Exact match. Not showing close matches.
PICList
Thread
'[OT] "Real" Random sequence sought'
1999\10\15@163845
by
Ben Stragnell

> Once recorded, does a random sequence cease to be random as can now be
> repeated at will, and is surely now  by definition  deterministic? Can
> randomness be quantified? Any mathematicians out there care to enlighten
> me?
For a sequence to be random, all that's really required is that each entry
in the sequence is completely independant of the preceeding entries. Now, if
I take a finitelength sequence of random numbers, and then play this
sequence endtoend repeatedly, then the longer sequence that I end up with
no longer obeys this requirement.
The easiest way to consider this is from the point of view of an "observer"
(eg. a PIC), who runs through the list one entry at a time. As long as they
run through the list only once, then the observer is unable to predict any
future entries, because they're independant of the preceeding ones. If the
list has been precomputed, then from *our* point of view, yes, it's
deterministic, but only because we have knowledge of the whole list.
Cheers,
Ben
1999\10\16@003123
by
Keith Causey
Another property of random number sequences is that there are no long
autocorrelations. A 64k snippet, or any other snippet for that matter will
eventually have a 100% overall autocorrelation. You may rethink your reasons
for needing random numbers and decide that pseudorandom numbers will do.
They are easy to generate algorithmically as you already have discovered.
The most economical solution that I have seen so far is the diode scheme.
Once that is generated then sampling it with an analog to digital converter
to the desired resolution is all that is needed. This should result in a
totally noncorrelatable sequence. Of course the catch22 is that to
determine if a sequence has 0 autocerrelation you have to look at it
forever.
1999\10\16@054758
by
Robert A. LaBudde

At 09:30 PM 10/15/99 0700, Keith Causey wrote:
>Another property of random number sequences is that there are no long
>autocorrelations. A 64k snippet, or any other snippet for that matter will
>eventually have a 100% overall autocorrelation. You may rethink your reasons
>for needing random numbers and decide that pseudorandom numbers will do.
>They are easy to generate algorithmically as you already have discovered.
>The most economical solution that I have seen so far is the diode scheme.
>Once that is generated then sampling it with an analog to digital converter
>to the desired resolution is all that is needed. This should result in a
>totally noncorrelatable sequence. Of course the catch22 is that to
>determine if a sequence has 0 autocerrelation you have to look at it
>forever.
We are apparently discussing two different types of randomness: 1) random
'numbers', which have some finite bit length, and 2) random bit streams,
which are sampled at some underlying rate.
The Platonic ideal of 'randomness' is useful in defining tests for
practical sequences. The fact that infinite bandwidth is impossible is
irrelevant in practical problems, since bandwidth is always limited
intentionally.
Random sequences are infinitebandwidth and 'white' in the same way ideal
opamps are infinitebandwidth and gain. Realworld implementations, of
course, only approach the ideal.
In the case of random sequences, the upper limit on bandwidth required is
1/2 the sampling rate. The 'white' bandwidth required is usually much less
than this. I.e., one is much more concerned with proximate correlation than
distal correlation.
'Whiteness' implies two principal properties:
1. 'Equidistribution', i.e., any subset is equivalent to any other subset.
2. 'Uniformity', i.e., correct DC component.
Since random numbers/digits are used in ntuples (singles, pairs, triples,
etc.) with 'n' a small number (usually < 10), then having zero loworder (<
10) autocorrelations usually suffices.
================================================================
Robert A. LaBudde, PhD, PAS, Dpl. ACAFS email: spam_OUTralTakeThisOuTlcfltd.com
Least Cost Formulations, Ltd. URL: http://lcfltd.com/
824 Timberlake Drive Tel: 7574670954
Virginia Beach, VA 234643239 Fax: 7574672947
"Vere scire est per causae scire"
================================================================
More... (looser matching)
 Last day of these posts
 In 1999
, 2000 only
 Today
 New search...