anrand: Analysis tool for a Hardware Random Number Generator
Copyright (c) 2016 Bart Massey
This tool is currently customized for analyzing the raw random bits produced by our AltusMetrum ChaosKey hardware RNG (formerly known as USBTRNG). The bits come out of the ChaosKey as 16-bit little-endian samples with 12 bits of possible range.
This code directly reads a file of these raw samples, and computes a number of graphs and metrics that are intended to verify the operation of the device. (In actual production use, the samples are whitened to 8 bits using a CRC, yielding -- as far as we can currently tell -- perfect hardware entropy.)
Build and Run
This is a pile of Haskell code. You will need to install GHC
and cabal-install on your platform, and figure out how to
operate it. The anrand.cabal file has a list of packages
you must install from Hackage. The code is intended to be
platform-independent, but I've only ever tried it on a
Debian Linux box.
Run the program with your unwhitened random bits on standard
input. The program will create an analysis subdirectory
with a bunch of text and PDF files in it.
The file prefixes are:
raw: The unmodified samples.low: Just the bottom eight bits of each sample.mid: Bits 8..1 of the sample, because the ChaosKey has a problem with the bottom bit or two.prng: All the statistics are also run against a PRNG so that you can see what white bits should look like.
The file suffixes are:
stats: Some summary statistics about the sample.hist: Sample histograms.ts: The initial time-series data.dft: Magnitude of DFT of a subsample of the data.
Adapt
You can pretty easily hack up the input reader to support other random sources of bits. You can also pretty easily add analysis to taste. See the Haskell source for details.
License
This work is made available under the "MIT License". Please see the file LICENSE in this distribution for license details.