MU-MIMO only works on the downlink, and kills beamforming


Last week we ran an article giving Qualcomm’s claims for MU-MIMO and while it is proving tough to find experts prepared to commit themselves on the subject of MU-MIMO at short notice, there are certainly more questions for Qualcomm to answer long term on this subject – you may remember last week it was unrepentant with its claim that MU-MIMO is 2 to 2.5 times faster in most situations than existing 802.11ac with SU-MIMO.

It is true however that over time much of the industry accepts that MU-MIMO is achievable and will deliver these kinds of benefits. What is less clear is whether or not Qualcomm has already achieved these benefits or whether they are still off into the future.

One thing to understand is that the gains which are currently obtained are highly contextual, meaning that under some conditions they may appear, and under others they may disappear.

At a high level, MU-MIMO multiplexes airtime by transmitting it in such a way that individual clients hear only their part of the message at any given time, and the rest is nulled to them. We think of this like the inverse problem of beamforming, where the device is trying to maximize the signal for a given client. In MU-MIMO the device is trying to minimize or null parts of the signal for all but one client, and it does this to two or three devices at a time. To do this it uses the same techniques as beamforming, so the two cannot be used concurrently.

Many providers are unwilling to accept that this can happen without the Access Point and the client both supporting it, which is why Qualcomm feels it can steal the entire market by saying that all the chips that it has been shipping for the past year can retrofit this capability in the client end.

The idea is that if only Qualcomm receiver chips can work with MU-MIMO, then you had better select these chips exclusively and drop Broadcom, Ralink, Marvell and Lantiq chips in phones. Of course that’s not going to happen given the current market distribution.

While this will be the marketing message of Qualcomm, there is some disagreement as to how efficiently MU-MIMO can be done at all given today’s state of the art. If the Qualcomm chips can demonstrate this in the field, then there will be a mad rush to close the gap on it, but that would provide Qualcomm with a 9 month or so lead, and it would dominate handset WiFi during that time and perhaps beyond.

One thing that will confuse the issue is the fact that the clients cannot reply to the Access Point in MU-MIMO – this is a downlink only technology that requires a lot of CPU power to resource and client STAs do not have enough. Since the client has to take turns with the downlink this reduces the gains by almost half.

Qualcomm’s initial explanation suggested to us that three, two antenna devices could each get 200 Mbps from a MU-MIMO Access Point, rather than a single device getting 200 Mbps in this way and logically this tripled performance. But if you say that this can happen half of the time, then the performance is half that amount so 1.5 times the performance improvement, and that’s achieved in a technology where it is always very difficult to see exactly what is going on – WiFi is notoriously chaotic to predict.

Also MU-MIMO has a hard cap on how many users you can simultaneously send to, based on the number of transmit chains on the AP — a 4×4 AP could theoretically only send to 4 stations simultaneously. We hear that realistically it’ll be at most one fewer than the number of chains, so 3 for a 4×4, which is more or less what Qualcomm admitted to .

In conclusion, given that MU-MIMO is mutually exclusive with beamforming, there is a tradeoff in total performance and if a company puts a huge amount of R&D resource only to find that its chips perform more or less at the same level, it would be a spending disaster, which accounts for the reluctance for others to head down this track as swiftly as Qualcomm.

Another issue is that the modulation scheme may well drop a notch when using MU-MIMO, so that the maximum may drop from an 867 Mbps PHY for two AC streams. If it has to drop to 64 QAM, the modulation scheme may only achieve 565 Mbps or 520 Mbps as its core PHY rate, further eating into the performance gains and making them illusory. It might be better off having one signal at full speed than two at a reduced speed.

What most people we spoke to said was that the real contribution of MU-MIMO might turn out to be at the highest modulation speeds and that means in scenarios where the mobile client is in the same room – so where a 4×4 MIMO AP is talking to multiple 2×2’s in the same room, where the signal is excellent.

In effect they think that cash rich Qualcomm has made a gamble in reaching for MU-MIMO first, and that it is unlikely that this will turn the market from a position where it is strong, but not dominant, in handset WiFi.

We may return to this subject once again as there are still some experts out there we have not yet been able to schedule a call with.