Too Many Zeros

Main forum to discuss RTL-SDR related topics.
Post Reply
Terry Bullett
Posts: 2
Joined: Sun May 17, 2020 7:00 pm

Too Many Zeros

Post by Terry Bullett » Sun May 17, 2020 7:17 pm

All,

I am using Airspy and SDR# for radio astronomy, and the baseband recorder to save 10 MHz data for post-processing.
A consistent problem I am having is that when you perform an amplitude probability distribution function on and I or Q channel recording, there are always twice as many 0 values as there "should be". "Should Be" is defined by the proper Gaussian distribution of points for all the other amplitude bins except zero (0). As shown in this figure:
Raw-Linear.png
Raw-Linear.png (10.64 KiB) Viewed 1231 times
This shows up in baseband recorder and IF recorder too. But only when I record integer values. The anomaly is absent in 32 bit float data. Since my entire process is I/O bound, I would strongly prefer to use the 16 bit integers. for half the file size. My best guess is some kind of integer round off error for values 0+ and 0- Does anyone know about this problem and/or a fix?
Cheers,
Terry

Terry Bullett
Posts: 2
Joined: Sun May 17, 2020 7:00 pm

Re: Too Many Zeros

Post by Terry Bullett » Thu May 21, 2020 6:31 pm

Update: Under the same test conditions, I recorded 32 bit floating point data, multiplied the float values by 32768 and performed similar analysis and plotting. The 'Too many zeros" problem goes away completely.
Fit-Lin-zoom.png
Fit-Lin-zoom.png (11.47 KiB) Viewed 557 times
The "ears" at about +/- 27 amplitude on this distribution are probably real.

Also, the fuzziness of the previous plot turns out to have a period of 8 integer counts.
8CountPDF.png
8CountPDF.png (10.8 KiB) Viewed 557 times
There is clearly something systematically wrong with the 16 bit IQ data values.
The same problem exists with the IF recorder module.

Any clues anyone?

Terry

Post Reply