Does MU-MIMO really expand Wi-Fi system capacity?
- 07 July, 2017 02:25
MU-MIMO enables multiple client stations to receive unique but simultaneous transmissions from a single access point (AP).
Until this advance, each associated station had to wait its turn via the normal contention process and subject to whatever vendor class/quality-of-service mechanisms might be implemented, available, and applied in a given installation.
Importantly, most clients, and especially highly mobile clients like handsets and tablets, are typically limited to only one or two MIMO streams. Consequently, clients typically have demands for throughput well below what can be provisioned in a single transmit cycle from a given AP. Given that Wave 2 enterprise-class 802.11ac APs usually implement three and increasingly four MIMO streams, significant additional network bandwidth and overall system capacity could thus be made available via MU-MIMO.
The question, of course, is how much of a gain can be expected. Most Wi-Fi implementations today enable literally dozens of tweaks to system settings, each with the potential to improve overall throughput and/or capacity, spread across APs, clients, drivers, and management consoles, and must also deal with the fundamental vagaries and uncertainty inherent in wireless communications of any form. It can therefore be difficult to evaluate the potential of any given enhancement, like MU-MIMO, under typical operating conditions. A more specialized testing environment is thus required.
So, for this test the objective was to determine what, if any, benefits might accrue from the application of MU-MIMO and to quantify these in an RF-isolated, repeatable test environment thus realizing the long-sought-after level-playing-field, apples-to-apples setting.
Recent advances in the state of the art of wireless testing and performance evaluation allowed us to configure exactly such a platform, and to perform the required testing and evaluation of the results quickly, repeatably, and efficiently – and with a high degree of confidence that our results do, in fact, provide guidance for the setting of both Wi-Fi customer and end-user expectations for MU-MIMO in production environments.
Test Environment and Procedures
Farpoint Group has been testing wireless products for more than a quarter of a century, with almost all of these tests performed in freespace – the open air in a given location and physical environment. Because the nature of RF propagation is non-deterministic, given such artifacts as multipath, building construction, and a number of forms of signal fading, we’ve always attempted to average-out externalities via a variety of procedural measures. Among these are monitoring tests using a spectrum analyzer (after an initial assessment of the state of the spectrum required using this equipment, of course), the use of turntables to eliminate antenna orientation at least on the battery-powered client end, and the averaging of multiple, relatively lengthy (1.5-3 minutes each) test runs, along with the rejection of any clearly anomalous results via substituting the results of yet another test run. While we believe such was indeed the best that could be done at the time and did in fact create a mostly level playing field, it was impossible to guarantee that such results would be repeatable in another physical environment, or even at another time, and consequently a degree of uncertainty existed no matter what steps were taken.
The above-noted state of the art in wireless testing has in recent years evolved significantly with the development of isolation chambers and a broad range of associated, highly programmable and instrumented test equipment. An isolation chamber is just that – a semi-anechoic environment completely isolated in terms of RF radiation from the outside world, and thus creating the much sought-after level playing field also noted above.
Placing devices under test (DUTs) in these chambers thus eliminates the variability inherent in freespace. Chambers can be interconnected with RF cables, enabling arbitrary test environments to be configured with relative ease. Variability in a given physical environment and the previously mentioned externalities are thus no longer factors in test results, and repeatability across runs as test and device settings and even the DUTs themselves are varied is guaranteed.
The test chambers we used for this testing are manufactured by octoScope, Inc. as part of their octoBox product line, shown below.
Additional elements from octoScope also used in this test were the PalTM 2 partner-device device emulators, which we used as the DUTs in this exercise to emulate both client devices and an 802.11ac Wave 2 access point. The Pal 2 is based on a Qualcomm Wi-Fi chipset, with custom drivers, firmware and related software, with all of this customization enabling a high degree of configurability and realized via a convenient browser-based user interface. We used one Pal 2 to emulate a single four-stream AP, with the only changes to its configuration between test runs being the enabling and disabling of MU-MIMO.
Three additional Pal 2 devices emulating single-stream clients were placed in a second octoBox, and separate tests were run with two and three clients enabled, each with MU-MIMO on and off. We used single-stream because we believe most clients will be configured this way to take maximum advantage of MU-MIMO optimizing for system–wide capacity over individual-device throughput. Here is a block diagram of the test configuration:
We thus believe this strategy is sufficient to demonstrate how well MU-MIMO might contribute to improved performance and system-wide capacity.
We used a beta release of octoScope’s new octoBox Software Suite (below screenshot), which unifies control of all octoScope products, including such elements as programmable attenuators, channel emulators, and related products, within a single browser-based interface.
It was thus very easy to specify, design, implement, and execute the testing required here, involving much less time and effort than the freespace approach and with much greater accuracy for the comparisons that are at the heart of the work here.
The octoBox Software Suite includes the popular iperf3 benchmarking tool for traffic generation, and for this testing we used 80-MHz. channels, both TCP and UDP (separate runs for each of course) as the transport layer protocols, specified unlimited bandwidth, and sent data in 128KB bursts (128KB packets) for one minute during each test run. RSSI was held at -41 dBm for each run (via tuning of the quadAtten unit), another major advantage of isolated testing environments.
Data from the first ten seconds of each test run were discarded to allow for any rate adaptation that might occur as associated clients begin to receive traffic. The following figure shows the output of a given test run, in this case three clients with MU-MIMO enabled.
The output traces shown to the right of the screen appear in real time as a given run progresses. The tables below contain the test results for all four runs using TCP and, separately, UDP.
As can be seen, and the bottom line here, a very meaningful improvement in performance was realized via the application of MU-MIMO in every case – again, no other changes to settings or configuration were made between runs.
Discussion and Conclusions
We do not, of course, mean to imply that any given combination of commercial APs and clients will realize the results we have presented here. As is always the case, specific combinations of endpoints within specific environments will yield a given set of throughput results that will likely vary with range, motion, time, and the nature of the specific physical environment.
But as our goal here was to examine the potential of MU-MIMO alone, we would have no reservations in suggesting MU-MIMO will indeed make a very significant contribution to maximizing the value of a given Wi-Fi installation. Also, MU-MIMO will likely reduce the need for additional capital investment in Wi-Fi infrastructure otherwise required in many cases to meet capacity objectives.
As the installed base of MU-MIMO clients builds – a process that will likely take two to four more years – the value of investments in 802.11ac Wave 2 infrastructure will be maximized. While 802.11ax will likely be backwards compatible with .11ac Wave 2, much of the pressure for immediate upgrades to that technology due to WLAN capacity limitations may be mitigated in many cases, extending the useful life of 802.11ac Wave 2 and maximizing RoI.
We expect that the MU-MIMO implementation in 802.11ax to be once again improved over the previous generation, so the staged/phased upgrades to this newer technology will also likely see excellent RoI over time, regardless.
The conclusion from this work, then, is that MU-MIMO is indeed going to yield very significant benefits on Wi-Fi networks that would otherwise be pushed to the limit. Perhaps the biggest gating items in realizing this vision now are upgrading infrastructure to 802.11ac Wave 2, as required, and the volume availability of appropriately equipped clients. Skepticism on the part of end-user organizations will, we believe, decrease rapidly from this point forward.
Even with a sophisticated test environment like the one we used here, it’s still important to consider the wide range of real-world scenarios that could further affect performance. These include roaming (which we did not test here, as such would have introduced only a brief period of no connectivity identical to that of non-MU-MIMO traffic), additional client loading, vendor-specific traffic prioritization and class of service capabilities, and similar concerns. Still, we expect MU-MIMO to yield a net benefit in essentially every venue where it’s put to work.
Special thanks to Fanny Mlinarsky and Andrew McGarry of octoScope for the use of the equipment involved and their assistance with this project.