It's been a long, slow slog for the engineers developing Wi-Fi 6 chipsets and products. I've been in the thick of it, trying to develop Wi-Fi 6 test methods that can show whether these draft-standard products deliver on the benefits they promise. That effort has been hampered by the fact that all the consumer Wi-Fi 6 products on the market to date have not had either OFDMA or AX MU-MIMO enabled.
That's right, despite what you read on their websites and product boxes, Wi-Fi 6 routers generally don't support key technologies that they advertise. The exceptions to this in the consumer Wi-Fi realm are NETGEAR's Qualcomm-based RAX120 and ASUS' Broadcom-based RT-AX88U. The former enabled OFDMA DL (downlink) only for both bands back in August, while ASUS pushed out a Beta firmware in late July that supports OFDMA DL in 5 GHz only.
MU-MIMO support has also generally been MIA for AX clients. When it is enabled, it will be for downlink only; AX MU-MIMO UL (uplink) won't be supported until a later revision of the 11ax spec.
My takeaway from this is that it's still too early to post meaningful results trying to determine whether OFDMA delivers on its promises of increasing total bandwidth capacity and reducing latency. So I'm going back to the basics for this article and just testing rate vs. range (throughput vs. attenuation). But unlike my first pass at this last December (First Peek At Wi-Fi 6: ASUS RT-AX88U & NETGEAR RAX80), this time I'm going to look at how AX routers perform using an AX client.
I tested the five draft AX routers for this go-around, shown in Table 1, three based on Broadcom's AX platform, one on Qualcomm's and one Intel based.
Both the ASUS RT-AX88U AX6000 Dual Band 802.11ax WiFi Router and NETGEAR Nighthawk AX8 8-Stream AX6000 WiFi Router are built on essentially the same Broadcom platform, i.e. a Broadcom BCM49408 64 bit quad-core @ 1.8 GHz processor and two BCM43684 4 stream an/ac/ax radios. So the most direct comparison can be made between these two products.
The ASUS ROG Rapture GT-AX11000 is the only tri-band router tested, but its Wi-Fi design is essentially the same as the RT-AX88U's. I didn't have NETGEAR's tri-band AX12 Nighthawk RAX200 on hand, which has essentially the same design.
The NETGEAR Nighthawk AX4 is the lone representitive of Intel's AX router efforts that I've been able to get my hands on. It's the only two-stream design in this test and the most aggressively priced at around $100 less than the others.
I didn't open up my samples, but gathered the information below from the soon-to-be departed WikiDevi and around the 'net.
Updated 9/23/19: Corrected values for RAX80.
64 bit quad-core @ 1.8 GHz
|1024 MB||256 MB||BCM43684 (x2)
4-stream an/ac/ax radio
64 bit quad-core @ 1.8 GHz
|1024 MB||256 MB||BCM43684 (x3)|
dual-core @ 800 MHz
|512 MB||256 MB||Intel WAV654 (x2)|
|$300||4/4||Broadcom BCM49408||1024 MB||512 MB||BCM43684 (x2)|
quad-core @ 2 GHz
|1024 MB||512 MB||Qualcomm QCN5054 (5 GHz x2)
Qualcomm QCN5024 (2.4 GHz)
Table 1: Component summary
How We Tested
The goal of this round of tests is to assess the performance seen by a two-stream AX (aka Wi-Fi 6) device by running classic rate vs. range (RvR) tests. All routers were set as follows:
- HE features enabled - This enables AX link rates and 1024 QAM modulation on both bands
- 2.4 GHz: Channel 6, 20 MHz bandwidth. Since most devices operate in crowded 2.4 GHz environments, I kept our traditional setting.
- 5 GHz: Channel 36, 20/40/80 MHz bandwidth
I did not run tests with 160 MHz bandwidth enabled on 5 GHz. Enabling 160 MHz requires using DFS channels, which aren't available in all countries or reliably where public safety (including airport) radar is in use. I also found using DFS channels can significantly increase 5 GHz association times because devices (STAs) can only passively scan for available APs. So instead of being able to send out a probe request and waiting for APs to reply, a STA using DFS channels must wait until an AP announces itself. You wouldn't think this would make much difference, but I found when switching from running 2.4 to 5 GHz tests, it took over a minute to find and associate to the router's 5 GHz SSID.
I did, however, also test the routers with an AC station. An octoScope Pal24 was used for the 2.4 GHz tests, set to use two streams with 20 MHz bandwidth. Since the Pal24 is configured to operate as an N device, this setting will limit the maximum link rate to 144 Mbps. An octoScope Pal5 was used for the 5 GHz tests, set to 80 MHz bandwidth to provide the maximum two-stream 802.11ac link rate of 867 Mbps.
Each router was centered on the turntable in an octoBox 38 test chamber, as shown below. This box is part of the TR-398 testbed and will likely be the one used once the next revision of the SNB wireless test process is finalized. Only the left-hand set of antennas was used in these tests.
ASUS GT-AX11000 in test chamber
I chose Intel's AX200 M.2 module as the Wi-Fi 6 STA (graciously supplied by Intel). Given it's the only game in town for laptop/notebook makers who don't want to embed a Wi-Fi chip directly on their board, the AX200 is likely to become the de facto Wi-Fi 6 laptop client and maybe SmallNetBuilder's Wi-Fi 6 standard test STA.
Unlike Broadcom's BCM4375 smartphone combo chip, which is used in Samsung's Galaxy S10 family and Apple's iPhone 11, the AX200 supports 160 MHz bandwidth. Although I chose not to test 5 GHz 160 MHz bandwidth performance this time, I want the option for future testing.
The last reason for choosing the AX200 was that it provides the option of using a cabled connection, maximizing signal level so that the maximum Wi-Fi 6 MCS11 can be reached. This equates to maximum two-stream 2.4 GHz link rates of 287 and 574 Mbps for 20 and 40 MHz bandwidths, respectively and 1201 and 2402 Mbps for 80 MHz and 160 MHz bandwidths in 5 GHz. So with our test bandwidths of 20 MHz @ 2.4 GHz and 80 MHz @ 5 GHz, these represent link rate increases of 2X for 2.4 GHz (287 Mbps / 144 Mbps) and 1.38X for 5 GHz (1201 Mbps / 867 Mbps).
Aaeon pico ITX board with Intel AX200
The AX200 is hosted on an Aaeon PICO-APL4-A10-0003 board, shown above. This is a pico-ITX format board with an Intel N4200 Pentium processor, 4 GB of RAM and 256 GB SSD, loaded with both Windows 10 Pro and Linux Mint 19.2. Although I prefer to run Linux to access more radio statistics, I found a significant uplink throughput problem that Intel is in the process of fixing. The AX200 Windows drivers are in much better shape, but you'll see there is perhaps still some interoperability work to be done. I used the latest 18.104.22.168 driver.
The Rate vs. Range (RvR) test shows how throughput (rate) changes as signal level is varied (range). Instead of using physical distance to vary the signal by walking a laptop around, the router is placed in an RF-tight anechoic chamber (the octoBox) and programmable RF attenuators are used to change the signal. The diagram below shows the test setup.
RvR test setup
For 2.4 GHz, the attenuation was ramped from 18 to 81 dB in 3 dB steps; for 5 GHz, the range was 0 to 54 dB in 3 dB steps. Each test ran five simultaneous connections of unlimited rate TCP/IP traffic for 35 seconds, with the first five seconds discarded from the test average at each attenuation level. The 18 dB of starting attenuation for the 2.4 GHz tests was necessary to avoid overloading the router, Pal or AX200 radios. The router under test was rotated 360 degrees during each test run and the router and STA were disassociated and reassociated between downlink and uplink test runs.
The test sequence was:
- 2.4 GHz @ 20 MHz downlink
- 2.4 GHz @ 20 MHz uplink
- 5 GHz @ 80 MHz downlink
- 5 GHz @ 80 MHz uplink
The published results represent single test runs. When results looked odd, selected runs were repeated, with best obtained results shown.