Wanted to get an idea how the V2+4 filter compares with the Lites. Using the same narrow band setup on both VNAs. Changing the Lite's averages from 1,2 & 3.
At least with the
original NanoVNA V2 firmware, the exact shape near the center is not so important, since the IF freqency resulting from mixing stimulus with LO always hits the peak at the center of the filter exactly (+/- phase noise) once the PLLs are locked, due to the chosen frequency plan
1). An important property of the filter
2) are however the zeros at f
center +/- N * f
IF, where N is an integer >= 1, in order to cancel IF harmonics completely, which emerge as mixing products of the the (square wave) harmonics of stimulus and LO. These zeros are therefore important in order to deal with the square wave drive.
1) Stimulus, LO, and ADC clock are derived from the same oscillator and are therefore frequency locked to each other. Stimulus and LO frequency are always integral multiples of the minimum frequency step, IF frequency is an integral multiple of the minimum frequency step, too, and ADC clock is an integral multiple of the IF frequency. Since the ADC clock is frequency locked, too, and an integral multiple of the IF frequency, the IF frequency can't miss the center of the digital filter once the PLLs are locked.
2) The IF filter is not a filter in the traditional sense, but the frequency response is eventually an implied property of the (digital) synchronous detector (mathematically the detector is equivalent to a windowed DFT, but calculated only for a single frequency bin). In addition to the synchronous detector, there is just an analog lowpass pre-filter between mixer and ADC, obviously acting as anti-aliasing filter.
Vector averaging of multiple acquisitions inevitably changes the effective frequency response of the detector, since the effective window (over which the DFT is accumulated) becomes a repetition of the primary window function then (possibly even with zero-valued gaps between the repetitions if there happen to be time gaps between the acquisitions which are averaged). Your screenshot did nicely visualize the change of the frequency response

. The averaging strategy of the Nano and the Lite obviously deviate from each other. The peak at the center and the zeros at f
center +/- N * f
IF (which are the properties of importance) are not affected by averaging and still retained, though.