[Cryptography] Whitening Algorithm

[Rob Seward]

  Hi,
I’m trying to whiten a random noise source (a reverse biased transistor)
with a low-powered microprocessor. I figured out a technique that seems to
work well, and I want to know if there is anything insecure or subpar about
it.

(Earlier, I had heard that XORing a random stream with alternating 0s and
1s could remove bias. However, this strikes me as very insecure, because an
attacker could reverse the mask by XORing the mixed stream with the same 01
mask. Also, I had used the Von Neumann algorithm
<https://en.wikipedia.org/wiki/Hardware_random_number_generator#Software_whitening>,
but this slows output significantly.)

The algorithm mixes new bytes derived from the noise source with previous
ones in an overlapping manner. Below is the source code, and a link to an
illustration of the process.

I’ve done some testing, and it appears to transform data that shows as much
as 5% bias to ~0% while passing ent chi-square. Some less rigorous testing
with NIST also had positive results.

Here’s an illustration of what the algorithm is doing:
http://imgur.com/itLWNyf

void loop(){
  currentByte = readByteFromSource()

    mixedByte1 = currentByte ^ previousByte;
  mixedByte2 = mixedByte1 ^ previousMixedByte1;
  mixedByte3 = mixedByte2 ^ previousMixedByte2;
  mixedByte4 = mixedByte3 ^ previousMixedByte3;
  outByte = mixedByte4 ^ previousMixedByte4;

  Serial.write(outByte);

  previousByte = currentByte;
  previousMixedByte1 = mixedByte1;
  previousMixedByte2 = mixedByte2;
  previousMixedByte3 = mixedByte3;
  previousMixedByte4 = mixedByte4;
}
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.metzdowd.com/pipermail/cryptography/attachments/20150722/02251b8a/attachment.html>