12-04-2020 01:22 PM - edited 12-04-2020 01:24 PM
I have been running some single tone tests to evaluate the performance of the ZCU111 ADC. One strange behavior that I have noticed is that the noise floor rises for some particular input frequencies. For example, my noise floor rises 20dB higher if I move my tone frequency from 1023.5MHz to 1023.7MHz with a sampling clock of 4096MHz. The higher noise floor goes away if I change the input tone frequency slightly or if I change the sampling clock frequency slightly. This strange behavior happens on many others input frequencies and sampling clock rate pairs too. I wonder if anyone has experience the same behavior and if this is a behavior documented by Xilinx.
12-09-2020 03:12 PM
I haven't come across this before, I was thinking something weird like a weird combination of the input signal and the broadband jitter on the sample clock but 20dB is too much here. What is the size of the FFT here? what does this look like when you try it in the Eval GUI or RF analyzer? the input is very close to FS/4 so perhaps this a calibration thing and the FFT size is too small to have the interleaving spurs above the noise floor of the FFT
Please give it a try in the zcu111 rftool evaluation design
12-09-2020 07:45 PM
These are time series data that I got from the output of the RFDC using ila core. The capture size is 8k here. The spectrum I showed is done by MATLAB sinad function. I have tried plotting FFT with various large FFT size but still shows the 20dB noise floor difference. It could be related to FS/4. If this is a calibration thing, how do I fix it here?
12-10-2020 02:41 AM
And is this repeated in the RF Analyzer /ZCU111 Eval tool?
12-13-2020 10:50 PM
The ZCU111 Eval tool also shows the same behavior but with a less noise floor difference which is about 5dB. The maximum FFT size from the UI is 32k so I don't know if this is limiting the noise floor.
12-14-2020 04:06 AM
You could switch the Eval design to use DDR instead of BRAM and try a bigger FFT?
12-14-2020 03:53 PM
this is not expected for me.
Is your spectrum good in terms of interleaving spurs in the working case? does it get worse when the noise floor rises?
can you try set the signal to 1023.5MHz, get a good capture, then freeze the calibration and move the signal to 1023.7Mhz. Does the issue still occur? If it does we can rule out the calibration.
another thing to try out is the dither .Try it with dither off and on to see if this is the issue
12-14-2020 04:00 PM
You mentioned that you tried a longer FFT but not sure if you meant longer FFT for the same set of data or you captured a longer data and tried a longer FFT.
In GUI, You may switch the memory to DDR and capture around 10^20 samples. Set FFT length to the same data length and see if that reduces the noise floor.