Hello ADI team,
On the AD9361, we are observing some cases where the Rx path passband noisefloor appears to not be flat when the Rx gain is below a certain threshold, and gradually becomes less flat as Rx gain is decreased. We know the FIR filter + HB filters response can cause some minor non-flat variance here in the passband, and we can correct for this with an appropriately built FIR filter since it is predictable (using the Matlab FIR filter tool or similar). However, for higher sample rates (> 35 Msamples/sec) where the channel bandwidth is a substantial fraction of the sample rate (roughly > 70%), we are observing that the edges of the passband before the normal FIR filter pedestals seems to increase in amplitude when the Rx gain is under ~40 or so. This looks very similar to the plots shown in this post:
From that previous post, it looks like the ADI team could confirm that the expected passband was non-flat (varied by almost 20 dB in the case above) when the Rx gain was set to 0. Obviously an Rx gain of 0 is exceptionally low, but for lower gain values (under ~40), we're also seeing a similar phenomenon. We wanted to try to get clarification regarding whether this is expected or not, what the cause may be, and if there is any way to improve this for low gain levels. It seems like needing to compose a table of FIR filter coeffs vs. Rx gain to potentially account for the amplitude variance would be one way, though this gets to be a very large data set very quickly.
For reference, one of our engineers plotted what we're observing in the passband for a similar situation, vs Rx gain values. This is for a sample rate of 36 Msamples/sec, where the FIR filter's decimation is set to 1 to accommodate the sample rate (with all of the preceding HB filters contributing at divide-by-2):
As can be seen, for Rx gain values > 40, the passband is relatively flat. However below that, the increase in amplitude at the edges is very similar to what was shown in the other forum post above. Further, if we decrease the sample rate to 35 Msamples/sec or below (i.e., the FIR filter's decimation is set to 2), we don't observe any increase in amplitude towards the edge of the passband.
Any insight as to the cause (and potential solution for flattening this out) would be appreciated.