802.11ac (Wave1): MORE Network Engineering Insights
In my previous blog on the 11ac series, I explored 80 MHz channel operation in 802.11ac in the context of data rate, OBSS (Overlapping BSS), network throughput, and autochannel assignment.
802.11ac (Wave1): Network Engineering Insights
In the present post, I explore the other speed factor of 1.33X that shows up in the Wave1 data rate equation: (2.16 x 1 x 1.33) x 450 Mbps of 802.11n rate = 1.3 Gbps. This 1.33X factor is attributed to the new modulation technique called 256QAM introduced in 802.11ac (802.11n had only upto 64QAM). Consistent with the theme of this blog series that the data rate equation does not bring out critical network engineering aspects, this post explores 256QAM from the enterprise network design perspective.
256QAM causes step function change in data rate near the AP
There are two newly added MCS’s (Modulation & Coding Scheme) in 802.11ac. They result in respective data rate increase factors of 1.21 and 1.33, over the highest possible data rate in 802.11n for a given channel bandwidth and number of spatial streams.
These two newly added MCS’s use the 256QAM scheme, which requires about 5 to 7 dB higher SNR (which is a lot given that dB is logarithmic scale) compared to the least SNR at which the best MCS in 802.11n (64QAM, R 5/6) can work with.
As a result, the 256QAM can only be used close to the AP. From the network engineering standpoint, the key point to note is that 256QAM to 64QAM is step function change, that is, as you move away from the AP, the data rate drops in step function from 256QAM rate to legacy 64QAM rate.
This observation is important to quantify cellwide benefit of 256QAM.


What is the cellwide impact of 256QAM?

In enterprise deployments, clients are distributed throughout the cell. In a sense, this is different from the home networking environment where many clients can be close to the AP. A wellknown principle in 802.11 is airtime unfairness, which means clients away from the AP consume more airtime due to their lower speed compared to those closer to it. By now, you probably can guess what I am getting at.
For illustrative purposes, consider four clients (let us call them C1, C2, C3, C4) at four distances from the AP, respectively, and having data rates (assuming 40 MHz channels and 2 antennas on clients) as follows:
 C1 @ 360 Mbps (256QAM rate with 1.33X data rate increase),
 C2 @ 270 Mbps (maximum 64QAM rate),
 C3 @ 216 Mbps (another 64QAM rate), and
 C4 @ 108 Mbps (16QAM rate).
I will compare this situation with the corresponding 802.11n data rates (no 256QAM) at the same distances for the same clients:
 C1 @ 270 Mbps (maximum 64QAM rate),
 C2 @ 270 Mbps (maximum 64QAM rate),
 C3@ 216 Mbps (another 64QAM rate), and
 C4 @ 108 Mbps (16QAM rate).
Below is the diagram depicting total airtime saved due to the use of 256QAM for clients close to the AP in the above example. Here, I have avoided using lower rates like 54 Mbps and 27 Mbps (which are for the QPSK and BPSK modulation schemes) for clients further away from the AP to favor 256QAM. The saving in airtime will be distributed to the clients in proportions of their data rates.

The above example shows about 4% saving in total airtime for the cell when the client close to the AP can use 256QAM.. Also a point to note here is that actual numbers of data rates and clients are not important and that relative proportions are important. You get the same saving number for the same relative proportions of the data rates.

More clients away than close (Area = Pi * Square of radius effect)

The area of coverage of the cell is proportional to square of distance from the AP (middleschool formula for the area of the circle).
So in reality, there are usually more clients away from the AP than as many close to the AP. This type of client distribution requires computation of weighted proportions of airtime consumption rather than simple proportions as I did above. With weighted proportions, the savings in total airtime due to the use of 256QAM close to the AP are below 5%.
For example, with one C1type client, two C2type clients, three C3type clients and four C4type clients, the total airtime saving because of C1 being able to use 256QAM comes out to be 1.5%.

Airtime fairness feature on AP

APs support airtime fairness feature which tries to prevent higher airtime usage by clients operating at lower data rates. Suppose the fairness feature is configured to equalize the airtime consumption across clients. Then, in the computation above (with simple proportions), without 256QAM, airtime would have been equalized as 25% each for each of the four clients. When 256QAM is used, only one of the 25% slices (representing client closest to the AP) see airtime reduction of about 25% (due to 1.33X data rate).
So when normalized over the entire cell, with equal airtime fairness implemented on the AP, the total airtime saving due to the use of 256QAM near the AP, comes to about 6.25%. As discussed earlier, in general there will be more clients away from the AP than those close to the AP. With weighted proportions computation as above, the total airtime savings is about 2.5%.

New radio implementations

As we can see from the previous examples, raising data rates of only those client that are close to the AP (like what 256QAM does), results in relatively small total airtime savings (this reminds me of an analogy from popular rhetoric: “what does it mean to the society if the rich become richer”).From the network engineering perspective, the clients that are away from the AP need more help. One hope is that 802.11ac clients may have better radio implementation than the 802.11n clients. This may enable the 802.11ac client at a given distance to achieve better SNR than the 802.11n client at the same distance. Introduction of low density parity check (LDPC) codes introduced in 802.11ac could also help a bit there, but that alone does not seem to be adequate. However, whether the net SNR boost will be adequate enough to raise the client at least one level up in the data rate (i.e., one layer up in MCS), remains to be seen until real life test results are out.
Overall, we see that 256QAM shows juicy 1.33X gain factor in the Wave1 data rate equation. However, from the perspective of cellwide impact, the airtime savings can be much lower. There needs to be a way to raise data rates of all the clients, particularly of those away from the AP, in order to achieve attractive airtime saving (and hence capacity and throughput gain for the cell). In that regards, 256QAM seems to be better geared towards home networking than enterprise networking.
For enterprise networking, we may have to rely on radio implementation improvements due to hardware and processing techniques enhancements over time, to be able to obtain blanket data rate increase over the cell. Alternatively, one can plan coverage of .11ac cells to raise the minimum data rate at the edge of the cell, but it has cost and cochannel interference considerations.
These network engineering insights are appreciated only if you think outside of the isolated data rate equation!

Addition Information:
 802.11ac (Wave1): Network Engineering Insights
 Don’t deploy 802.11ac without thorough RF planning
 Abandoning the 2.4 GHz junk band – Moving WiFi to 5 GHz by Lisa Phifer in Webtorial
 The 802.11ac Paradox by Lee H. Badman
 BOM Math for Secure WiFi Deployments
 WiFi networks in 5 GHz: a few observations

Devin
Thanks for your feedback on the blog and additional insights into 11ac network engineering. This is an area where lot more needs to be analyzed before we can declare victory on 11ac. I am confident victory will happen once relevant issues are brought into focus. Also good catch on long/short GI rates.
Hemant
Hi Hemant,
This is an excellent blog. Thanks for writing it. I would like to make some specific comments.
1) The data rate numbers you use for C1C4 are with long guard interval (nonSGI). The reason I mention it is that most folks are used to seeing 300Mbps as the maximum 64QAM data rate on a 2SS device. The 270Mbps might be a little confusing to those who are less familiar. Nevertheless, your numbers are accurate.
2) Your notes about computational weighting and the number of clients further from the AP often being greater than those close to it is something that most folks wouldn’t consider. We hear all the time, in the marketing propaganda, about how 11ac is airtime efficient due to 256QAM. While it would be if all clients were within range to use 256QAM, the reality of its overall effect on cell efficiency might be minimal due to client range — EVEN with airtime fairness enabled. I like how you pointed that out so clearly. Kudos.
3) Your point about 11ac clients is an important one.
4) Your point about the ACI and CCI ramifications due to cell planning around 256QAM is a VERY important one and one that I’ve been harping on for a while now. I don’t believe that cell planning around 256QAM is a good idea at this point given how much power is required to maintain those MCS rates. This seems to call for one or more mechanisms whereby clients can be urged to gather more closely around each AP (rather than being in the netherregions between APs).
Thanks again for the relevant blog. It’s an excellent reality check for implementers of 11ac, regardless of vendor.
Devin