5

I am trying to build a noise source based on reverse biased PN junction (i am trying to replicate this project https://altusmetrum.org/ChaosKey/ ). At this point i tried multiple iterations but so far this is my best result achieved. The noise source is sampled by STM32 microcontroller and sent over USB.

Here is the schematics of the noise source itself (MCU and power supplies omited for clarity)

VAA is 23.1V

noise source schematics

Here is frequency spectrum of the signal (software dc blocked):

frequency spectrum

This looks flat and good

Here is histogram of the same signal (amplifier is inverting, so the original signal will have the same histogram, but mirrored):

histogram

This looks weird. I would expect normal (Gaussian) distribution, but i am getting this weird shape. Why could this happen?

For reference, here is the signal as seen by oscilloscope (ac coupled) after C12:

scope measurement

I tried to vary the VAA - when i lower the voltage, the histogram changes and at around 10V, it looks more or less correct, but i would like to understand why is this actually happening? What is the source of this weird distribution?

pigster
  • 51
  • 2
  • 1
    If you want gaussian white noise, use thermal resistor noise. Here, I wouldn't expect a normal distribution: The breakdown happens "in bunches" meaning the voltage at the bases would be some kind of sawtooth or telegraph waveform perhaps, which is then passed through an exponential transfer function by the gain stage. – tobalt Jun 27 '23 at 19:33
  • 1
    So after lot of digging and experiments, here is my findings: TLDR; R4 value is way too small - the current through the reverse biased junction is way too large. Good value found for this particular case is 470k. See http://holdenc.altervista.org/avalanche/index.html @tobalt actually it will be lognormal, but very close to normal anyway – pigster Jul 27 '23 at 20:05

0 Answers0