ESP32 WiFi link quality measurement

Other Hardware 3rd Party Boards
1

Most arduiners evaluate WiFi (or LoRa) signal quality in a very simple way: they look at RSSI.
The larger the value (it is negative so closer to zero), the better the signal.

Typically, the standard WiFi library is used for this. It can scan available networks and return their RSSI values. Based on this RSSI, the program decides whether it should connect to the network or not.

However, this approach is not entirely correct.

RSSI only shows the signal level, but says nothing about the noise level.
In reality, the signal-to-noise ratio (SNR) is what determines the actual link quality.

Measuring the RF noise floor on ESP32

Fortunately, on ESP32, ESP32-S3 and ESP32-S2 (other chips were not tested) there is an internal ROM function that measures the RF Noise Floor.

The function is not declared in the official headers, but it is present in ROM/linker scripts and can be used like this:

extern "C" {
  int rom_check_noise_floor(void);
}

The return value is in 1/4 dBm units. (result must divided by 4)

If WiFi is not initialized, the function returns a positive value (>0).

Link quality estimation

could be done like this (pseudo-code)

int rssi = ...;                       // read RSSI from WiFi library
int nf = rom_check_noise_floor() / 4; // convert noise floor to dBm

int snr = rssi - nf;                  // calculate signal-to-noise ratio



| SNR      | Link quality   |
| -------- | -------------- |
| 25–40 dB | excellent      |
| ~20 dB   | good           |
| 10–15 dB | poor           |
| <10 dB   | near the limit |

This was tested on an ESP32-S3 using a simple WiFi example sketch from the WiFi library and several nearby access points.

Hope this helps.

2

For sure, which is the main reason Arduiners use the SNR reading produced by a LoRa device as the important indication of proximity to link failure.

3

Deeper analysis showed that using ram_check_noise_floor() instead is better.

It is a patched version of the ROM function. It is shipped as part of the WiFi driver binary blob.

4

Even deeper analysis showed that sometimes (rare occurences) NoiseFloor reads as a positive number.

In this case the result must be negated manually OR it can be re-read. The sign (positive sign) has a meaning: PHY driver reports that "Noise floor measurement took longer than 10 milliseconds". In really noisy environments sometimes it takes more than 10 ms to measure.