Hi @davefes,

Thanks for noticing this, it's a bug. It should be based on this section of the SX1276 datasheet:

Meaning the implementation should be something like:

# units: dBm
rx_packet.rssi = self._reg_read(_REG_PKT_RSSI_VALUE) - (157 if self._rf_freq_hz > 525_000_000 else 164)

Are you in a position to test this version and submit a PR to change it? If not, I'll get to it as soon as I have time.

Not until the weekend. Submit a PR ... it would be a first for me. As long as it works it's way into library I am OK with the "under-reporting".

I did check the RSSI at the receiver increased by turning off PA_BOOST at the receiver. Of course I didn't get an ACK at the sender. Another odd behaviour noted: the first RSSI after a re-boot is 5dB better then subsequent RSSI readings.

Thanks @davefes! Will leave this issue open until there's a fix.

Another odd behaviour noted: the first RSSI after a re-boot is 5dB better then subsequent RSSI readings.

Interesting. I assume that's probably an artifact of the module, although I don't know what it would be down to.

Change VALUE to VAL
rx_packet.rssi = self._reg_read(_REG_PKT_RSSI_VAL) - (157 if self._rf_freq_hz > 525_000_000 else 164)
and RSSI values around -117dBm are when errors started appearing. Still see 4 or 5dB higher on the first RSSI reading. I am sure I haven't seen this happen on other implementations, but I will go back and check.

Edit:
just re-checked Wei's implementation and it reads -123 or -124dBm (at onset of errors) after a re-boot and that stays the same after that. The distance between RX and TX remained the same, so TX power and RX sensitivity appear to be identical with both S/Ws.

I believe you are both using the same RSSI register value. I did notice in the datasheet that there is another RSSI register:

3. As signal strength increases (RSSI>-100dBm), the linearity of PacketRssi is not guaranteed and results will diverge
   from the ideal 1dB/dB ideal curve. When very good RSSI precision is required over the whole dynamic range of the
   receiver, two options are proposed:

Rssi in RegRssiValue offers better linearity. Rssi can be sampled during the reception of the payload (between
ValidHeader and RxDone IRQ), and used to extract a more high-signal RSSI measurement
When SNR>=0, the standard formula can be adjusted to correct the slope: RSSI = -157+16/15 * PacketRssi (or RSSI = -
164+16/15 * PacketRssi)

Practically-speaking, I think the most useful aspect of RSSI is to give the user some rough idea of the excess link budget one has. If I see errors at -117dBm and my RSSI is -100dBm then I know I have ~17dB of margin.

Confirm that RSSI is the same on the first loop and subsequent loops with your chips ... then I'll just say it is problem with Hope Radio RF96 chips and I will close this issue.