USRP decimation filter performance

GNU Radio
1

Hi all

Is there information anywhere about the performance on the USRP
decimation filter?

I have a setup with an external radio which has a bandwidth of 5MHz
around a 10.7MHz IF. I think the attenuation of the radio is quite poor
relatively far away from the 10.7MHz centre (e.g. -20dB at +/-10MHz).

If I use the USRP with a decimation of 8, I should get a bandwidth of
8MHz (minus the filter roll-off) I assume - is this correct?

If so, then I’m worried about images in my passband - i.e a signal at,
say, 18.7MHz will wrap around and be right in the centre of my passband,
but attenuated by 20dB by the radio itself. I assume there is
attenuation from the decimation process (the CIC/half band filter) - can
anyone give me an idea of how much decimation I can expect outside of my
passband? (or put me right if I’m completely wrong with all of this!)

Thanks for your help

Rich

2

Richard M. wrote:

If I use the USRP with a decimation of 8, I should get a bandwidth of
8MHz (minus the filter roll-off) I assume – is this correct?

Yes. You get about 80% usable bandwidth due to the filter rolloff.

If so, then I’m worried about images in my passband – i.e a signal at,
say, 18.7MHz will wrap around and be right in the centre of my
passband, but attenuated by 20dB by the radio itself. I assume there
is attenuation from the decimation process (the CIC/half band filter)
– can anyone give me an idea of how much decimation I can expect
outside of my passband? (or put me right if I’m completely wrong with
all of this!)

About you get >80 dB of attenuation outside of the passband.

If you set the center frequency to 10.7 MHz and the decimation to 8,
then you are correct that 18.7 MHz would be right in the center of your
passband. However, the attenuation is well over 120dB in this case.

To clarify, you do NOT need to worry about aliasing into your signal.
There is no measurable aliasing as long as you are in the center 80% of
the bandwidth.

You need to determine how wide your signal is. Then set the decimation
so that it fits your signal. Anything outside that range will be gone.
Nothing else to worry about.

Matt