Re: white noise generation

On Jul 4, 9:48=A0am, Jerry Avins <j...@[EMAIL PROTECTED]
> wrote:
> esfield wrote:
> > The xor****ft generator I'm using has three xor operations and three
****=
ft
> > operations; which, if I understand correctly, are supposed to work
toge=
ther
> > to create a 32-bit random number. =A0
>
> > Here is the code (I'm doing an asm version of this):
> > unsigned long xor(){
> > static unsigned long y=3D2463534242; //seed
> > y^=3D(y<<13); y^=3D(y>>17); return (y^=3D(y<<5)); }

wow, i nearly can understand this, but not quite.  the ****ft-register
sequence operation i was thinking of is sorta described at

   http://www.dspguru.com/info/tutor/mls.htm

now maybe your operation above can be broken down into a normal
looking LFSR sequence, the operation would be (in C code)


static primitive_poly =3D 0x82D4E1B8;
static int state =3D 1;   // can be anything non-zero

int random_bit(void)
   {
   if (state&0x00000001)
      {
      state =3D (state>>1)^primitive_poly;
      }
    else
      {
      state =3D state>>1;
      }

   return state&0x00000001;
   }

an operation similar to that, is what i understand to be a
pseudorandom ****ft register sequence.  the LSB of "state" are bits
that are virtually white in their correlation (except for DC), but "
"state" itself, taken as a binary PCM value, does not make for a good,
white random number.


> > When you refer to the xor****ft method only producing one random bit at
=
a
> > time,
=2E..
> I didn't work through your code. There may be something clever that
> randomizes all the bits in the register, but I doubt it.

yeah, i'm skeptical too.  but i also hadn't worked through it.  maybe
these 3 ****ft and XOR operations can be shown to be equivalent to the
LFSR code above for a primitive polynomial that is somehow implied.
i'll have to look more closely after the holiday.

r b-j