Your browser is not supported. Please update it.

23 October 2014

Qualcomm-Atheros pounds the MU-MIMO drum one year in

Qualcomm once again this week waved their MU-MIMO (multi-user Mimo) WiFi implementation at us and repeated its claim that it can offer WiFi which is three times faster than “traditional” WiFi. It bothered us because usually WiFi chips work well in some conditions and crash and burn in other conditions, and it’s very tricky to back up hard and fast claims like that one.

Other WiFi players have urged us to ignore MU-MIMO saying that it is only of any use in high density environments like stadiums, and suggesting that even Qualcomm would admit to this. It didn’t. Qualcomm is describing the move to MU-MIMO as version 2.0 of 802.11ac almost as if this will be the “de facto” choice of operators and handset vendors going forwards, its spokesman Todd Antes, VP of products at Qualcomm Atheros, perhaps demurred at declaring its 3 times as effective as SU-MIMO, but was certainly happy that it was at least twice as fast most of the time, and 3 times as fast some of the time. He suggested we settle on 2.5 times.

Rivals often play down the achievements of other companies so it was worth reaching out to Qualcomm and get its side of things, and we have also asked rivals to give us their take, but no-one was able to get back to us prior to writing this article this week, but they have promised to pursue it. We also needed to have a better understanding of how MU-MIMO works in order to make a call one way or another.

It must be noted that the recent announcements of Qualcomm Atheros really were two fold and relatively minor. It was last year that it came out with its first MU-MIMO chip and this is it is quietly filling out its product line. This week it has new MU-MIMO capable chips both for Access Points and others for receiving devices, one of these being new, the QCA9377, but this is a bottom end addition to the range, which can receive two or more streams at once from a multi-user Access Point. The other chips were an MU 4 x 4 chip to build Access Points with and the announcement that it has its first design win at AVM. It also told us later about another supplier Devolo that is also taking the Multi User chips.

Neither or these customers are specifying devices for massive operator networks – if these had deals with Deutsche Telekom or some such, we’d be more impressed, but they are both players in the German retail device market. So MU-MIMO today cannot claim any operator design wins that will lead to millions of chip shipments.

Antes warmed to his explanation, “In Single User mode, each client can only receive one stream at a time. In any MIMO format whether it is 2 x 2 or 3 x 3 or more, one stream is a maximum of 433 Mbps in 802.11ac and 150 Mbps in 802.11n.

“What happens if you have a device which can send 4 streams of 802.11ac is you have a maximum theoretical throughput of 1.7 Gbps. Prior to MU-MIMO if there is just one receiving device, and it had 2 antennas, it could only receive 2 streams – so its maximum receive rate is 867 Mbps, and since the system works sequentially, the other streams simply cannot be used when speaking to that device. So in that situation, half of the capacity is going unused.” Antes makes the point that the great many devices out there (perhaps 40%) are single antenna and most of the others have just two antennas (maybe 50%).

“If you have three devices and they are dealt with sequentially by single user MIMO, then PHY throughput would be just one third of that 867 Mbps. By the time you take overhead out of the PHY rate you may get a usable 600 Mbps, so that’s 200 Mbps each.

“If you have an MU capable Access Point then all 4 streams can be sent at once, and they all get that speed,” so we assume they all get 600 Mbps all things being equal.” Of course things rarely are equal in WiFi as the condition of the airwaves can change dramatically every few milliseconds, but that’s Antes logic and it seems sound for a 3x ratio.

Our concern is that what your AP is doing, depends on the condition of those airwaves. If you are trying to send to what we call a bad apple, a client device which is experiencing acute interference, and which lies at the edge of the WiFi reach, then there are variety of tricks that an Access Point can try.

If the device does have two antennas, then the transmitter can use two spatial streams to send the same information, expecting diversity benefits to either get the message through intact or get it through faster. But if your device is able to send multi user, then your other antennas can be sending simultaneously to different devices, not waiting for the prolonged period that such a bad apple can create.

But that’s not the only variation when an AP is trying to deal with a bad apple. It can also to drop to a different modulation scheme, one which is  more robust, but a lot slower. The shorter your bursts to keep up with a single user using SU-MIMO, the more time you have available to deal with your remaining clients. But in every single user configuration, there comes a time when a bad apple is right on the edge of its function and it makes the radio sending to it slow right down, and in a single user MIMO that can kill the network.

It must be remembered that all of these parallels spatial streams use the same spectrum. They are distinguishable from one another because each signal is coming from a different direction, often using different Modulation Coding Schemes. Each antenna must be sufficiently isolated from each other and the industry rule of thumb is that they need to be in the transmit device about half a wavelength physically apart.

Antes admits there is a cost to offering Multi-User MIMO in that there are more calculations and they are more complex, and these have to be carried out on the chip. But he insists, “the costs implications are not substantial,”  they are not as big as say the upgrade from 802.11n to 802.11ac and mostly you can use resources that are already there in version 1.0 chips.”

“In N,” he says, “the implementation of beamforming was  vendor specific, and they have fixed than in AC.  Everyone now does beamforming the same way and all the chips should work with all other clients in the same way.”

MU-MIMO sends using beamforming, There are two major different forms of beamforming – implicit which is the creation of a beam with what little the AP knows about the channel states, and explicit, which is based on sending something to the client and then getting it back so the system can measuring the distortion created to what are known values.

“In explicit beamforming you are making intelligent decisions about how to shape the energy for a specific client. It’s not always the best solution to use explicit beamforming, but typically it is useful in the mid-range of conditions. There is an overhead for instance in channel sounding, which has to be conducted regularly and prior to sending any information.”

So this means you are sending a channel sounding to every device, and this has to be done in hardware, rather than in software, otherwise it takes too long, as some designs did in the 802.11n generation. MU-MIMO can get a response simultaneously from 3 or 4 clients, and it makes sense to send this out during the Short Interframe Space, the gap between a data frames and their acknowledgement.

Antes insists that the MU know-how all comes out of Qualcomm’s understanding of cellular base stations sending in multi-user mode to individual handsets. Antes also said that handsets that had been shipped at any time over the past year which have Qualcomm chips in them, are already capable of MU-MIMO, so that a software update could allow these devices to take advantage of it.

“MU-MIMO will be better where there are high numbers of clients, especially 50 to 100 simultaneous clients, but this isn’t just arenas, it is convention centers, hotels, airports, and throughout enterprises, at venue WiFi (retail, café bars and restaurants) and in classrooms.”

Our biggest problem at Faultline with this entire discussion is that WiFi has to operate under a maximum EIRP (effective or equivalent isotropically radiated power) basically a power limit to prevent it interfering with other users of WiFi spectrum. This varies between 50 milliwatts at the lowest end, up to 1 watt, depending on which particular part of the spectrum it is in and which country you are in. We put this to Antes.

“I can’t tell you all our secrets, but I will tell you one, which is that MU beamforming is unlike single user beamforming in that it has to create 2 or 3 beams at a time, and this does use up a lot of power. But in order to allow as much energy as possible to be used on those beams, MU beamforming tries not to send any of the transmission out in every direction. It doesn’t only deal with active beams, but also active nulls. That means we specifically cancel the beam in certain directions, so that other clients don’t hear it and think it’s for them, but also so that the overall power usage is kept lower.”

“If you don’t believe me, ask some of our equipment providers like AVM and Devolo,” said Antes, and we hope to reach out to these and other chip makers and WiFi providers in a follow up article to do just that.

But our feeling is that this is really difficult to do, that WiFi tends to be chaotic, and that when WiFi goes wrong, it goes badly wrong, and with this added complexity, if there is more to go wrong, then it will. If you have a beam that is pointing at a remote client and the signal is not getting through and you are nulling the rest of the signal, an Access Point might tie up one or two antennas for some time, doing not very much.

But whether now or at some later stage, MU-MIMO will come into its own, and when that happens there is every chance that Qualcomm Atheros will still have a market lead.