802.11ac (Wave-1): Network Engineering Insights
Have you noticed all the buzz on 802.11ac especially now that Wave-1 has arrived?
How could you not! 802.11ac is the new Wi-Fi standard and it has new techniques to increase the wireless data rate above the existing 802.11n standard.
11ac is slated to arrive in two Waves – Wave-1 this year and Wave-2 next year.
At its full potential (Wave-2), the standard is characterized as: 802.11ac max data rate of (4.33 x 2 x 1.33) times the 600 Mbps max data rate of 802.11n, which comes out to be about 6.9 Gbps.
The current version (Wave-1), which is commercially limited to 80 MHz channels and 3 spatial streams per AP, the standard is characterized as: 802.11ac max data rate of (2.16 x 1 x 1.33) times the 450 Mbps current data rate of 802.11n, which comes out to be about 1.3 Gbps.
It is also important to note that Wave-1 does not have MU-MIMO. The MU-MIMO effect does not explicitly show up in the data rate equations, but it can have profound benefits in practice.
So what is missing in the above characterization?
Network engineering insights – that’s what is missing! There are several network engineering nuances which do not show up in the above equations. I will discuss them in a series of blog posts starting with this one. In this initial post, the focus is on the nuances of operating 80 MHz channels in the practical network deployments. These 80 MHz wide channels are responsible for the first multiplicative factor of about 2X in the Wave-1 data rate equation.
However, when considering the 80 MHz channels, special attention needs to be paid to OBSS. OBSS (Overlapping BSS) is the term that the standards community uses to indicate that other APs overlap with the BSS (AP) using the 80 MHz channel.
The OBSS can be the APs in your neighborhood. Or, your AP can be OBSS to neighboring 80 MHz APs. OBSS can also be in your own network, particularly in dense high capacity enterprise deployments. All that to say that OBSS has profound impact on how and if the 2X factor propagates to throughput gains.
When no transmission from the OBSS is detected anywhere in the 80 MHz span, the BSS transmission happens with full 80 MHz bandwidth at about twice the data rate of the 40 MHz channel.
However, when the transmission from the OBSS is detected anywhere in the 80 MHz span, either
a) the entire 80 MHz remains idle if the interference overlaps the primary 20 MHz, or
b) the non-interfered primary 20 MHz or 40 MHz within the 80 MHz
is used for transmission. Whether it is a) or b) depends on the exact nature of channel overlap.
This bandwidth adaptation is a feature in 802.11ac that allows the transmitter to utilize primary 20 MHz or primary 40 MHz channel for transmission when interference from the OBSS is detected only in the secondary portion of the 80 MHz channel. There is an additional RTS/CTS based bandwidth adaptation available in 802.11ac to address interference at the receiver.
In RTS/CTS based adaptation, after the 80 MHz channel is sensed idle, the RTS/CTS handshake is performed in each 20 MHz component of the 80 MHz channel to sense interference situation at the receiver and the transmission bandwidth is adjusted as discussed above based on the outcome of the handshakes.
The OBSS Phenomenon
Below is a very simple illustration of the OBSS phenomenon. This illustration is based on the premise that both the 80 MHz BSS and the (collective) 40 MHz OBSS have enough traffic to transmit all the time, so they effectively take turns in winning the channel contention. In this representation, the benchmark chosen is that of two separate non-overlapping 40 MHz channels like in 802.11n, with “T” being the nominal throughput of each.
As can be seen, without the bandwidth adaptation, total throughput (4/3 of T compared to 2T) and individual throughputs (2/3 of T compared to T) of both BSS and OBSS are reduced compared to the two separate 40 MHz channels benchmark. With bandwidth adaptation, total throughput (2T) is only even with two separate 40 MHz channels benchmark. Here, 80 MHz throughput is increased (4/3 of T compared to T), but at the cost of OBSS throughput (2/3 of T compared to T).
Additional channel alignment issues also impact overall throughput. For example, from the previous example, if the 40 MHz OBSS overlapped with the primary 40 MHz of the 80 MHz channel, then the 80 MHz transmission is completely blocked, with or without bandwidth adaptation. This is because, primary channels always need to be included for transmission, and hence, the 80 MHz channel does not have option to ever transmit when the 40 MHz OBSS is transmitting.
More sophisticated traffic models considering multiple OBSS configurations, varied traffic patterns, and addressing additional intricacies such as backoff windows are possible. One example is this study which simulates the probability that OBSS has interfering transmission.
Another consideration is that in practice transmission patterns will be more disorderly than what is shown in our illustration. However, the main conclusion is: Wider 80 MHz channels introduced in 802.11ac (Wave-1) may result in the data rate increase of about 2X over the 40 MHz channels, but in the presence of OBSS, throughput results can vary widely based on the amount of channel blocking and bandwidth adaptation. This matches up with the actual testing we conducted on early 11ac APs.
Observations on bandwidth adaptation
As we see in the above illustration, with bandwidth adaptation, the overall network throughput performance is better than the one without. An interesting point however is that the bandwidth adaptation may help the 80 MHz channel, but it could be at the expense of the 40 MHz OBSS. So, if you are an OBSS and your neighbor is 80 MHz, then you are out of luck.
Another related point that comes out is that the total throughput of the 80 MHz BSS and the 40 MHz OBSS may not be better than the total throughput with the two non-overlapping 40 MHz channels, even when the 80 MHz channel performs bandwidth adaptation! Accordingly, if OBSS is in your own network, it can create throughput imbalanced cells, rather than increasing overall throughput.
Auto-channel techniques for 80 MHz channels
Many deployments use auto-channel assignment in their networks. These auto-channel algorithms work with the heuristic to keep your BSS as much away as possible from the OBSS. This is most effective scenario is when there are many non-overlapping channels available to select from. However, when working with 80 MHz channels, the auto-channel techniques may find it difficult to find free spectrum where they can steer the AP’s channel because of the OBSS in your own network or the OBSS from the neighbor’s network.
Any sub-optimal positioning of 80 MHz channel by the auto-channel techniques can have unpredictable effects on the various throughputs due to the different factors discussed above. Accordingly, the ability of the auto-channel techniques to effectively maneuver or maneuver around the wide 80 MHz channels with the cognizance of associated throughput impact, becomes important practical consideration in real life deployments.
Network engineering insights are important to estimate practical gains from deploying new technology. In Wave-1 of 802.11ac, the main technical advances resulting into data rate increase come from 80 MHz channels and 256-QAM.
However, operating 80 MHz channels is akin to driving a wide vehicle on a busy highway –> this requires greater care, better maneuverability, but also creates higher potential for lane blocking.
In particular, in the presence of OBSS which can be in your own network or in the neighbor’s network, the doubling of data rate with 80 MHz channel may not predictably translate to throughput gains due to various overlap aspects and the behavior of auto-channel techniques. Once experiences from real early deployments start coming in, we will truly know whether 80 MHz channels can indeed be effectively operated in the dense enterprise environment or whether 40 MHz channels is the better choice.
IEEE 802.11ac: Dynamic Bandwidth Channel by Minyoung Park