marketing v. reality in multi-antenna wi-fi...page 3 marketing v. reality in multi-antenna wi-fi...
TRANSCRIPT
Marketing v. Reality in Multi-Antenna Wi-Fi
Ruckus Wireless | White Paper
Executive Summary A number of Wi-Fi vendors have recently begun to market
beamforming features, attempting to follow our lead in the
field. As rapidly growing numbers of mobile devices and
increasing bandwidth demands hit wireless LANs everywhere,
it’s no surprise that customers and suppliers alike are focusing
more attention on the fundamental radio performance of their
networks.
Most new vendor marketeer converts to beamforming are
unfortunately glossing over (as marketeers will do) a number
of critical limitations in the chip-based beamforming their APs
offer, including the absence of current client support for the
standard, incompatibility with the higher data rates in 802.11n
that require spatial multiplexing, and at best only small perfor-
mance gains.
The problem is that this marketing gloss is beginning to con-
fuse those responsible for WLAN purchase decisions, who are
typically not radio-frequency engineering specialists and there-
fore unable to see past the gloss to make adequately-informed
choices. As we’ll show, not all multi-antenna techniques in Wi-Fi
are equally capable of solving the basic WLAN challenge — de-
livering fast, reliable connectivity to many clients in challenging
environments.
This paper goes back to the basics of how radio signals work in
Wi-Fi systems, explaining the functions multiple antennas can
serve, using plain English, lots of pictures, and only as much
trade jargon as necessary, defined carefully and clearly. Using
A Back-to-the-Basics Guide to 802.11 Beamforming for Discerning Buyers
this firm foundation of the basic principles, we then compare
and contrast the main varieties of multi-antenna techniques
employed in current Wi-Fi systems, concluding with a look at
real-world performance comparisons.
We don’t want to spoil the ending, in part because understand-
ing the results we show in the wrap-up comparison below
requires you to go through all the building blocks to see why
this is the case. But we do want to emphasize up front that the
network performance differences between the WLAN technol-
ogy choices you face can be very large, and they can have an
equally large impact, for better or for worse, on the experience
of your WLAN users. So please read on, and let us help you
become a better-informed buyer.
Ruckus 3x3:2 APwith BeamFlex
4-Year Aggregate Client Throughput per AP, Terabytes
Generic 3x3:3 APwith ExplicitBeamforming
440
more capacity for client servicewith Ruckus
260
70%
WLAN lifetime purchase-decision impact of BeamFlex adaptive antennas v. conventional 3x3 AP with generic beamforming: 70% difference in client service capacity.
Page 2
Marketing v. Reality in Multi-Antenna Wi-Fi
1. IntroductionIt’s often said that imitation is the sincerest form of flattery. At
Ruckus, we are indeed flattered that so many in the WLAN indus-
try have embraced the idea of what is commonly called beam-
forming, since we have for several years now been pioneering the
application of multi-antenna technology in Wi-Fi. In fact, we have
built our entire business on the performance, ease of use, and
cost-effectiveness gains this technology can provide, when done
right. Our rapidly growing community of customers is enjoying
the substantial benefits our approach to shaping radio energy in
their networks brings every day, in terms of higher client through-
put rates, better coverage, more resistance to interference, and
much more reliable Wi-Fi performance.
For organizations currently shopping for new WLAN gear,
on the other hand, many vendors’ approaches to the whole
subject of beamforming are rather less than helpful. As is all too
common in the wireless industry, marketing departments have
gotten a little out of control and are promoting “beamforming”
capabilities that have little or no connection to the real world of
Wi-Fi radios, planting the seeds for significant customer confu-
sion and disappointment in the process. As we’ll show in the
following material, it’s easy to imitate beamforming messages
in pretty brochures (with several vendors stretching the truth
while they’re at it), but another thing entirely to deliver genuine
and material performance gains in day to day operation.
We’ve put together this brief paper to help clear up some con-
fusion about how different versions of multi-antenna systems
work, what kind of performance gains have been validated for
them in the real world, and when they will be commercially rele-
vant given chipset adoption and interop certification timelines.
2. What’s this “beamforming” thing again?This will take some explanation. One of the reasons for ven-
dor 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-
ence in the arcana of advanced radio-frequency engineering
— in other words, the majority of you. We certainly don’t mean
that observation as a slight to any of you, since it’s just fine with
us that you leave the gnarly technical bits of this area to the ex-
perts. No one would expect you to engineer your own compact
fluorescent light bulbs or LCD televisions, either. The downside,
though, is that it’s easy for vendors to spew nonsense about RF
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
really understand this stuff, either), and certainly with little
chance of savvy customers calling the technical foul.
Since all this stuff matters so much to the performance of your
network, and the experience of your users, we’re ready to take
up the challenge of getting you enough knowledge to make
good Wi-Fi network design decisions, however. To build the
foundation required to accurately assess claims and likely
performance benefits for multi-antenna systems, we’re going
to go back to the basics here, using lots of pictures and defin-
ing carefully the necessary jargon along the way to try to help
make things very clear.
We start with an old-fashioned single-antenna access point,
shown in Figure 1 (below), with a common omni-directional an-
tenna, or an “omni”. When this device transmits, as the anten-
na’s name suggests, it sends the same signal in all directions.
While this approach has a certain satisfying design simplicity, it
has substantial performance disadvantages. The vast majority
of this radio energy is completely wasted, since an access point
can only talk to one client at a time. This excess energy creates
self-interference in the WLAN, stepping on neighboring APs
and their clients and reducing the possibility of channel reuse
nearby. Meanwhile, the tiny fraction of transmit energy that
actually reaches the client yields a lower throughput rate, as
we’ll show shortly, than would the case if the energy could be
focused more tightly (since client throughput is directly related
to available signal strength).
Next we introduce another omni antenna to begin to explore
the tools this might provide us for better control of the radio
signal. As shown in Figure 2 (see next page), the combination
of two copies of the same signal transmitted from two neigh-
boring omni antennas creates a set of intersecting troughs
and peaks, much like the wave rings that might be achieved by
tossing two separate rocks into a still pond at the same time.
In some locations, the peaks of the signal from transmit
FIGURE 1: Radio signal distribution pattern from an access point with one omni-directional antenna.
Omni Transmit (Tx)Pattern
Page 3
Marketing v. Reality in Multi-Antenna Wi-Fi
antenna 1 (“Tx 1”, in the jargon) line up in space and time with
the peaks from Tx 2 — this is referred to as constructive com-
bination. In other locations, the peaks of Tx 1’s signal are lined
up with the troughs of signal from Tx 2, which yields destruc-
tive combination. If a receive (Rx) antenna is placed in the zone
of perfect constructive combination, it would pick up roughly
twice the signal strength of a single Tx antenna’s output, with-
out doing any intelligent work on its own — its analog receive
electronics simply sum the signals received automatically. In
contrast, a zone of complete destructive combination would
yield zero signal, a phenomenon useful in reducing intra-AP in-
terference at the network level (more on this later). The repeat-
ing patterns in radio communication signals allow us to use
the concept of phase to describe the peak or trough match-up
relationship between two different signals.
One way in which the phenomena of constructive and destruc-
tive combinations of multiple transmit sources can be put to
productive use is through the addition of active control of indi-
vidual transmit signal phases. This is the narrow (and most tech-
nically correct) definition of the term beamforming, and the
type of multi-antenna processing that is coming into fashion in
a growing number of vendors’ Wi-Fi products. (This is not what
Ruckus products do today, which we’ll explain in section 4.) In
beamforming systematic manipulation of the phase of signals
transmitted from multiple antennas is used to place zones of
constructive combination that fall ideally at the location of the
client of interest. We illustrate this in Figure 3. The depiction of
the transmission pattern has been cleaned up here in order to
show only the areas of constructive combination, which is the
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”)
ConstructiveCombination
Tx Antenna 1
In Phase
180º Outof Phase
Tx Antenna 2
Time2x signal
Signal strength
ReceiveAntenna (Rx)
DestructiveCombination
No signalTx 1
Tx 2
Rx
FIGURE 3: How signal phase control works in beamforming.
DestructiveCombination
Client antenna
Tx 1
Tx 2
ConstructiveCombination
Phase adjustment per antenna to directlocation of constructive combination on client
Rx
Page 4
Marketing v. Reality in Multi-Antenna Wi-Fi
tially the equivalent of geographic contour maps, where height
in this case is signal strength. You can see where the term
beamforming arises, since the resulting patterns tend to have
lobes of constructive combination areas that look somewhat
like “beams” of energy shining out from the antenna array,
much like the beam of a flashlight, that are “formed” by the
system controlling the individual phases of the antennas. “Con-
trolling phase” in this context means essentially “changing
when you start transmitting.” Outside the lobes of the pattern
as drawn are areas of destructive combination.
Phase is adjusted by the system to compensate for different
travel times between each antenna and the client of interest,
so that the signals from Tx 1 and Tx 2 arrive at Rx with their
peaks aligned in time, maximizing signal strength at the client.
So far, so simple. Things need to get a little more interesting to
assess with clarity what’s being used in Wi-Fi today.
A key underlying assumption in beamforming is that both Tx 1
and Tx 2 are transmitting the same signal. To understand why
that is important, we need a short word on the nature of the
signals themselves. So far we’ve been using simple sinusoidal
curves to depict our wireless signals, so they may appear to be
just generic energy levels, and it might not be obvious why one
segment of the orange curve couldn’t be combined at Rx with
any other. The signal-shape reality of today’s encoded wireless
transmissions is much more complex. In order to achieve high
throughput, many bits are transmitted at the same time on a
single signal “wave”, in a format called a constellation, where at a
single snapshot in time each bit holds a particular place in a ma-
trix in the real and imaginary number space. Fortunately for the
many of you we’ve just lost with that last sentence, this particular
batch of complexity is not important to understand in the con-
text of evaluating multi-antenna processing technology. For sake
of keeping things as simple as possible, we’ll show “real” wireless
signals through squiggles we borrowed from the audio electron-
ics world, in order to illustrate some key concepts.
We put this into practice first in Figure 4. Remember that the
receive processing expected of the client in a phase-based
beamforming system is just simple summation of the signals re-
ceived at any given time. Figure 4 illustrates visually a point we
can also make through a music analogy: if you had two audio
speakers side by side blasting two different tunes (say, Bach’s
Brandenburg Concerto #1 and Black Sabbath’s Iron Man) at the
same time, what you’d hear would be essentially just noise. If
they were both playing the same tune at exactly the same time,
you’d hear a louder version of the tune.
The other key underlying assumption in beamforming is that
the system knows where the client is, in an RF-signal-path
sense of the term “where”, in order to choose phase adjust-
ments to point or “steer” one of its beams in the right direc-
tion. In the Wi-Fi world there are two different approaches
being used for educating the AP about client “location”:
Implicit Beamforming
In the normal course of communication from client to AP, the
AP can detect from its multiple antennas the different phases
of arrival of a signal from the client on each of the AP’s anten-
nas. This is roughly analogous to the way human ears process
sounds that arrive at each ear at different times and therefore
give an indication of the direction from which the sound came.
It’s worth noting that for the same reason our ears can be de-
ceived by sound bouncing off acoustically reflective surfaces,
the AP’s impression of the client achieved by measurement of
signal arrival phase differences is not a terribly reliable indica-
tion of actual physical location — because of signal reflection
off surfaces in the environment in which the AP and client are
operating. In implicit beamforming the phase differences are
used as just that — phase differences that should be applied to
the AP’s transmit antennas to achieve maximum constructive
combination on the next transmit to that client.
FIGURE 4: Phase-based beamforming can only be used with copies of the same signal.
Tx 1
In Phase
Tx 2
Garbage
Different waveforms
Tx 1
In Phase
Tx 2
2x signal strength
Same waveform
Rx Rx
Page 5
Marketing v. Reality in Multi-Antenna Wi-Fi
The flaw in using the radio-space characteristics of the uplink
from client to AP as a model for what should be used to ma-
nipulate signals in the downlink is that signal behavior can dif-
fer substantially between the uplink and downlink paths. We’ll
show quantitatively in our subsequent section on performance
assessment that implicit beamforming really just doesn’t work
very well, for primarily this reason.
Explicit Beamforming
To improve on the poor performance of implicit beamform-
ing, the alternative that is just now coming into infrastructure
products in Wi-Fi involves communication from the client to
the AP of what would work best for the client (in terms of the
AP’s transmit phase and other settings), given the client’s cur-
rent vantage point in the radio space. No Wi-Fi clients on the
market today support this feature, however, so this approach
faces some substantial commercial challenges for broad adop-
tion. More technically, while it certainly improves the quality of
the AP’s understanding of the characteristics of the best radio
path to the client, it remains subject to a significant handicaps
endemic to small-antenna-count beamforming in the context
of Wi-Fi networks. We elaborate on the nature of these chal-
lenges in the performance section.
One final note on this category: since the manipulation of
signal phase must be done at the PHY layer (at the lowest level
of hardware), both implicit and explicit beamforming function-
ality must be built into the Wi-Fi chipset. For this reason, these
techniques are often referred to as “chip-based beamforming”
in the industry.
3. OK, got beamforming. Now what’s “spatial multiplexing” ?Phase-based beamforming was actually the simpler story. The
more complex topic that is more essential to the high data rates
built into the 802.11n Wi-Fi protocol is spatial multiplexing, or SM.
In the less technically-minded segments of the Wi-Fi community
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
cleaning this one up at the outset with a couple definitions:
MIMO = an acronym for “multiple input, multiple output”. De-
fined from the vantage point of the air between an AP and a cli-
ent, the term refers simply to a system design where more than
one transmit antenna put signals into the air (multiple input),
and where more than one receive antenna extracts these sig-
nals from the air (multiple output). MIMO systems can be used
to do a variety of different forms of multi-antenna processing,
but not necessarily all at the same time, as we’ll discuss below.
SM = spatial multiplexing. A system in which a single flow of
bits is broken up into two or more streams that are transmit-
ted into the air from Tx antennas separated in space (with one
stream per antenna). These multiple streams are then extracted
from the air by the same number of Rx antennas, also sepa-
rated in space, and then recombined into the original single bit
flow. SM requires a MIMO antenna architecture, i.e. multiple
antennas on both client and AP. The point of the exercise is to
get bits over the air interface faster by adding additional paral-
lel “lanes” for traffic on the same spectrum.
Figure 5 illustrates how this works from a radio signal perspec-
tive. It may appear at first glance to break the rules for signal
FIGURE 5: How spatial multiplexing works.
Tx 1
Tx 2
Different coded waveforms (on same frequency)
Matrix decode,not simple sum
Data stream split
and cross-encoded for Tx
...0010
...1011
...0010
...1011Rx 1
Air, with
spatial diversity
or
Rx 2
Data stream recombined...1101001101 ...1101001101
Page 6
Marketing v. Reality in Multi-Antenna Wi-Fi
combination we introduced in section 2, since the blue and
orange signals received on each client antenna look like our
metaphorical Bach and Black Sabbath arriving at the same ear-
drum, just creating noise. The details of exactly how SM signals
are coded on transmit and decoded on receive go beyond the
scope of this paper (and would require reviving cool matrix-
math tricks you learned in the undergrad linear algebra class
you’ve certainly forgotten by now). For our purposes here we
can just say that the combination of special pre-encoding and
spatial diversity (signals differing based on where in physical
space they are received) are leveraged in receive processing to
disentangle the streams that got combined in the air between
Tx and Rx. Note that this technology can be used by clients to
send multiple streams to APs as well. Also note that without
spatial diversity between the streams, decoding fails.
One Wi-Fi vendor has recently introduced APs with three anten-
nas capable of both transmit and receive (a “3x3:3” architecture
in Wi-Fi parlance, which means 3 transmit, 3 receive, and capabil-
ity for 3 spatial streams). This vendor also claims performance
gains from the explicit beamforming processing that has just
been implemented in the new generation of Wi-Fi chipsets.
We’ll address those alleged performance gains quantitatively in
a moment, but first this product form presents a logical problem
related to spatial multiplexing that we need to explore.
As we’ve shown, spatial multiplexing requires that Tx antennas
produce different signal waveforms in order for the system to
code and decode the multiple streams, while phase-based
beamforming requires the transmission of multiple copies of
the same signal waveform. With two transmit antennas (and
two receive antennas on the other end of the air link, to com-
plete the MIMO requirement for SM), it’s obvious that a system
can do spatial multiplexing or phase-based beamforming, but
not both. The purpose of a third transmit antenna introduced
on a conventional AP is unclear in a world filled with at best
two-antenna clients: what are you going to do with it, exactly?
In Figure 6 we illustrate the scenario that gives us pause. If you
use two of your three Tx antennas for SM in order to achieve
the substantial data rate gain it can provide, you have the
choice of either leaving the third Tx antenna quiet, or doing
something active with it. You can’t do more SM, since you only
have two antennas on the client (a third stream would required
a third client Rx antenna). What happens if you try to do phase-
based beamforming with it? Let’s say, for sake of argument,
that you replicate the second of the two encoded streams
and transmit it on the third antenna. Let’s also assume (and
this is a generous assumption that would be nearly impossible
to achieve in practice) that you can adjust phase on the third
antenna so that the simple sum of signals from Tx 2 and Tx 3
result in a higher-amplitude version of stream 2 arriving at both
Rx 1 and Rx 2 with the same diversity that would have been the
case if Tx 3 wasn’t involved at all. The SM decoding process on
the client would then yield a normal version of stream 1 and a
stronger-signal version of stream 2. So what? If you anticipated
this result and tried to use a higher modulation class (i.e. higher
bit rate) on stream 2’s signal, you would break the process of
multiplexing and demultiplexing the original bit stream, since
it’s essentially a simple round-robin algorithm that assumes the
bit rates of the two multiplexed streams are the same. If, on
the other hand, you leave the modulation class alone, then you
have a lot of extra signal strength on stream 2 that services no
useful purpose for the target client, but that does spew more
co-channel interference into the surrounding network.
FIGURE 6: How SM fails in combination with phase-based beamforming with 3 Tx antennas.
...0010
Tx 1
Tx 2
Duplicated forbeamforming?
Weaker signal strength on stream 1 than on stream 2, so lower data rate?
Rx 1
Rx 2 Data stream recombinationout of sync
...?0?0?1?01
... 1 1
Tx 3
...0010
...0010
...1011
...1101001101
or, alternatively, extra signal strength in stream 2 goes unused for higher throughput, causing unnecessary and harmful self-interference in thenetwork
Page 7
Marketing v. Reality in Multi-Antenna Wi-Fi
802.11 PHY Rates Overview
802.11b 802.11a/g 802.11n
Peak Bit Rate, Mbps 20 MHz channel
40 MHz channel
Modulation Bit Rate Mbps
Modulation Coding Bit Rate Mbps
MCS Index
Spatial Streams
Modulation Coding 800nsGI 400ns GI 800ns GI
400ns 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
FIGURE 7: Importance of spatial multiplexing to higher 802.11n bit rates.
1
10
100
1,000
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 23 31
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
Peak bit rate, Mbps
MCS number
and a graphical view:
Page 8
Marketing v. Reality in Multi-Antenna Wi-Fi
We’ll return to the question of the merits of a 3x3:3 architecture
in today’s market in section 7.
To put this question and our performance assessment section
6 in context, we repeat for convenience in Figure 7 the well-
known tabulation of bit rates defined for the various modula-
tion and coding schemes in the IEEE’s Wi-Fi standards series.
The highlighted part of the 802.11n table emphasizes that all
the more interesting rates require two or more spatial streams.
In other words, sensible 802.11 systems will bias their designs
toward maximizing the use of spatial multiplexing for clients
that can support it.
4. What does Ruckus BeamFlex™ do, then?While we often use the term “beamforming” for the multi-an-
tenna processing we do in our APs, as a matter of convenience
when we don’t want to have to get into a long discussion of
distinctions between different multi-antenna architectures,
that’s not an entirely accurate term for what we do, given the
strict definition we introduced in section 2. As we show in
Figure 8 below, our approach involves digitally switching a
selection from a large number of antenna elements to connect
with the individual radio chains in the RF front end of off-the-
shelf Wi-Fi silicon. A “radio chain” is the RF engineering term
for the analog radio part between the chip doing the digital Wi
Fi protocol processing and the antennas. In BeamFlex there is
no analog adjustment of phase on each radio chain. Instead,
an optimal combination of antennas is selected on a packet-
by-packet basis to focus patterns of radio energy in the right
direction based on the inherent characteristics of the antenna
elements themselves. The selection for a given client is based
on the throughput last achieved with that client, confirmed
through the ACK packet that is a standard part of the 802.11a,
b, g, and n Wi-Fi protocols and that is supported by all clients
today. [Note that some vendors like to make up for their lack of
intellectual property contributions in Wi-Fi by de-positioning
BeamFlex as “non-standard” or (gasp) “proprietary” — which
couldn’t be further from the truth. Our APs are absolutely 100%
compliant with the 802.11 protocols and require absolutely zero
special behavior on the part of clients.]
Of the many terms used in the general area of multi-antenna
processing techniques (such as smart antennas, beam switch-
ing, beamforming, and so on), the most accurate classification
of BeamFlex would probably be in the category of adaptive
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.
As with most true innovation, we’ve moved beyond existing
categorization, so we tend to just stick with BeamFlex.
There are two primary advantages in our ability to use a com-
bination of multiple antennas on individual Wi Fi radio chains:
better antenna patterns, and compatibility with spatial multi-
plexing. We’ll look at each in turn.
Better antenna patterns
Better than what? The baseline is the kinds of beam patterns
that can be created with phase-based beamforming — for
which, see Figure 9 on the next page. This approach is limited
to using only as many antennas as it has radio chains. With
only two or three antennas to work with, the shapes that can
be created are fairly limited in structure, and they are marked
consistently by a couple of substantial issues: they are symmet-
ric, and their lobes or beams tend to be relatively narrow. The
symmetry means that from the perspective of a target client,
half the energy transmitted is wasted. From the vantage point
of neighboring APs in the WLAN, this energy is worse than
wasted — it means louder co-channel interference. The narrow
beams mean they are pretty unforgiving about inaccurately
pointed beams. If an AP’s estimate of the right phase combina-
tion to use for its antennas is a little off (either because it was
using imperfect implicit feedback, or because the higher-qual-
ity explicit feedback forthcoming in Wi-Fi systems has gotten
out of date because of delays in its use, high client mobility,
or rapid changes in the environment like a door closing), the
beam formed will fall where the client isn’t, and an area of
destructive combination will fall where the client is, making the
whole exercise worse than useless.
Ruckus BeamFlex, in contrast, creates highly asymmetric pat-
terns that have much more forgiving lobe shapes, and with
huge variety across physical as well as polarization space — see
Figure 10. Because the n elements of a Ruckus antenna matrix
can be switched combinatorially to the radio chains of a Wi-Fi
chipset, the number of possible patterns on each chain is 2n.
FIGURE 8: High-level view of Ruckus BeamFlex architecture.
n antenna elements
RF
RF
RF PHY MAC Host
Off-the-Shelf 802.11 Silicon
Page 9
Marketing v. Reality in Multi-Antenna Wi-Fi
The asymmetry of these patterns provides very significant
benefits when you look at a WLAN with many access points.
As Figure 11 shows, a typical BeamFlex pattern has as much
as 10 to 15 dB of inherent self-interference suppression over
more than half of the total coverage area. As a result, Ruckus
APs tend to be better neighbors of each other in a network
than is the case for conventional approaches.
Compatibility with spatial multiplexing
BeamFlex is the industry’s only multi-antenna approach that can
support both spatial multiplexing and constructive signal combi-
nation at the same time using only two transmit radio chains. An
example of how this works is shown in Figure 12 below.
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).
Three Antennas
Two 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
FIGURE 11: Typical BeamFlex antenna system pattern and inherent interference reduction.
-30
-25
-20
-15
-10
-5
0
5
10 dBi Client
10–15 dB interference mitigation over more than 180°
Typical BeamFlex Pattern
Page 10
Marketing v. Reality in Multi-Antenna Wi-Fi
5. A proper multi-antenna processing taxonomyTo summarize the multi-antenna technologies we’ve reviewed
here, and to set the stage for our final assessment sections 6 and
7, we present here a thorough taxonomy of current approaches.
Attributes Implicit Beamforming
Explicit Beamforming
BeamFlex
802.11 protocols supported
a, b, g, n n a, b, g, n
adaptation effectively open loop (uses guesswork based on uplink)
closed loop closed loop
client behavior requirement
none must send AP transmit charac-teristics ‘recom-mendation’
none
source of feedback
measurement of uplink signal from client
client’s recom-mendation
client ACK packet on previous transmission
supports 802.11n spatial multi-plexing
NO* NO* YES
typical perfor-mance gain (see section 6)
none 2 dB 6 dB
network self-interference reduction
NO* NO* YES
vendor examples
none 2 dB 6 dB
network self-interference reduction
none none 10–15 dB
vendor examples
Cisco, Meraki HP Ruckus
* would require 2 or more radio chains and antennas per spatial stream, a configuration no commercial AP supports today
6. On to what matters most: performanceAs we’ve developed a baseline understanding of how these
things work, we’ve noted a few characteristics of the various
multi-antenna processing approaches that affect their perfor-
mance. Now we pull these observations together, along with
external validation, to quantify these technologies’ typical per-
formance gains in real networks. We’ll discuss the key metrics
and validation for each of the three categories in turn, and then
assemble a depiction of their relative performance gains in a
complete rate v. range comparison.
Implicit Beamforming
A small number of vendors have embraced the chip-based ap-
proaches to beamforming over the past year, largely just to add
“beamforming” to their marketing materials, since we’ve done
so much to popularize the idea. Given that the explicit version
requires client functionality that has yet to reach the market,
all but one recent entrant into the beamforming race are using
implicit beamforming. As we saw in section 2, implicit beamform-
ing suffers from fundamental flaws: the absence of any corrective
feedback about whether or not a set of antenna phase decisions
has been effective at all, the reliance on uplink characteristics to
estimate the downlink (an unreliable metric), incompatibility with
802.11n spatial multiplexing rates (without adding impractical
numbers of RF chains), and patterns of coherent combination that
FIGURE 12: The unique ability in BeamFlex to use spatial multiplexing and coherent signal combination gains at the same time.
1
2
3
4
5
6
7
8
9
10
11
12
Rx 1
Rx 2
Two-stream encodingas in Figure 5.
Decodingas in Figure 5.
BeamFlex engineassigns streams to antennas selectedfor best signal propagation
Higher signal strength enables higher modulationclass (bit rate) on both streams, for higher 802.11n MCS
Better, and balanced, coherent combinationof both signals on bothRx antennas
Tx RF
Tx RF
Page 11
Marketing v. Reality in Multi-Antenna Wi-Fi
are both sensitive to phasing inaccuracies and very symmetric
(generating more concentrated co-channel interference to neigh-
boring APs).
As a result, it’s reasonable to expect at best modest performance
gains. In fact, results we’ve seen from 3rd-party testing (see
Figure 13) suggest that the gains can be closely approximated
by the number 0. We exclude implicit beamforming from further
analysis here for this reason.
Explicit Beamforming
At this writing, just one vendor has brought a new AP to market
based on the new generation of Wi-Fi chipsets that include ex-
plicit beamforming. Since there are no client devices that support
this technology yet, we have not seen any performance testing
from third parties.
We do have close working relationships with the chipset vendors
who are enabling this next step in phase-based beamforming. In
conversations far from the limelight of AP vendor marketing ef-
forts, the engineering teams at the Wi-Fi chipset suppliers (many
of whom have long histories in the area of multi-antenna process-
ing techniques and therefore credible views on the subject) have
provided us the following information:
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
FIGURE 13: Example third-party comparative test results for implicit beamforming
43.4
69.85
72.9
107.63
0 20 40 60 80 100 120
Aruba AP125
Cisco Aironet LAP1142N w/o BF
Cisco Aironet LAP1142N w/ BF
Ruckus 7962
Minimum @ 50% (average throughout)
ZapUDP Mb/sec, 2.4 GHz
Location 2
0.85
25.875
23.7
35.775
0 5 10 15 20 25 30 35 40
Aruba AP125
Cisco Aironet LAP1142N w/o BF
Cisco Aironet LAP1142N w/ BF
Ruckus 7962
Minimum @ 50% (average throughout)
ZapUDP Mb/sec, 2.4 GHz
Location 5
expectations for the technical value of implementing the explicit
beamforming part of the 802.11n standard.
2. In fact, their own lab testing has shown that the technique is
only marginally more effective than implicit beamforming —
yielding gains that range from a fraction of a dB to at most 2 dB.
3. It follows naturally that commercial implementation has always
been a low priority for the chip vendors, and its entry into chips
now was motivated almost exclusively by pressure from their
larger AP vendor customers who wanted to add the capability to
their sales story.
For the purposes of our quantitative comparison in the balance of
this paper, we will give our esteemed competitors a little benefit
of the doubt and assume the chipset vendors’ figure of 2 dB in
performance gain is a reasonable scenario.
BeamFlex
We have a large number of external validation points from cus-
tomers and other 3rd parties that indicate our APs perform about
twice as well as conventional APs from any of the others. This can
take the form of 2x the throughput for a given client distribution,
or 2x the coverage. This kind of performance improvement can
most easily be summarized as a 6 dB gain in link budget, averag-
ing across many different situations.
Putting it all together
We summarize these performance perspectives in Figure 14 on
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
deliver, please note the conditions assumed for this comparison.
The peak 3-stream rate of 450 is raw bits in a 40 MHz wide 5 GHz
channel at 400 ns guard interval. We prefer to frame the analysis
in a domain relevant to a real world still well populated with 2.4
GHz-only clients, and one in which actual usable packet through-
put (net of 802.11 protocol overhead) is of more interest. So our
analysis depicts UDP downstream traffic in 2.4 GHz with an 800
ns guard interval, which cuts the 450 down to 195 because of the
shift to a 20 MHz channel. Recall Figure 7. The final adjustment
to the peak rate of about 170 on the chart is the shift from gross
bits (195) to UDP packets — reflecting some “tax” paid for 802.11
overhead. The other dimension of realism is client to AP range —
we’ve assumed here ETSI requirements for EIRP (100 mW) and a
healthy dose of interference and fading margin for busy indoor
Page 12
Marketing v. Reality in Multi-Antenna Wi-Fi
conditions (15 dB). As most everyone has experienced, through-
put declines, and quite quickly, as a function of distance from the
AP.
Now to explain each of the curves:
[1] 802.11n 1x1:1 omni. This is the baseline, depicting the kind
of performance expected from a conventional omni-equipped
AP with no multi-antenna processing, communicating with a
single-antenna client like a smartphone.
[2] 802.11n 1x1:1 + Explicit Beamforming. As we explained in
section 2, the addition of explicit beamforming on a 3-antenna
AP (the only model extant today) provides the system the
choice of using explicit beamforming (EB) or spatial multiplex-
ing, but not both. With only 2 dB of incremental gain to offer,
EB does not provide enough of a performance benefit to out-
weigh the data rate gains from SM in any case other than that
for which SM is not an option, the 1x1:1 system. As can be seen,
2 dB doesn’t buy much, in terms of either throughput increases
or range extension — roughly a 20% increase.
[3] 802.11n 2x2:2 omni. This reflects the two spatial streams
performance of an AP equipped with 2 omni antennas commu-
nicating with a 2 Rx equipped client such as a laptop. Note that
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
decode the two streams effectively. The performance of an AP
with implicit or explicit beamforming capabilities would look
exactly the same as this curve, because of the mutual exclusiv-
ity of spatial multiplexing and phase-based beamforming.
[4] 802.11n 3x3:3 omni. This adds a 3rd transmit antenna to
curve [3]. Note the rapid convergence of the 3x3 system’s
performance with that of the 2x2 system. The higher modula-
tion classes (based on 64 QAM, a very complex constellation)
in combination with 3 spatial streams are only effective, in
practice, at very short ranges.
[5] 802.11n 1x1:1 + BeamFlex. This adds the 6 dB BeamFlex
link budget gain to curve [1], showing a roughly 2x increase in
either throughput or range.
[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
increase.
To net this all out in terms of the impact of a vendor selection
made today over the life of an organization’s WLAN, we need
to look at one more piece beyond the AP — the evolution of
client capabilities over time.
7. Client capabilities over time matter, tooMore often than not, an AP’s throughput is limited by what
can be supported by its clients. This is especially true for the
explicit beamforming and three-stream SM modes, which have
yet to reach the client base. For the purposes of analyzing AP
productivity over time, we’ve assembled a rough forecast of
the market penetration of different client capabilities, based on
feedback from our chipset suppliers and various analyst views
(see Figure 15 on the next page).
One theme that runs across each of the three categories we
depict is the timing of introduction of EB-enabled devices.
While the Wi-Fi chipset supply community is now releasing
the capability in the latest generation of their products, it will
take some time for the handset, tablet, and laptop vendors to
incorporate them into their devices. The larger uncertainty is
actually the timing for formal interop and certification pro-
gram development at the Wi-Fi Alliance. Because of the very
marginal benefit of the technology, some chip suppliers have
expressed doubts that the WFA timeline will be short, since the
technical challenges of proving in multivendor interop for new
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-ence, near LoS link conditions, indoors.
FIGURE 14: Rate and range comparison of various multi- antenna technologies. See text for additional explanatory notes and sources.
0
20
40
60
80
100
120
140
160
180
0 5 10 15 20 25 30 35 40 45 50
[1] .11n 1x1:1 omni
[2] .11n + Explicit Beamforming (1 stream)
[3] .11n 2x2:2 omni
[4] .11n 3x3:3 omni
[5] .11n 1x1:1 + BeamFlex
[6] .11n 2x2:2 + BeamFlex
RF Technology Comparison
Peak Throughput, Mbps
Range, m
Page 13
Marketing v. Reality in Multi-Antenna Wi-Fi
behaviors introduced at very low levels in the protocol (and
the feedback required from clients in EB certainly qualifies as
such) are substantial. For purposes of illustration here, we have
assumed a scenario where these matters will be sorted out on
a fairly optimistic timeline — by the end of 2011 — and that
the progression to widespread adoption in mobile devices will
begin in 2012. These figures could easily be 6 if not 12 months
too optimistic.
Smartphones
The vast majority of smartphones today are single-stream
devices. A small number of legacy (801.11b/g) protocol devices
persist in the market, but those are expected to be aged out
of relevance quickly, given the typically short lifespan in this
category. Power and space constraints will prevent much head-
way for 2-stream models, but consumer pressure for higher
throughput over time will eventually lead to some adoption
of 2x2:2 architectures. The presence of EB in the chipsets will
cause this technology to diffuse quickly, mostly because of
short device lifespans rather than inherent merit.
Tablets
Like smartphones, the majority of tablet devices today are
1x1:1. Because of relatively lower pressure from battery and
form factor constraints, there will be a growing segment of
2x2:2 architectures in the tablet category, gaining EB capabili-
ties within the planning period as lifespans are short here, too.
Laptops
Since the laptop installed base is retired more slowly, the
diffusion of EB capabilities into this category will take a little
longer. Relatively less severe constraints on space and power
have already given two-stream architectures a head start here
— the introduction of three-stream APs will be accompanied
by three-stream support on laptops, too, but on the relatively
longer timeline for laptop replacement.
So what’s our point?The last step in the analysis here is to assess the relative merits
of a couple of relevant purchase decisions made today, given
both the device futures we’ve depicted as well as the perfor-
mance characterization of APs working with these different de-
vice types. To illustrate the trade-off, we consider the options
of building a WLAN network based on either (a) Ruckus 3x3:2
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
point highlighted in the figure) with the client capabilities
roadmap in Figure 15 yields the total aggregate traffic sup-
ported comparison in Figure 16. Given the qualitative narrative
we’ve presented here, it should come as no surprise that the
quantitative reality of the 3x3:3 + EB story trails far behind its
marketing. EB’s performance gain is minimal, and limited to
single-stream devices that support the feedback protocol, of
Sources: Wi-Fi chipset vendor expectations, various analysts.
FIGURE 15: Scenario-based forecast of client capabilities over time.
0%
100%
2011 2012 2013 2014 2015
legacy
:1n
:2n
:2n EB
:3n EB
Laptops
0%
100%
2011 2012 2013 2014 2015
:1n
:2n
:1n EB
:2n EB
Tablets
0%
100%
2011 2012 2013 2014 2015
legacy
:1n
:1n EB
:2n EB
Smartphones
FIGURE 16: WLAN lifetime impact of BeamFlex v. 3x3:3+EB purchase decision.
Ruckus 3x3:2 APwith BeamFlex
4-Year Aggregate Client Throughput per AP, Terabytes
Generic 3x3:3 APwith ExplicitBeamforming
440
more capacity for client servicewith Ruckus
260
70%
Page 14
Marketing v. Reality in Multi-Antenna Wi-Fi
Ruckus Wireless, Inc.
880 West Maude Avenue, Suite 101, Sunnyvale, CA 94085 USA (650) 265-4200 Ph \ (408) 738-2065 Fx
w w w. r u c k u s w i r e l e s s . c o mCopyright © 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 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
which there are unlikely to be very many until late in the analysis
period. Likewise for 3-stream spatial multiplexing. The net re-
sult is a Ruckus 2x3:2 AP with BeamFlex will provide 70% more
capacity for client service over the life of a WLAN.
8. In conclusionWe’ve been through lots of details here. Let’s recap the story,
to make sure things are perfectly clear.
Other AP vendors are piling on the “beamforming” band-
wagon, since we’ve demonstrated how valuable multi-antenna
approaches to Wi-Fi can be.
There are issues with their approaches, though:
1. Implicit beamforming provides effectively zero performance
benefit in practice.
2. Explicit beamforming (new to the scene) does a little better,
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
3. There won’t be very many clients around to take advantage
of explicit beamforming or 3 spatial streams for a long time
Meanwhile, Ruckus BeamFlex continues to offer a 2x perfor-
mance advantage over other APs.
Despite competitor marketing to the contrary, the Ruckus solu-
tion is 100% standards compliant, works with all Wi-Fi clients
today, and can be used directly with spatial multiplexing to
give the benefits of both SM and beamforming, but without
anywhere near as much concentrated co-channel interference
introduced into the network.
Summed over the life of a WLAN, we can provide 70% more
aggregate throughput per AP than the best of the rest.
We think the right choice for your WLAN is obvious. Please do
not hesitate to contact us for more details.
Go to www.ruckuswireless.com for more information.