1. Rate My Wi-Fi
Ruckus Wireless | White Paper Finding the right balance between optimum
performance and reliability with adaptive data
rate algorithms
When buying a sports car, we often focus on engine size, top Instead of selecting the fastest mutual speed, 802.11 stations
speed, horsepower, and 0-60 time. But utilizing those capabili- attempt to find the best speed, based on a tradeoff of reliabil-
ties requires a well-designed transmission. ity and performance—note that uplink and downlink conditions
are different, so the AP and client have data rate autonomy.
In Wi-Fi, we often focus on maximum data rate, MIMO configu-
ration, channel size, and fancy antennas (guilty as charged). We
rarely talk about the mechanism that switches Wi-Fi gears. What’s a data rate?
Thanks to marketing departments, Wi-Fi speeds and feeds are
Like a car, Wi-Fi devices have a transmission too. In Wi-Fi, it’s fairly well known by Wi-Fi people. However, fewer people inti-
called dynamic rate adaptation (aka. rate control, rate switch- mately understand why there are different data rates and why
ing, or rate selection). dynamically changing data rates can improve communications.
Rate adaptation is the function that determines how and when Fundamentally, it’s critical to understand that higher data rates
to dynamically change to a new data rate. When it’s tuned are more “complex” than lower data rates. With lower data
properly, a good adaptation algorithm finds the right data rates, the modulation and coding mechanisms are simplified,
rate that delivers peak AP output in current RF conditions – which makes them less efficient, but more reliable.
unstable as they are. Though often ignored, rate adaptation is
a critical component to any high performance system. Each data rate is the product of some specific combination of
modulation and coding—as well as other factors like channel
bandwidth and spatial streams.
Choosing a rate: Ethernet vs. Wi-Fi
On a wired Ethernet link, endpoints connect and auto-negotiate
an interface speed at the fastest mutually supported signaling Modulation
rate. Because Ethernet link conditions are static, the rate remains Modulation is the process of changing the properties of a car-
the same. Simple. rier wave to represent information bits. There are three basic
types of modulation: amplitude, frequency, and phase.
In Wi-Fi, link conditions change more often than HP changes
CEOs. To find the best data rate in an undulating sea of Figure 1 shows a simplified concept in which each modulation
unlicensed spectrum (mobile clients, transient devices, RF change represents a single bit of data (the baseband signal).
interference, temporary networks, bursty traffic, etc.), smart We could also visualize modulation on a constellation map, as
rate adaptation is essential. Wi-Fi systems must handle chang-
ing conditions in stride, adapting communication rates based
on a complex set of variables.
2. Page 2
Rate My Wi-Fi
FIGURE 1: Basic Types of Modulation FIGURE 3: Phase and Amplitude Modulation with 16-QAM
1 0 1 1 0 1 0 0 16-QAM Q b0b1b2b3
00 10 01 10 11 10 10 10
Baseband Signal TIME
Q +3
de
itu
00 11 01 11 11 11 10 11
pl
Phase°
Am
+1
I
Amplitude Shift Keying (ASK) -3
00 01
-1
01 01
+1
11 01
+3
10 01
I
-1
00 00 01 00 11 00 10 00
-3
Frequency Shift Keying (FSK)
16-QAM (quadrature amplitude modulation)—4 bits per
symbol—is shown in Figure 3. By now, I’m sure you can see the
Phase Shift Keying (PSK) complexity and efficiency pattern emerging. Higher-order modu-
lation is more efficient—more bits of data per sample. But, better
FIGURE 2: Phase Modulation with BPSK and QPSK signal quality is required for reliable mapping by the receiver.
QPSK 64-QAM—6 bits per symbol—is also used in 802.11a/g/n, and
Q BPSK Q Q b0b1
b0 802.11ac will introduce 256-QAM—8 bits per symbol.
01 11
+1
Phase° +1
0 1
I -1 +1 I -1 +1
I Coding
00 10
-1 -1
Another important mechanism that controls efficiency and
1 Bit — 180° Phase Shifts 2 Bit — 90° Phase Shifts
reliability is the coding rate. Also known as forward error cor-
rection, coding is the process of adding redundant information
bits to a data stream to improve reliability over unreliable
in Figure 2. In Figure 2, we’re focused on phase shift modula- mediums. In other words, x number of data bits are converted
tion, which is common in Wi-Fi. When the receiver receives into y number of coded bits to improve error recovery. We
a modulated signal, the phase of the signal is aligned with a express this as a ratio of data bits to coded bits.
constellation (i.e. the red dots) on the map and represents a
specific bit pattern (e.g. 00, 01, 10, 11). More complex modula- Data Bits Coded Bits Coding Rate Efficiency Reliability
tion types have more bits on the map. 1 2 1/2 Less More
2 3 2/3
With BPSK (binary phase shift keying), there is only one bit of
3 4 3/4
information. The receiver detects the phase of the signal and
5 6 5/6 More Less
matches that phase with a bit pattern (either the right or left of
the constellation map): either a 0 or a 1. So the margin for error As with our modulation methods, the tradeoff for coding is
is quite large, making this a highly reliable modulation method. efficiency versus reliability. The benefit of data redundancy
However, one bit per symbol is inefficient. over noisy RF channels is a priority, but the goal is always to
find the right balance.
With QPSK (quadrature phase shift keying), there are four
possible constellation points—2 bits of data. Compared with
BPSK, the receiver must detect the signal’s phase with more Modulation and coding schemes
precision. Thus, the signal quality must be higher, but the tech- We use the term data rate to indicate the speed of a wireless
nique is more efficient. connection. Data rates are determined by a number of vari-
ables, but the primary elements that we can dynamically control
To add efficiency, the next higher order of Wi-Fi modulation
are modulation and coding schemes. The table on the next
uses both phase and amplitude shifts, as in Figure 3.
page shows how the data rate increases or decreases based on
the efficiency of the modulation and coding methods.
3. Page 3
Rate My Wi-Fi
Data Rate (Mbps) Modulation Coding Rate OFDM Subcarriers Coded Bits per Coded bits per Data bits per
subcarrier OFDM symbol OFDM symbol
6 BPSK 1/2 48 1 48 24
9 BPSK 3/4 48 1 48 36
12 QPSK 1/2 48 2 96 48
18 QPSK 3/4 48 2 96 72
24 16-QAM 1/2 48 4 192 96
36 16-QAM 3/4 48 4 192 144
48 64-QAM 2/3 48 6 288 192
54 64-QAM 5/6 48 6 288 216
When we look at the guts of modulation and coding, it How do we know whether a few errors are an anomaly or a
becomes clearer why rate adaptation is necessary. Access predictor of normal conditions for this network? The reality
points are responsible for choosing the best combination of is that we don’t. Every RF environment is different, so simple
modulation and coding at any point in time for each connected thresholds are a poor answer to rate shifts. Based purely on
device. Again, it’s always a tradeoff of efficiency (higher data assumptions, the best reaction to errors and retries is unknown.
rates) and reliability (lower data rates).
In fact, the best response might be to use higher data rates
The 802.11 specification introduces the term dynamic rate because they occupy the wireless channel for a shorter period
switching and acknowledges the fundamental issue: with mul- of time and are less likely to be corrupted by momentary inter-
tiple data rates, there is a need to dynamically adjust based on ference. Commonly enough, a data rate downshift causes more
RF conditions. But, they don’t lend any help. So if you look at errors, which causes another downshift. And suddenly, the safe
ten different Wi-Fi companies, you’ll see ten different rate con- and reactionary rate switch leads to a rate shift sinkhole–gob-
trol algorithms. bling capacity as it tanks to the bottom.
In other words, purely reactive algorithms are sub-optimal in
What’s so smart about Ruckus? their myopia. Math is the better way. Statistics tell us more
Wi-Fi engineers have been led to believe, and—for better or about the implications of transient interference, short-term hic-
worse—site survey software validates the belief, that data rates cups, and longer-term trends. Accordingly, we can adjust—or
can be reliably predicted based on a metric like RSSI or SNR. perhaps more importantly, not adjust—the data rate to opti-
And some product manufacturers use simple metrics like these mize both short-term and long-term performance and capacity.
to determine the right rate. This is also why spectrum analysis doesn’t help much. Identify-
ing an interference source does not readily tell us how exactly
Ruckus approaches rate selection with a unique focus. Instead
that source will impact our network. Even the best heuristics
of using unreliable signal measurements to hope for the best
aren’t as accurate as statistics.
data rate, we focus on the math. Our rate selection algorithms
are statistically optimized, which is our engineer-chic way of For delay- and jitter-sensitive applications, the best data rate
saying that we pick the best data rate based on historical, sta- is also the one that consistently delivers the frame to its des-
tistical models of performance for each client. tination in the shortest amount of time. Our statistical rate
selection model ensures that too.
Without the right algorithm, the optimal rate for any client at
any given moment in time is a crapshoot. And when you’re Another unique advantage with Ruckus is our test and valida-
guessing, the safest guess is to err on the side of reliability, tion rigor. Because of our custom AP hardware and software,
which sacrifices throughput and capacity and causes other we test and test and test everything some more. One such
unwanted problems. monotonous test is for data rate performance. Believe it or
not, we test every individual MCS rate at different ranges and
Let’s look at an example. The normal thought process in
conditions to ensure that our performance is a bulwark of reli-
Wi-Fi is that frame corruption and Layer-2 errors should lead
ability. And this is no small feat. 802.11n MCS options are far
us to downshift to a more reliable data rate. It’s a reasonable
more complex than 802.11a/g.
assumption. However, interference is bursty and transient by
nature, so the best response is not necessarily to downshift.