On professional VNAs, stimulus source is calibrated for flatness on factory with a bolometer. On nanoVNA you can't control stimulus output power, so we should probably calibrate the audio codec gain to get most of it's dynamic range at each frequency, or at least for each band. However multiple mixing products, others than the desired one, could be present at audio frequencies, so this must be done with care.
This could also lead to a debate about needed headroom to measure DUTs with gain. IMHO, it would be better to sacrifice having a lot of headroom and just insert an attenuator if an amplifier S21 is to be measured.
I have not taken a look to the code in much detail yet, maybe this is being done, but probably not. I have read somewhere that people have measured their LNA's gain, so current code must be setting the codec gain too low if it has so much headroom and the S21 70-80dB noise floor we are seeing could be in fact being limited by the codec dynamic range.
I also miss an averaging mode, as some measurements are quite noisy, and it would help for duplexer adjustments with notches near the VNA noise floor. Maybe just a moving average low pass filter with just 2 taps to avoid increasing memory requirements (new=0.9*last+0.1*meas). David mentions using averaged measurements for isolation, and in fact most professional VNAs use averaging when performing calibration.
Carlos
