True Random Number

T Chap

Anyone have an idea on generating a True random number?

?X is not it

Incorporating the CNT into a formula may work, but even then it is derived. I am looking for the closest thing to the brain picking a number between -2B and 2B, preferably not time based.

I suppose a computer can do nothing without some formula even in generating randomness.

T Chap

This is the closest I can find.

   Num := CNT
   Num := ?Num

Paul Baker

The only means for generating a true random number is by measuring a random system, it is simple not possible to compute a random number, hence all computationally generated numbers are called psuedo-random.

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Paul Baker
Propeller Applications Engineer

Parallax, Inc.

CJ

why not just let a couple of pins float and add them to an accumulator for a little while, to create your seed

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Parallax Forums - If you're ready to learn, we're ready to help.

T Chap

CJ, that is the idea, something by which some external, unknown source can affect it.

Paul Baker

Unfortunately a pin floating isn't a truely random process (its a biased system) and would require a mathmatical transform to unbias it. If you are looking for a random seed generator, the easiest is a pushbutton recording the cnt. If you want a hands off method there is a means of taking a noisy process (such as a zener diode) and amplifying it into a stream of binary noise. But since it seems what you are really asking for a relatively random seed and not a true random number generator, a circuit such as http://www.discovercircuits.com/DJ-Circuits/noisegen.htm·would likely work just fine.

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Paul Baker
Propeller Applications Engineer

Parallax, Inc.

Post Edited (Paul Baker (Parallax)) : 3/17/2007 3:23:05 AM GMT

cgracey

Hah! How about this:

The internal ~20KHz oscillator is a free-running analog oscillator and is used to sync between clock selections when CLKSEL is written to (via CLKSET(mode, frequency). It takes two of these clock cycles to get through a CLKSET instruction in Spin, but how long this actually takes varies, because of the phase relationship to the ~20KHz clock (~50us to ~100us of delay will occur). So, if you do a:

t := cnt

clkset(clkmode,clkfreq)·· 'rewrite CLKSEL with the same data

t := cnt - t

...While running at 80MHz, there will be a random ~50us variance in t which, at 80MHz, will translate into ~4000 units. The only problem with this is that the whole chip's master clock will slow to ~20KHz for one to two cycles, so a big time hiccup occurs.

I think that practically speaking, if you are waiting on some asynchronous external event to begin some process that needs a random number, the CNT will be effectively random, especially if the random ? operator is used.

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔


Chip Gracey
Parallax, Inc.

CJ

I thought CNT was clocked by whatever the chip is running at, so how would that get you a random number? unless CNT is remaining clocked by the 80Mhz during the 20Khz steps

please correct me if I am off

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Parallax Forums - If you're ready to learn, we're ready to help.

T Chap

Thanks for the ideas. There is a start event of the cycle, by any means really, including F10 or buttons. I have some real goals in mind for the random number, but, it's the weekend and fun stuff takes over now!

I want to be able to affect the outcome of the number with thoughts. If the number is above 0, a Green LED and a tone of 1k sounds on a speaker. If < 0, a Red LED and a of tone of 700 plays.

I was studying the supposedly "verifyable" effects of thought on the outcome of physical events from people concentrating on the test. If a ball can be affected as to where is lands after striking a number of randomly placed obstacles in it's downward path, then one of two choices of LEDs should be able to get affected as well. We'll soon see if this is true. I am testing Chips code now.

Post Edited (originator) : 3/17/2007 3:51:39 AM GMT

CJ

chip, I think I see what it gets you now, it is still ticking along at 80M until the 20k is ready, hops over for a couple, then hops back, leaving an indeterminate amount of ticks on CNT

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Parallax Forums - If you're ready to learn, we're ready to help.

Phil Pilgrim (PhiPi)

Here's something that appears to work. I connected a 100K resistor from A7 to A6, and another 100K resistor from A6 to ground. Then I programmed A7 to be the output pin for an A/D converter and A6 to be the feedback pin. 'No caps, just resistors. In this condition, A7 outputs at about about an 80% duty cycle, with random phase shifts. By reading A6 and shifting it into a 32-bit variable, then applying the ? operator a number of times, I'm getting what look to be random numbers. Of course, it will take further analysis to determine just how random they are.

Here's the program listing:

[b]CON[/b]

  [b]_clkmode[/b]      = [b]xtal[/b]1 + [b]pll[/b]16x
  [b]_xinfreq[/b]      = 5_000_000

  out           = 7
  inp           = 6 

[b]OBJ[/b]

  pr : "tv_wtext"
       
[b]PUB[/b] start

  pr.start(12)
  [b]dira[/b] := 1 << out
  [b]ctra[/b] := constant(%01001_000 << 23 | out << 9 | inp) 
  [b]repeat[/b] 32
    pr.hex(random, 8)
    pr.[b]str[/b]([b]string[/b](" "))
  [b]repeat[/b]

[b]PUB[/b] random | v

  [s]v := $12345678[/s] [i]Not needed. It was part of a different experiment.[/i]
  [b]repeat[/b] 32
    v := v << 1 | [b]ina[/b][noparse][[/noparse]inp]
  [b]repeat[/b] 32
    ?v
  [b]return[/b] v

-Phil

Post Edited (Phil Pilgrim (PhiPi)) : 3/17/2007 5:39:05 AM GMT

T Chap

Interesting idea, I'll plug it in as well, thanks Phil

Phil Pilgrim (PhiPi)

For your range of numbers, you can change the print statement to:

    pr.dec(random >> 1 // $56 - $2B)

-Phil

cgracey

You got it. It keeps running at 80MHz until the unrelated 20KHz edge occurs.

CJ said...
chip, I think I see what it gets you now, it is still ticking along at 80M until the 20k is ready, hops over for a couple, then hops back, leaving an indeterminate amount of ticks on CNT

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔


Chip Gracey
Parallax, Inc.