MIMO and some history...

[sfx2000]

This is a great summary -- back in the Wireless Holy Wars, I was working on 3GPP2 as well as IEEE 802.16 and 802.11

 

[user114514]

I want to understand what limits the further development of WiFi MIMO. STAs remain at 2x2, and APs stay at 4x4. I suspect economic factors are the dominant reason.
I won’t pretend to fully understand the deep mathematics behind MIMO, but from video explanations, it’s obvious that bandwidth and MIMO are the only two factors that can almost linearly increase capacity, while SNR is a painfully logarithmic factor.
I know many people say STAs stick to 2x2 due to power or size constraints, but if the target throughput is fixed (say, 1Gbps), then a 4x4 configuration should allow achieving the same throughput with a lower SNR. This means transmit power could be greatly reduced, since MIMO offers near-linear gain while SNR gain is only logarithmic. As for idle power, SMPS exists to handle that. Regarding size constraints, the fact that LTE UEs have widely adopted 4x4 proves that size isn't really a limiting factor — most LTE bands have much longer wavelengths than WiFi bands.

 

[thiggins]

Part of it is economics. Phones have limited real estate and there is a lot of competition for space and power consumption.

But MIMO is old news, OFDMA and now MLO are newer ways to increase bandwidth. Plus, new acronyms to plaster on consumer boxes drive the replacement cycle.

 

[RMerlin]

Wifi 8 is also expected to lean in that same direction (stability rather than more speed):

Wi-Fi 8 won't improve transfer speeds — but the new standard will enhance reliability and user experience

User experience is the key.

www.tomshardware.com

 

[user114514]

Part of it is economics. Phones have limited real estate and there is a lot of competition for space and power consumption.

But MIMO is old news, OFDMA and now MLO are newer ways to increase bandwidth. Plus, new acronyms to plaster on consumer boxes drive the replacement cycle.

My experiments with my own AP show that OFDMA helps control peak latency, but has no obvious impact on throughput.

MLO is theoretically a major breakthrough, but in reality it is not that powerful.
As is well known, MLO has two modes: EMLSR and STR.
My understanding of MLO is as follows:

1. EMLSR enables fast switching across different frequency bands on the same AP and can avoid roaming and the WPA 4-way handshake. This is essentially an enhanced version of MBO.
2. EMLSR will always dynamically monitor multiple links and select the best link for communication. The AP can also recommend that the EMLSR MLD switch the active link and renegotiate the link. This is similar to an improved version of BTM.
3. STR means “full duplex.” STR can also be combined with EMLSR, such as STR(2.4G, EMLSR(5GL, 5GH), EMLSR(6GL, 6GH)).

However, current vendor implementations of MLO are imperfect.

1. iPhone and Pixel devices with bcm4390 and bcm4398 only establish EMLSR links but do not perform multi-link monitoring and do not accept steering requests from the AP. I call this pseudo-EMLSR. Unless the main link breaks for some reason, they will never switch to another link.
2. Many APs still disable dynamic MLO link reconfiguration (for example, many Qualcomm APs have mlme_mlo_reconfig_reassoc_enable=0), making MLO’s flexible switching meaningless. For example, when you return home and first detect 2.4GHz but 5GHz/6GHz are still not visible, the STA will immediately associate with 2.4GHz but will not reconfigure to establish MLO association even after 5GHz enters the range.
3. STR, as a MAC-layer aggregation mechanism, even if supporting asynchronous transmit and receive, is fundamentally different from PHY-layer asynchronous transmit and receive in full-duplex Ethernet. Each STR link at the lower layer is still “half duplex”-like, so the latency advantage of STR is not as strong as advertised.
4. Many STAs only support STR in 2.4GHz + 5/6GHz, and 2.4GHz does not even support 40MHz or 1024QAM. Such a huge throughput disparity renders STR almost meaningless. The scheduling overhead of MLO itself can even offset any throughput gains from EHT20 MCS7. For the reasons mentioned earlier, using it as an EMLSR link is also not very effective. Qualcomm released the “HBS” MLO with WCN7850, allowing STR MLO between high-frequency bands 5GHz and 6GHz: 5GL+5GH, 5GL+6GL/H (6GL+6GH is not feasible due to hardware filtering limits). HBS allows 160MHz+160MHz MLO, but it has a fatal flaw: the maximum aggregated bandwidth is capped at 320MHz, so 160MHz+320MHz is impossible. This means no additional bandwidth advantage compared to non-HBS, with a maximum of 320MHz, but requiring two extra high-frequency FEMs and two HBS duplexers. This directly caused all 8Gen2/3 phones, although equipped with WCN7850, to not support HBS due to lacking FEMs. Qualcomm did not solve this hardware limitation in the subsequent WCN7880/7860, resulting in no 8Elite products supporting HBS. Therefore, the NCM865 released in 2022 is almost the only product supporting HBS, with no follow-up products supporting it. I can claim HBS is dead.
5. Combining EMLSR + STR. Broadcom’s bcm4390 seems to support STR(2.4G, EMLSR(5G/6G, 5G/6G)) combinations, but due to the aforementioned problems, it is practically meaningless. Qualcomm’s latest Linux driver supports STR(2.4G, 5GL, EMLSR(5GH, 6G)), but because of the 320MHz total bandwidth cap and HBS’s failure, it remains practically meaningless. Moreover, due to Android GRF, existing devices will not receive this driver update. Although NCM865 supports HBS, Windows driver development enthusiasm is low (you need to wait half a year for a driver update) and it does not support the feature.
6. Assuming HBS MLO is resurrected and popularized in the future, EMLSR will become meaningless, because it is expected that under heavy load all links will become congested, and under light load EMLSR is unnecessary anyway.

From the above, we can see that current MLO is almost a joke, with a failure degree even more exaggerated than DL MU in the 11ac era.

 

[sfx2000]

Part of it is economics. Phones have limited real estate and there is a lot of competition for space and power consumption.

But MIMO is old news, OFDMA and now MLO are newer ways to increase bandwidth. Plus, new acronyms to plaster on consumer boxes drive the replacement cycle.

True - marketing will always push the theoretical rates, and they've been doing this for years...

Yes, MIMO is old news, but it is still the major driver on performance - OFDMA is more about efficiency/capacity than overall performance.

MLO has potential, as it's similar to 4G/LTE-Advance and 5G-NR for what they refer to as Channel Aggregation - CA for short...

I'm still of the opinion of the following:

1) 5GHz - 802.11ac Wave 1 - hands down the best improvement in Wifi overall, everything else has been incremental
2) 6GHz - 802.11ax - if available, it's quite nice, and 6GHz is zero legacy for the most part even for WiFi7
3) 2.4GHz - 802.11ax - This is a standards based big jump from 11n (WiFi4) to 11ax (WiFi6) - one can get surprising results here if done right (e.g. delete 11b compat)

Going back to the economics - good/fast/cheap - pick any two - and this is kind of where many of the client chipsets live in the good/cheap realm - Intel is a good example here with their BExxx cards...

 

[thiggins]

My experiments with my own AP show that OFDMA helps control peak latency, but has no obvious impact on throughput.

MLO is theoretically a major breakthrough, but in reality it is not that powerful.
As is well known, MLO has two modes: EMLSR and STR.
My understanding of MLO is as follows:….

Thanks for the insight. Too bad this is in a thread about MIMO! (Not a criticism. Just makes the info harder to find.