Substantive noise difference between runs

Hi there,

I’d like to preface this by saying thanks for replying to the previous
issue I had about the internal filters of the USRP. I have managed to
equalise these out okay. However over the 10MHz wired interface (2m or
so coaxial cable), I am sometimes getting runs (0.5 seconds) which are
quite nice and smooth and others which are substantially more noisy. I
have changed daughterboards, USRPs, SMA-BNC adapters, USB cables,
interpolation and decimation rates. Gain figures have also been kept
relatively low to avoid distortion. Currently I believe the issue is
caused by corruption of data at either the interpolation end or the
decimation end. I make sure now that there are no overruns or underruns
(uU or uO) during the runs, because I think this may cause corruption of
the flat noise spectrum I am sending.

A figure is attached, I have interpolated a pseudo-random noise signal
to 20% of its original bandwidth in Matlab and sent this to the USRP
with a interpolation rate of 64. Thus the spectrum is 10MHz±1MHz, with
my noise signal taking up 10MHz±200KHz. In this figure I have not
equalised the effect of the CIC filters. As you can see, one run is
substantially more noisy than the next. Nothing much has changed between
these runs, perhaps only the receiver programmable gain. Sometimes the
noise is at quite a low figure for many runs and sometimes its quite
high for many runs. The level of noise seems to be quite stable for a
period then changes dramatically. Any ideas?

Thanks in advance,