to look at one more piece beyond the AP — the evolution of Notes on conditions: 2.4 GHz, 20 MHz channel, 800 ns guard interval, ETSI EIRP level, downlink UDP traffic, medium level of Wi-Fi and other interfer- client capabilities over time. ence, near LoS link conditions, indoors.

7. Client capabilities over time matter, too Now to explain each of the curves: More often than not, an AP’s throughput is limited by what [1] 802.11n 1x1:1 omni. This is the baseline, depicting the kind can be supported by its clients. This is especially true for the of performance expected from a conventional omni-equipped explicit beamforming and three-stream SM modes, which have AP with no multi-antenna processing, communicating with a yet to reach the client base. For the purposes of analyzing AP single-antenna client like a smartphone. productivity over time, we’ve assembled a rough forecast of the market penetration of different client capabilities, based on [2] 802.11n 1x1:1 + Explicit Beamforming. As we explained in feedback from our chipset suppliers and various analyst views section 2, the addition of explicit beamforming on a 3-antenna (see Figure 15 on the next page). AP (the only model extant today) provides the system the choice of using explicit beamforming (EB) or spatial multiplex- One theme that runs across each of the three categories we ing, but not both. With only 2 dB of incremental gain to offer, depict is the timing of introduction of EB-enabled devices. EB does not provide enough of a performance benefit to out- While the Wi-Fi chipset supply community is now releasing weigh the data rate gains from SM in any case other than that the capability in the latest generation of their products, it will for which SM is not an option, the 1x1:1 system. As can be seen, take some time for the handset, tablet, and laptop vendors to 2 dB doesn’t buy much, in terms of either throughput increases incorporate them into their devices. The larger uncertainty is or range extension — roughly a 20% increase. actually the timing for formal interop and certification pro- gram development at the Wi-Fi Alliance. Because of the very [3] 802.11n 2x2:2 omni. This reflects the two spatial streams marginal benefit of the technology, some chip suppliers have performance of an AP equipped with 2 omni antennas commu- expressed doubts that the WFA timeline will be short, since the nicating with a 2 Rx equipped client such as a laptop. Note that technical challenges of proving in multivendor interop for new the performance of the 2x2:2 system converges to that of the 1x1:1 system at longer range, as the realities of RF propagation reduce the ability to achieve spatial multiplexing with reliability — at longer ranges there isn’t enough spatial diversity at Rx to Page 13 Marketing v. Reality in Multi-Antenna Wi-Fi

behaviors introduced at very low levels in the protocol (and FIGURE 15: Scenario-based forecast of client capabilities over the feedback required from clients in EB certainly qualifies as time. such) are substantial. For purposes of illustration here, we have Smartphones 100% assumed a scenario where these matters will be sorted out on legacy a fairly optimistic timeline — by the end of 2011 — and that :1n the progression to widespread adoption in mobile devices will :1n EB begin in 2012. These figures could easily be 6 if not 12 months :2n EB too optimistic. 0% 2011 2012 2013 2014 2015

Smartphones Tablets 100% The vast majority of smartphones today are single-stream :1n devices. A small number of legacy (801.11b/g) protocol devices :2n persist in the market, but those are expected to be aged out :1n EB of relevance quickly, given the typically short lifespan in this :2n EB category. Power and space constraints will prevent much head- 0% way for 2-stream models, but consumer pressure for higher 2011 2012 2013 2014 2015 throughput over time will eventually lead to some adoption Laptops of 2x2:2 architectures. The presence of EB in the chipsets will 100% legacy cause this technology to diffuse quickly, mostly because of :1n short device lifespans rather than inherent merit. :2n :2n EB :3n EB Tablets 0% Like smartphones, the majority of tablet devices today are 2011 2012 2013 2014 2015 1x1:1. Because of relatively lower pressure from battery and Sources: Wi-Fi chipset vendor expectations, various analysts. form factor constraints, there will be a growing segment of 2x2:2 architectures in the tablet category, gaining EB capabilities within the planning period as lifespans are short here, too. FIGURE 16: WLAN lifetime impact of BeamFlex v. 3x3:3+EB purchase decision.

Laptops 4-Year Aggregate Client Throughput per AP, Terabytes Since the laptop installed base is retired more slowly, the Ruckus 3x3:2 AP diffusion of EB capabilities into this category will take a little with BeamFlex 440 longer. Relatively less severe constraints on space and power have already given two-stream architectures a head start here Generic 3x3:3 AP — the introduction of three-stream APs will be accompanied with Explicit 260 Beamforming by three-stream support on laptops, too, but on the relatively longer timeline for laptop replacement. more capacity 70% for client service with Ruckus So what’s our point? point highlighted in the figure) with the client capabilities The last step in the analysis here is to assess the relative merits roadmap in Figure 15 yields the total aggregate traffic sup- of a couple of relevant purchase decisions made today, given ported comparison in Figure 16. Given the qualitative narrative both the device futures we’ve depicted as well as the perfor- we’ve presented here, it should come as no surprise that the mance characterization of APs working with these different de- quantitative reality of the 3x3:3 + EB story trails far behind its vice types. To illustrate the trade-off, we consider the options marketing. EB’s performance gain is minimal, and limited to of building a WLAN network based on either (a) Ruckus 3x3:2 single-stream devices that support the feedback protocol, of APs with BeamFlex, or (b) 3x3:3 APs equipped with explicit beamforming. Assuming a client mix of 50% smartphones, 20% tablets, and 30% laptops over the next four years, combining the performance profile in Figure 14 (taken at the 10m range Page 14 Marketing v. Reality in Multi-Antenna Wi-Fi

which there are unlikely to be very many until late in the analysis 3. There won’t be very many clients around to take advantage period. Likewise for 3-stream spatial multiplexing. The net re- of explicit beamforming or 3 spatial streams for a long time sult is a Ruckus 2x3:2 AP with BeamFlex will provide 70% more Meanwhile, Ruckus BeamFlex continues to offer a 2x perfor- capacity for client service over the life of a WLAN. mance advantage over other APs.

8. In conclusion Despite competitor marketing to the contrary, the Ruckus solu- We’ve been through lots of details here. Let’s recap the story, tion is 100% standards compliant, works with all Wi-Fi clients to make sure things are perfectly clear. today, and can be used directly with spatial multiplexing to give the benefits of both SM and beamforming, but without Other AP vendors are piling on the “beamforming” band- anywhere near as much concentrated co-channel interference wagon, since we’ve demonstrated how valuable multi-antenna introduced into the network. approaches to Wi-Fi can be. Summed over the life of a WLAN, we can provide 70% more There are issues with their approaches, though: aggregate throughput per AP than the best of the rest.

1. Implicit beamforming provides effectively zero performance We think the right choice for your WLAN is obvious. Please do benefit in practice. not hesitate to contact us for more details.

2. Explicit beamforming (new to the scene) does a little better, Go to www.ruckuswireless.com for more information. but just like its poorer implicit cousin, it’s mutually exclusive with the essential spatial multiplexing modes in 802.11n — with 2 or 3 radio chains and antennas, you can do phase-based beamforming, or spatial multiplexing, but not both

Ruckus Wireless, Inc. 880 West Maude Avenue, Suite 101, Sunnyvale, CA 94085 USA (650) 265-4200 Ph \ (408) 738-2065 Fx

Copyright © 2011, Ruckus Wireless, Inc. All rights reserved. Ruckus Wireless and Ruckus Wireless design are registered in the U.S. Patent and Trademark Ofﬁce. Ruckus Wireless, the Ruckus Wireless logo, BeamFlex, ZoneFlex, MediaFlex, FlexMaster, ZoneDirector, SpeedFlex, SmartCast, and Dynamic PSK www.ruckuswireless.com are trademarks of Ruckus Wireless, Inc. in the United States and other countries. All other trademarks mentioned in this document or website are the property of their respective owners. 803-71273-001 rev 01