CODEWITHSUNDEEP
integerrandomsequenceprng
Sam Washburn
Sam Washburn

Reputation: 1827

Pseudorandom Sequence Generator not just a number generator

I need an algorithm that pretty much will turn a unix timestamp into a suitably random number, so that if I "play back" the timestamps I get the same random numbers.

And here's what I mean by suitably:

  1. Most humans will not detect a loop or pattern in the random numbers.
  2. It need not be cryptographically secure.
  3. All numbers must be capable of being generated. (I've found that LFSR don't do this)
  4. The numbers are 32 bit integers

And I would like it to be fairly fast.

So far my idea is to just seed a PRNG over and over, but I'm not sure if that's the best way to handle this.

Any thoughts and ideas will be much appreciated.

Thanks.

Upvotes: 1

Views: 719

Answers (4)

ephemient
ephemient

Reputation: 204698

I would suggest that the easiest thing to do is feed your time to jrand48. Something like

#include <stdlib.h>
int mix(int t) {
    unsigned short x[3] = {t, t<<16, t};
    return jrand48(x);
}

It's reversible (216·x+n≡0x5deece66d·(232+1)·t+0xb mod 248 ⇒ t≡0xdfe05bcb1365·(216·x+n-0xb) mod 248 where n∈[0,216)) but since it's the high 32 bits out of 48-bit, it's actually not too easy. (You can apply jrand48 to x more than once too; as long as you don't apply it 248-1 times, the same sorts of properties will hold.)

Upvotes: 1

L̲̳o̲̳̳n̲̳̳g̲̳̳p̲̳o̲̳̳k̲̳̳e̲̳̳
L̲̳o̲̳̳n̲̳̳g̲̳̳p̲̳o̲̳̳k̲̳̳e̲̳̳

Reputation: 12560

(timestamp ^ 0x12345678) + 12345678 Is this subtle enough?

If you don't care about reversibility of it you could crc32 each timestamp.

Upvotes: 0

Jonathan Leffler
Jonathan Leffler

Reputation: 753675

I suggest looking at the POSIX-compliant drand48() family of functions. They give decent (but certainly not cryptographic) random numbers, and srand48() takes a 32-bit seed value. They are deterministic, so reusing a given seed will regenerate the same sequence of numbers again.

Upvotes: 0

outis
outis

Reputation: 77400

If it doesn't need to be statistically random, perhaps feed the timestamps to MD5 and truncate the hash. The main issue is that I don't know if this would be surjective. Other hashing algorithms might work better.

Upvotes: 2

Related Questions