Dear USRP and GnuRadio folks,
I'm not sure whom to whom I should address this question, since I don't know if the problem arises in the GnuRadio code, the drivers, or the actual physical USRP.
I was attempting to run an extremely simple GRC graph (attached to this email) where the input antenna of the USRP is grounded, the signal is taken in with UHD: USRP Source, then the samples' magnitudes are squared, then fed through a moving average (100,000 points), and then the result is displayed on the screen in dB.
As I altered the sampling rate, I observed a very strange thing: the average for the noise readings were, by and large, smoothly increasing with the sampling rate, except at certain frequencies (specifically, 30*2^k, k=0,1,2,3, etc). At these specific sampling rates, the noise level would shoot up by about 20 dB.
I'm attaching a graph of the some of the values I got, where the big jumps are marked by their respective sampling rates.
Can someone explain why these jumps occur? I'm guessing that when I adjust the samp_rate, it selects the front-end filter from a filter bank and combines that with decimation. Maybe at certain samp_rates (30*2^k, k=0,1,2,3,...) it's right at a threshold edge where the noise is worse. Or maybe the problem is entirely at the other end and GnuRadio has a bug in its assignment of samp_rate values?
Two attachments included. Comments much appreciated.
Thanks and regards,
Ale