[Discuss-gnuradio] sampling artefact because of two step mixing?

[Luis Simoes]

Hello everybody,

I am currently using the flex2400 board and I feed it with some designed 
signals from a sophisticated signal generator. I analyse the spectrum of 
interest by fft and pass all information to a file sink. I plot the file data 
in Matlab to evaluate the results.
As I feed the USRP with a single sine tone with a frequency of 2.444 GHz and 
an amplitude of -50dbm I saw on my plot a nice peak at 2.444 GHZ but also a 
second peak at 2.452 GHZ but attenuated by 15 db's when daughter board is 
tuned to 2.452 Ghz. To verify the accuracy of the signal generator I 
connected it to a high quality spectrum analyser. The spectrum analyser 
verified that the output of the signal generator is a clean peak without any 
side peaks. However, the usrp_fft.py tool from the gnuradio examples shows 
the same phenomenon including the second peak.
The parameters I use in my application are : 
Flex2400 daughterboard
Decimation factor 8
complex samples at 16bit I and 16 bits Q each
fft size 64 ( corresponds to 125 kHz bin resolution)

My first idea points to the effect of the second mixing in the DDC from the 
remaining frequency offset after the analog mixing in the daughterboard tuned 
the centre frequency as close as possible to baseband.
When the tune method is set to the centre frequency of 2.452 Ghz, the flex 
mixes with 2.448 GHz and the DDC with -4MHz. By mixing with a cos wave we get 
two peaks, one at (f-f0) and one at (f+f0), but both with half signal 
strength. The resulting peak from mixing with the double frequency (f+f0) can 
now explain the appearance of this side peak in my plot.
But: 
1. Why is the second peak attenuated? If it is a result of mixing it should be 
as high as the original signal?
2. If the assumption of the two peaks is correct, why are the assumed and the 
real measured peaks mirrored in other configurations (other signal frequency 
and center frequency of the usrp)?
3. The flex2400 is able to tune to every frequency between 2400 and 2500 MHz 
in steps of 1 MHZ. Why can I not tune the flex directly to the centre 
frequency without another mixing stage in the DDC? The DDC frequency is 
allways between -2 and -5.5MHz. Would this effect disappear if no second 
stage mixing is needed?

I found almost no documentation about the configuration of the DDC. Which 
filters are implemented and what are the parameters used in the logical steps 
of mixing, decimating and low pass filtering?

Is there any way to avoid this physically not existing signals and if not is 
there a detailed explanation why this phenomenon occurs in an irrational (it 
seams so) way?

I am very grateful for any advice,
Perhaps Matt and Eric are the experts in this matter. So this question is 
specially directed to you.
Thanks a lot

Luis