Readers risk being either bored or frustrated with Multi-User MIMO discussions that we have had over the past few weeks, but we will risk one more visit to put the subject into perspective.
It all began when Qualcomm announced last Spring its first chip that supports MU-MIMO and has recently introduced a wider family of these chips and is unrepentantly claiming that the throughput when talking to existing handsets is 3 times faster than existing WiFi chips.
No-one else has MU-MIMO as yet and all the market movement has come from companies like Quantenna and Celeno, in making chips that are 4 x 4 MIMO which demonstrate higher throughputs, in more challenging circumstances. Celeno added a 4 x 4 MIMO chip in September while Quantenna has even added an 8 x 8 chip which is sampling.
Broadcom so far has not covered these moves and yet it remains the chip company of choice within WiFi, and Atheros the WiFi chip builder acquired by Qualcomm, has played second fiddle to Broadcom from the start of the WiFi market.
The question of MU-MIMO is important because there is the potential for a mistake that concedes the entire WiFi market to Qualcomm or to a Chinese chip firm. We would typify Broadcom as the company that squeezes cost out of a BOM (Bill of Materials) in chips, and takes a chip to commoditization, and Qualcomm as the pioneer chip-maker that will both invent new technologies and also make the first effective chip for a market. If Qualcomm has genuinely “cracked” this market it is a big challenge for Broadcom and others.
So this week we talked to Manny Patel, Director of WLAN Business Development at Broadcom, asking when he thought MU-MIMO would begin to shape the market.
“When we do a simulation on MU-MIMO and assume no processing overhead and perfect signal processing, then the type of claims that Qualcomm is making are in the right order of magnitude,” Patel begins.
“But those claims don’t make it into practical reality. And these calculations rely on most clients chips out there being 1 x 1 chips, and that is most definitely changing. MU-MIMO works best when there is no extra receive chain to allow for diversity on the receive chips.
“This is not true, as Qualcomm claims, when you begin to work with clients that are 2 x 2. Most new clients coming out now, except for the latest iPhone 6, are all 2 x 2 WiFi clients, with 2 antennas and 2 radios. This all changed with the Galaxy S5.
“And that means that the Qualcomm claims are not working from a fair reference point. All this is well understood from an academic perspective in doing the mathematics to simulate the outcome.
“At Broadcom, we believe that there will be benefits from MU-MIMO and it is still on our silicon roadmap, but we think about it differently from Qualcomm, and will use a different approach.” Here we tried to get more out of Patel about the way Broadcom will approach this market, but he said he could not go into the simulations he has done in detail, or the math, without giving away silicon plans and he wasn’t ready to do that.
“In a 1 x 1 environment the math says that the performance will improve over normal SU MIMO by 1.75 times up to 2 times, depending upon the quality of the signal. But a chip in this situation is really stretched when it is asked to sustain 3 signals if the signal to noise ratio is poor, as when WiFi is connecting through a thick wall or competing with other signals around it.”
Patel is convinced that doing MU-MIMO mostly for 1 x 1 clients results in 3 clients being the optimal number of simultaneously signals to support. “The math changes when you work with 2 x 2 clients when it becomes better to work with groups of 4 clients.”
When speaking to 1 x 1 clients we use a type of beamforming, where you send a stronger signal to the target clients and null the signal going elsewhere. What you are doing with an N minus 1 approach (4 signals less one is 3) is you have one degree of freedom (one spare antenna) to help strengthen the signals, beamforming style, to each of the three targets. When you have 2 x 2 clients, that extra degree of freedom is in the client and makes for better beams.
“If you have 5 or 6 clients, there is still a calculation of which 3 clients or in the case of 2 x 2 clients, which 4 clients to support first, and which clients do you support sequentially,” and he is right that in many US homes there are already a dozen or so WiFi devices and certainly there can be more than 4 used at once.
Patel says that multiple antennas are the first thing you need to improve performance, citing beamforming and diversity as two effects which then come into play and he points out, “When we put MU-MIMO into the 802.11ac standard, we took that decision at a time when the world was full of 1 x 1 clients.
Another thing that affects the performance is the relative orientation of the clients devices to one another. “It’s better if they are further apart,” says Patel “and when we test like against like” this factor has to be controlled.
“Also if you have a bad apple problem (when a single constrained client takes up too much time to receive a weak, interference prone signal) then under low SNR (signal to noise ratio) the benefits of MU-MIMO drop off dramatically and there is practically no benefit. So bad apples cause as much of a problem to MU-MIMO as they do to SU-MIMO chips.
The Wireless Broadband Alliance is talking about MU-MIMO coming late in 2015, which is why there will be no certification for MU-MIMO prior to March 2016.
That all begs the question, will an operator install Access Points based on MU-MIMO when the market has a single source, and there has been no impartial interoperability testing between the MU MIMO APs and other clients devices?
The truth is that if the market leader proved that its chips would work with a MU-MIMO access point, then perhaps MU-MIMO would make it into tenders prior to March 2016. But until then this looks like it will only be used by existing Atheros clients, to prepare for a post-2016 marketplace.
We await a significant deployment win for the Atheros chip in a major operator Access Point, and suspect that it will be rather a long time coming.