At 10:40 AM 7/28/98 +0100, Mok-Kong Shen wrote:
David Honig wrote:
The output of a good block cipher in feedback mode will pass Diehard tests, though it is not crypto-secure.
I often see the phrase 'pass Diehard test' though I don't see from the documents of Diehard how to evaluate the volumenous printout of Diehard to say exactly whether the test is passed or not. Furthermore the component asc2bin.exe of Diehard is buggy.
M. K. Shen
My rough understanding: the 'P' value is a measure on the hypothesis that the test sample is a truly random sample, where truly random is defined by the expected statistical properties being measured. Eg in 100 bits you expect to find 50 1's; if you count 48, is your 100-bit sample consistant with it being unpredictable? If you get values near 1.0 your sample is not likely taken from a random pool. Try this: generate 10Meg from a block cipher feeding back on itself. Diehard will pass these. (Diehard needs 10M samples) Now run FM hiss into your soundcard. Sample this at 8Khz (to avoid temporal correlation) and save to a file til you have 10Meg. Diehard will reject this. Make a larger file, and then gzip it down to 10Meg. (That it shrinks indicates its symbols don't carry a full bit.) Run Diehard on this. It will pass more tests but not all. Take the FM hiss, feed it into a stream cipher, and start burning those OTPs. Do this with a detuned *video* tuner for more bandwidth. honig@alum.mit.edu "Speech is not protected simply because it is written in a language" Federal Misjudge Gwin on the Bernstein Case