Moo-MIMO

This reply-to-self may seem strange, but it has culminated from a extremely beneficial conversation with a very good (and extremely knowledgeable) friend in the industry. My interpretation of my takeaways from our conversation around the topics within this blog are listed below in a Q&A-to-Self. I would like to profusely thank this unnamed (at his request) engineer for his time, effort, and insights in contributing to the accuracy and value of this blog.

Q1: Why would we worry about a device having WiFi Alliance certification for TxBF?

A1: I would think that TxBF is like much else when it comes to interoperability, and I would even guess that given the complexity of it, that it would be very important to get it certified. It is already part of the WiFi Alliance’s certification as best I can tell, but it’s optional as you’d expect it to be. I would assume that the WiFi Alliance added it to their list of things to certify because they thought it would be hard to make it interoperable otherwise.

Q2: There’s no harm, in short-range uses cases, in adding additional SNR due to using TxBF, even when it’s already high.

A2: In general, I agree. Specifically my concern is the per-client airtime overhead (and possibly latency for mobile voice devices) that would be introduced for very little gain.

Q3: At longer ranges (however you may want to define “longer”), TxBF seems to add more value than anticipated. There is a small amount of testing data showing this.

A3: I would LOVE to see that data. Since there’s a bunch of marketing hype around TxBF, the WiFi Alliance has an optional certification for it, and MU-MIMO is predicated on TxBF being supported on both sides of a link, there’s some importance in understanding what is and isn’t factual about TxBF operation and overhead. I will add that we don’t actually want clients to be at longer ranges, which is why we are always suggesting that APs support a minimum basic rate of either 12 or 24Mbps and why we have infrastructure features that instruct the AP not to talk to clients with an SNR below X dB. Therefore, even if TxBF does work better than expected at longer ranges, it becomes a case of “just because you can doesn’t mean you should.”

Q4: TxBF overhead isn’t that large, so at longer ranges, the benefits outweigh the overhead.

A4: I don’t have any hard data (only word-of-mouth assertions I’ve heard in the past from esteemed engineers) to set a baseline for “large” or “small” overhead, and I would guess that the amount of overhead depends on the total amount of traffic moving. The biggest conceivable negative impact that I see is that in high density environments, even a little extra utilization from each client would result in much higher overall channel utilization. I guess this one comes down to “how much extra data would be “normal”?

Q5: There are a few clients that are already on the market that can have their firmware upgraded to support MU-MIMO, but even if there were none, we probably want to plan for client capabilities that are ~1 year down the road, and in 1 year, there will surely be lots of MU-MIMO capable client devices.

A5: I can’t factually argue what will or won’t be here in a year, but I hope that there are MU-MIMO clients readily available within a year so that we can capture meaningful data around the value of MU-MIMO in various environments. Today, the value is either zero or close to it. In a year, we’ll see.

Q6: 40MHz channels are OK.

A6: My thoughts on using 20MHz vs 40MHz channels has to do with two very specific things: 1) density of client devices, 2) size of network. More channels mean more separated collision domains, which equates to more simultaneous conversations, which equates to more system capacity. If you use a 40MHz channel, you have essentially borrowed from system-wide capacity in order to give more capacity to a single BSS. When the client device density is high, you want more channels. When the client device density is low, you could feasibly move to 40MHz channels – but with one major caveat: size of network. Significantly minimizing CCI in a multi-floor building, large building/campus, or the like requires as many channels as you can use in the channel re-use plan. The less channels in your channel re-use plan, the harder it is to get rid of CCI, and CCI is a HUGE capacity killer. In a small-to-medium sized network, 40MHz may be fine, but in larger networks, no way, no how. Splitting your collision domains offers far more system capacity.

Q7: The 4th transceiver (transmitter/receiver) is more beneficial for control than for extra throughput on a very few clients (in the future). 4SS SU-MIMO is not that interesting. Also, as long as the 4th transceiver doesn’t affect your power budget, you’re in good shape.

A7: Couldn’t agree more. Getting an extra (theoretical) 1.23dB of gain on the downlink is a good thing (which comes from 6dB (for 4Tx) – 4.77dB (for 3Tx) = 1.23dB), especially when it’s increasingly hard to wrangle an extra dB or two from antennas. TxBF seems to be important at mid-range for the same reason, provided the overhead isn’t large. Every extra dB helps to some degree. Totally agree about the power budget.

Q8: 256QAM is the the same for both Wave-1 and Wave-2, and if the additional benefits of Wave-2 (e.g. TxBF, better Rx Sensitivity, 4 Transceivers, etc) add enough value, then we could increase SNR and therefore get a little extra range out of 256QAM.

A8: Yes, agree, but unsure as to how much extra range or under what circumstances (e.g. type of client, RF environment, etc). This is hit-or-miss, but if it adds value, it does. I just wouldn’t want to pay for it given that it’s a complete unknown.

Q9: While it’s agreed that many networks are very poorly designed and configured — and rarely tested — it’s important to have the infrastructure add as much value as it can so that these poorly deployed networks perform as well as possible under the circumstances.

A9: I can’t argue against that. Give me all of the benefit you can from the infrastructure. However, the analogy I would use to put it into perspective is that you’re putting a new Ferrari in the hands of a 10 year old who has never driven a car. The capabilities are there, but the knowledge of how to use them isn’t.

Q10: There will be a significant differential on how well vendors implement MU-MIMO, which is arguably the best part of Wave-2.

A10: Yes, agreed. IF it adds value in normal use cases, then I can see that the complexities therein could be handled to different levels of expertise by different vendors. How well various vendors handle MU groups with varying degrees of orthogonality will be a big one.

Q11: Even if the benefit of MU transmissions is only something modest, like 30% (which of course wouldn’t apply to all situations), then as a zero-cost feature it’s worth it. NOTE: “Zero-cost” means that you don’t have to trade off some other benefit to get it).

A11: I agree that it’s zero-cost from a technical standpoint, and that’s a fair point. However, I don’t see this as a zero-cost (monetarily speaking) since Wave-1 APs are cheaper than Wave-2 APs. I agree with the premise that free extra capacity is always a good thing though.

Q12: There are still many unknowns when it comes to queuing issues, but the hope is that in client environments where 50-75% of the clients are MU capable, that we’ll see at least a 1.5X capacity gain, and perhaps in downlink-heavy environments with mobile devices, as much as 2X gain.

A12: I hope you’re right. I’m not naysaying for the heck of it, but I am skeptical. If we see a 1.5X or greater overall capacity, I’d be very happy about it. The main issue I still see though is that even though infrastructure manufacturers are giving their best to bring us new technology that has the possibility of better performance, implementations are so poor (in general) that many folks who purchase these systems will simply not see those benefits. 🙁 I would also like to add that MU-MIMO won’t be useful in many vertical markets because those markets either have: 1) legacy clients, 2) no HD or VHD requirements, and/or 3) no MU capable clients (e.g. healthcare). That will further narrow the use cases and target market for Wave-2 systems.

Q13: The normal lifetime of a WiFi network is 3-5 years.

A13: Some salient points to consider about the current and on-going value of Wave-2 could include:

MU-MIMO is the only possibly meaningful upgrade of Wave-2, and it’s going to take a year to get a reasonable amount of MU-MIMO clients on the market
We have to wait nearly a year for DFS certification (of Wave-2 products) to have reasonable system-wide capacity (due to more channels being available for use)

Therefore, the beneficial time period of Wave-2 APs is reduced (20 – 33%), which makes it even harder to justify a Wave-2 purchase (due to the price premium).

Q14: Is there any other value in buying Wave-2 APs that you can think of?

A14: The place where I find real value is in the fact that vendors launch new features alongside new high-end hardware. Those features may get put into older/lesser platforms at a later date, but when launching new high-end hardware, they add such special features in order to more significantly differentiate their existing high-end platform(s) from the brand new high-end platform.