Hi Danny and Dan, Thanks for your advices! The main cause was indeed the quantization and the FFT bitwidth. I did increase FFT bitwidth to 25 and increase the quant block output to ufix_64_25 To do so I had to also increase the vacc bitwidth and store the output in 64bit (two 32bit bram) After all this bit growing I got a nice flat noise floor 75dB below the test tone I put (which was at the ADC full scale = 0dbm). the response is totally linear all the way! The optimal shift seems to still be FFF.
Now the problem will be to put two of this spectrometers in a ROACH and compile for high speed (1.0 - 1.5GHz) ! I am getting into this now ! Cheers, r On Thu, Jul 26, 2012 at 9:40 PM, Danny Price <[email protected]> wrote: > Hi Ricardo > > I think along with the FFT shift, the quant block is a likely culprit. I'd > recommend increasing the quant block from 6 bits to somewhere above 8 bits > and seeing if the problem persists. (You'll have to modify the vacc block > too). > > Also, you might want to consider digitizers with more bits if dynamic range > is critical. The iADC and 083000 have 8-bits (255 levels), so the dynamic > range is 20*log10(1/255) = 48dBm, a tad lower than the 50dB figure you > mentioned. > > Regards > Danny > > On 26/07/2012 21:01, Ricardo Finger wrote: >> >> Hello Casper community, >> >> I am working with the tutorial 3 spectrometer (slightly modified to work >> with the iADC083000 board). >> I need to analyze strong CW tones with a dynamic range of at least 50dB. I >> am using the standard libraries and running at 500MHz ADC clock. >> I started reducing the digital gain of the 'quant' block, and adding noise >> to the tone to reduce the digital artifacts/harmonics. >> So far so good, I got a strong line 75dB above the noise floor. The >> problem is that those 75dBs are not the effective dynamic range because >> there is a strange nonlinearity in the response of the spectrometer at the >> lower power end. Please see the attachment "200MHz tone with 0 to 40dB >> attenuation.pdf" >> >> When I reduce the tone power in 10dB steps I can see the peak going down >> accordingly for the first two steps (20dB) but with the next 10dB of >> attenuation, it drops more than 20dB, and with the following 10dB of >> attenuation (40dB in total) it does completely disappear, dropping more than >> 35 additional dBs. >> Have any of you seen this 'low end' nonlinearity before? >> Is it part of the ADC response or an artifact of the digital processing? >> is it maybe related to the re-quantization before accumulation? >> >> The nonlinear response also applies to the noise floor as it can bee seen >> in the attachment "200MHz tone with 5 steps digital gain.pdf", where for a >> gain of '0F 00 00 00' the noise floor has a slope of around 10dB, but for a >> gain of ' 00 80 00 00" the slope is more than 50dB. >> >> Regards, >> >> Ricardo. >> > > -- Ricardo Finger Camus Electrical Engineer Astronomy Department University of Chile Of: 56(2)9771122 Casilla 36-D, Santiago. http://www.das.uchile.cl/lab_mwl/
