USRP Gain

GNU Radio
1

I have been attempting to calibrate my USRP system. I am running four
channels and feeding the various channels to
fft sinks following de-interleaving and channel filtering. I am using
the Basic RX boards, and the PGA in front of the ADC is set at 20 dB.
For large decimation, the sensitivity of the system seems to much
larger than I expected,
and is a function of the decimation factor D. Overall, I can’t
account for a gain of between 60 and 80 dB.

For decimations below 96, the gain is fairly constant, changing +/- a
couple of dB. However, for decimations above about
96, the voltage gain in dB is growing approximately linearly with D:
(for example, when I change D from 100 to 160 the overall gain
increases by about 5.5 dB. When I change D from 160 to 222, the
voltage gain increases by another 5.5 dB). So the gain itself is
growing exponentially with decimation rate.

I believe that the gain of a basic CIC filter is proportional to D
(but not exponentially). Is this the source of the gain variation
that I am seeing, or is some scaling or normalization going on
elsewhere that I am overlooking?

The documentation indicates that 4-stage CIC filters are used, and the
4-channel configuration eliminates the half-band filters.
Is there any information on how the decimation is distributed across
the 4 stages as a function of overall decimation?

Thanks for the help.

Dick…

2

On Mon, Sep 1, 2008 at 4:26 PM, Richard J.
[email protected] wrote:

of dB. However, for decimations above about
96, the voltage gain in dB is growing approximately linearly with D: (for
example, when I change D from 100 to 160 the overall gain increases by about
5.5 dB. When I change D from 160 to 222, the voltage gain increases by
another 5.5 dB). So the gain itself is growing exponentially with
decimation rate.

I’m not an expert by any means, but this CIC documentation is telling
me the gain is:

g = (RM)^N

Since M and N are defined (M=1, N=4) and you’re changing R between 160
and 220, finding the difference in gain in dB:

10*[ log10(220^4) - log10(160^4) ]
5.5 dB

Do you agree?

Thanks for the help.

Dick…

Brian

3

On Sep 1, 2008, at 9:55 PM, Brian P. wrote:

than I expected,
by about
Since M and N are defined (M=1, N=4) and you’re changing R between 160
and 220, finding the difference in gain in dB:

10*[ log10(220^4) - log10(160^4) ]
5.5 dB

Do you agree?

4-channel configuration eliminates the half-band filters.
Is there any information on how the decimation is distributed
across the 4
stages as a function of overall decimation?

Thanks for the help.

Dick…

Brian

Your analysis looks good. I checked the rest of the gain data I had
taken for R = 96 through 256,
and g = K (R^4) seems to provide a good fit to the data.

Thanks again.

Dick