# How to analyse the distribution of the noise being received by my RTL-SDR device?

I'm very new to RTL-SDR and signal processing in general, but have managed to get my device (NOOELEC NESDR Mini) working and data into Python (via the pyRTLSDR library).

I am able to output a complex-valued array of predefined length. I understand these are the I/Q values. The distribution of these values is what I'm interested in for now (rather than applying FFT and whatever comes after).

In a given sample, I found that the distribution of both the real and the imaginary parts of the array formed a truncated normal:

I had some questions regarding this:

• How do I interpret the noise that is on top of these I/Q values?
• Why are the values truncated between -1 and 1? Is the limitation of being inside the unit circle just a hardware thing, or a theoretical thing?
• Can I extract just the noise from these values to get an array of true random numbers?

Further, can someone point me towards some literature so I better understand this initial step of the process?

Note that the SDR settings were left as default except the centre frequency, which is tuned to a local FM station:

# configure device
sdr.sample_rate = 2.048e6  # Hz
sdr.center_freq = 107.6e6  # Hz
sdr.freq_correction = 60   # PPM
sdr.gain = 'auto'

• An RTL-SDR outputs 8-bit integer IQ values (0..255, or -128..127). How are you (or the python library) scaling these integer IQ pairs to get a range of -1.0 to 1.0 ? Jul 21, 2021 at 21:10
• That graph looks like a normal distribution of noise, except for the two end-most bins, which are higher than their adjacent neighbours. That looks like signal clipping, where the SDR's ADC limits are exceeded, The usual procedure for "out-of-bounds" ADC samples is clipping. Could be the result of auto gain setting? Jul 21, 2021 at 21:12
• The auto gain setting on an R2832U RTL-SDR is often too high a gain (as it was likely designed for a different type and bandwidth of DTV signal). Try using manual gain, and reduce it so as not to see any (or too much) clipping. Jul 21, 2021 at 21:24
• Hi Kris, and welcome to ham.stackexchange.com! Jul 22, 2021 at 1:29

• Thanks. It does seem as though the pyRTLSDR library is normalising the samples. Suppose I tuned into a local radio station playing a song; how would I go about doing what you mentioned at the end? (i.e "filtering out your assumed-non-noise signals in the frequency domain")