Re: [Discuss-gnuradio] Spurious Signals and DC Offset

[Lee Patton]

The spurious harmonics I was seeing in the DBSRX output go away after
the board warms up for about 0.5 and 1.0 hr. (See attached plots.)  I
also see significantly less drift at 1800 MHz after this warm-up time.
I don't know however, if the drift I see is due mostly to the DBSRX or
the ancient signal generator I have.

I still don't understand the DC offset I'm seeing in the data. It seems
more pronounced that expected. I think that has to do with some flaw in
my data capture.


On Fri, 2005-12-09 at 16:59 -0500, Lee Patton wrote:
> Thanks, Matt. I appreciate the help.
> 
> On Thu, 2005-12-08 at 23:25 -0800, Matt Ettus wrote:
> > >   1) noise_ave_pwr.jpg
> > >     * normalized average power over time
> > >     * is there any reason why the power would start high then go
> low?
> > 
> > The first samples could be corrupted because the pipeline may not be
> > cleared.  Drop the first samples
> 
> Is there a programmatic way of clearing the pipeline before collecting
> data, or at least know how much data I have to throw away for each
> collection? It looks like the first 0.02 seconds worth.
> 
> > >   2) noise_ave_psd.jpg
> > >     * average PSD of a chunk of the data
> > >     * why the very large energy at DC?
> > 
> > It's less than 10 dB above the noise floor.  Turn up the gain and you
> > won't see it.
> 
> I see this no matter what gain I use. That goes for the DBSRX and
> BasicRx boards. Please see attached plots.
> 
> > 
> > >     * why the spikes at +/- 1 MHz and -2 MHz?
> > 
> > You are seeing synthesizer spurs or harmonics of the reference.
> You'll
> > notice that they are only about 3 dB above the noise floor.
> 
> I figured it was something like that.  The weird thing is that they're
> often much larger than 3dB above the noise floor.  I even got rid of
> them for a very short period of time when I was experimenting with
> enabling/disabling the halfband filter (although I'm not sure that had
> anything to do with it.)
> 
> 
> > 
> > >   1) wifi_psd.jpg
> > >     * fft of 0.5 sec. of data
> > >     * very pronounced DC component
> > 
> > How many bins in the data?
> 
> This was a full half second at fs = 4MHz. I think matlab zero pads to
> reach the next largest power of two for fft. So, that would be 
> Nfft = 2^( ceil(log2( 0.5*4e6))) = 2097152, right?  No windowing was
> done either. That should be about 2 Hz resolution.
> 
> > 
> > >    
> > >   2) wifi_psd_ave.jpg
> > >     * average PSD of entire data collection
> > >     * DC component not as pronounced
> > >     * I guess the half-band filter in the FPGA is causing
> attenuation at
> > > the edges of the band?
> > > 
> > > I have attached the script I am using to collect the data. It is
> just a
> > > customized version of usrp_rx_file.py.  I seem to see these affects
> > > regardless of frequency, pga gain, or whether I write complex or
> short
> > > samples to file.  On a related note, besides space considerations,
> are
> > > complex or short samples preferable one to another?
> > 
> > What command line options are you using for the script?
> 
> ./capture -f 2.412G -g50
> 
> I've attached two more plots. One is the BasicRx with no antenna, and
> the other is DBSRX with no antenna.  Two log files are attached as well
> so that you can see the settings.  These were taken at 1800 MHz.
> 
> Note:  the PSDs shown here are calculated as follows:
> 1) 0.5 second data stored in x
> 2) x is reshaped into length(x)/1024 rows by 1024 columns
> 3) fft taken across columns of x
> 4) average magnitude squared of rows is taken and plotted.
> 
> I guess I just don't know if what I am seeing is expected or not.  If it
> is, that's fine.  I just want to make sure there is nothing wrong with
> how I'm collecting the data.
> 
> Thanks,
>   - Lee
> _______________________________________________
> Discuss-gnuradio mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/discuss-gnuradio

dbsrx_ave_psd.jpg
Description: JPEG image

dbsrx_nospurs_2.412G.jpg
Description: JPEG image