Hi @charles2, thanks for looking into this. Following up on my post above, here are answers for my GL-MT3000 (Beryl AX), with before/after measurements. I log the router's WiFi figures every minute, so I have the day before the upgrade on 4.9.0 and eight days on 4.11.0, for the same clients in the same places.
Note: I used Claude AI to help collect and format this message, but I'm a real person and this is real data from my smarthome. I'm a retired software engineer with 35 years of experience and lover of gadgets and smart home stuff. 
Your questions
Distance. All of these clients stay in one place, except the watch:
| Client |
Band |
Distance from the router |
In between |
| Single-board computer (Orange Pi Zero 2W, Linux) |
5 GHz |
about 4 m |
nothing, line of sight |
| Air-quality monitor (ESP32-C3) |
2.4 GHz |
about 4 m, beside the computer |
nothing, line of sight |
| Smart display |
5 GHz |
about 3 m, in the opposite direction |
one wall |
| IP cameras A and B |
2.4 GHz |
one about 3 m, beside the display; the other about 5 m, 2 m beyond it |
one wall |
| Smartwatch |
5 GHz |
worn, so it moves around |
|
I haven't checked which of the two cameras is the nearer one.
At 4 m with nothing in between, the air monitor's −75 dBm on 4.11.0 is far too weak, and its −19 dBm on 4.9.0 is too strong. The driver's best antenna reads −38 dBm, the same as the device's own reading.
Antennas. Both antennas are upright, and the unit sits flat. I haven't touched the router at all, before, during or after the upgrade: it wasn't moved, and nothing was unplugged or adjusted. I upgraded from the web admin page, by accepting its pop-up offering the update.
What changed at the upgrade (4.9.0 → 4.11.0, 29 September)
These are medians of the iwinfo assoclist signal for the 24 hours before the upgrade and the 8 days after it, with the driver's own figure and the client's own reading:
| Client |
Band |
iwinfo, 4.9.0 |
iwinfo, 4.11.0 |
Change |
Driver table, best antenna (4.11.0) |
Client's own reading, before → after |
| Single-board computer |
5 GHz |
−28 |
−73 |
−45 |
−59 |
−37 → −40 |
| Smart display |
5 GHz |
−31 |
−76 |
−45 |
−62 |
– |
| Smartwatch |
5 GHz |
−29 |
−73 |
−44 |
−59 |
– |
| Air-quality monitor |
2.4 GHz |
−19 |
−75 |
−56 |
−38 |
−39 → −38 |
| IP camera A |
2.4 GHz |
−22 |
−78 |
−56 |
−45 |
– |
| IP camera B |
2.4 GHz |
−26 |
−82 |
−56 |
−54 |
– |
All figures are in dBm, and the changes are in dB.
Nothing on the radio side changed:
- The clients' own readings didn't move. The single-board computer logs
iw dev wlan0 link every 30 s, and the air monitor reports its own RSSI. Hour by hour, neither shows a step at the upgrade.
- Link rates are the same:
- 433 Mbit/s (VHT-MCS 9, 80 MHz) for the computer and the display;
- 72 Mbit/s for the air monitor;
- 129–143 Mbit/s for the cameras.
- Pings are the same. The computer pings the router every second: 0.006 % loss and a 1.5 ms median before, 0.002 % and 1.5 ms after.
- The radio settings in
/etc/config/wireless are identical before and after: 2.4 GHz on channel 11 at 20 MHz, 5 GHz on channel 36 at 80 MHz, country US.
What I think the bug is
On 4.11.0, the per-client signal that iwinfo shows is the average of the four RSSI0/1/2/3 slots in the driver's station table. The average includes the -109 placeholder for antennas the radio doesn't have, and it is rounded towards zero. I took two snapshots a minute apart. All 7 clients matched to the dB in both, 14 of 14 readings. Four examples:
iwpriv stainfo RSSI0/1/2/3 band sum / 4 iwinfo assoclist
-50/-39/-109/-109 2.4 GHz -76.75 -> -76 -76 dBm
-54/-52/-109/-109 2.4 GHz -81.00 -> -81 -81 dBm
-58/-66/-56/-109 5 GHz -72.25 -> -72 -72 dBm
-56/-55/-58/-109 5 GHz -69.50 -> -69 -69 dBm
The size of the error depends on the band. On this MT3000 the table shows two live antennas on 2.4 GHz and three on 5 GHz. So 2.4 GHz has two placeholders in the average and 5 GHz has one:
- On 2.4 GHz, 4.11.0 reads 28–37 dB below the best antenna.
- On 5 GHz, it reads 14 dB below.
A band with four live antennas would have no placeholder in its average. That might be why your MT6000 showed a smaller difference.
4.9.0 erred the other way. It read 19–31 dB stronger than the best antenna reads now. Its figures fit (antenna 0 + antenna 1) / 4 to within about 3 dB for every client, as if the other slots counted as 0. I didn't save a driver table on 4.9.0, though, so this part is a guess.
On the antennas: every live antenna slot gives a normal level for every client, −37 to −69 dBm in every table I looked at. I see no sign of an antenna problem.
Two smaller things, in case they help:
- On 4.11.0 the assoclist prints
unknown (SNR -76). The noise is unknown, so the "SNR" is just the signal.
- iwinfo's TX rate to both cameras read 6 Mbit/s on 4.9.0 and 143 Mbit/s on 4.11.0. Their RX rates didn't change.
Workaround
I now read the signal from the driver's table instead, using the strongest live antenna per client and ignoring −109:
iwpriv rax0 show stainfo >/dev/null # on the MT3000, one call lists the clients of both bands
dmesg | tail -n 300 | grep -oE '([0-9A-Fa-f]{2}:){5}[0-9A-Fa-f]{2} +STA( +[0-9]+){6} +-?[0-9]+(/-?[0-9]+)+'
The last field is RSSI0/1/2/3, and the newest table is at the bottom.
Let me know if there's anything else I can provide to help diagnose the issue!