Wifi performance poor despite strong (-9) RSSI, how debug/diagnose?

[WiFive]

Yeah so if you are seeing this problem on only some units the likely possibility is poor assembly of that bodge. Also the crystal load caps look like maybe they could be better placed/routed and the antenna looks like it is sort of shifted off it's pads during reflow.

[novalight]

It's not a crystal, it's an oscillator (thus no load caps, only one decoupling). I've verified this botch. The antenna here was hand assembled, the other units are machine assembled.
Since I tested all units in the lab I am not sure if this could even be the issue (especially the oscillator thing, since the ESP works normal otherwise).

[ESP_Angus]

40MHz oscillator?

At least once we have seen similar symptoms caused by clock stability hardware issues. The ESP32 locks onto a slightly wrong frequency, due to the input frequency being off or unstable. The ESP32 is quite liberal in what it accepts (hence the strong RSSI)[*] but perhaps the AP is not so liberal in what it accepts so the communication struggles.

If you had access to the hardware install I'd suggest either a spectrum analyzer or (much cheaper option) running a WiFi interface in Monitor/raw capture mode to get more of the picture here.[**]

[*] Or possibly it only receives a few frames with strong RSSI, and is dropping the rest.

[**] Raw capture mode will not give you the full picture, but if you, for example, see the ESP32 constantly having to retry frames that the AP hasn't ACKed then you can guess that probably the issue is with the ESP32->AP transmissions. Or vice versa.

[novalight]

That is why we shifted to using oscillators, since our products experience a wide temperature range. This one has a maximum error of 2ppm +-2ppm over temp range if I remember correctly.

[ESP_Angus]

Ah, I remember this thread now. That's strange.

To double-check, you have the same hardware design deployed in multiple locations. In some the RF performance is normal but in some it's degraded?

[novalight]

That's correct!

[WiFive]

Yeah the bodge would still be my #1 suspect and digital logic running fine doesn't say anything about rf.

[novalight]

Mhh, I see this oscillator thing more like an "all or nothing"-problem.
I mean it's a digital signal running through a 4-5mm bodge wire. The Vcc and GND connections are fine and it's properly decoupled.
The 40 MHz Signal has slew rate control, so it should not radiate any harmonics and the short wire should also not radiate the 40 MHz too much.
So as soon as I get to it I will test some things in the field, but that can be in one or two months. So I would appreciate anything that can be diagnosed remotely.

I think it would be helpful in general to have more info on RF-calibration (internal and maybe external) and WiFi performance out there, especially for people designing their own boards. So it would be interesting to know how you handle these things internally when you design boards and also how the RF-cal works. If some of these also fall under some kind of NDA I would be happy to sign it if that's an option.

[ESP_Alvin]

Hi, Novalight

Please see some feedbacks as below.

1. There is no issue for RX of ESP32 at RSSI (-9~-20), and there is high chance that your board may have some issues.
2. Do you have a wifi sniffer which could be used to catch the packet of the process for the connection? If you could provide the captured packets, we can help analysis.

Thanks.