Good point. But buying good resistors of varying values and mixing them up
is not very practical in production. They use reel components.

I will use a simply algorithm to distribute the random number. I shift the
number so that the lest-significant bit becomes the most significant and the
most becomes the lest. This way the number will be distributed fairly even.
For example, two numbers 42 (0b00101010) and 43 (0b00101011) are next to
each other; but after I shift them, they will become 84 (0b01010100) and 212
(0b11010100) -- broadly separated. So I am not too worry about the
distribution of the random number. As long as they are unique each board /
each time, I can distribute it myself.

Tony

----- Original Message -----
From: "Don Hyde" <DonH@AXONN.COM>
To: <PICLIST@MITVMA.MIT.EDU>
Sent: Tuesday, August 21, 2001 11:06 AM
Subject: Re: [PIC]: Generating a "truly" random number on the application of
power


> If you want your boards to have a random dispersion due to component
> variation, 20% resistors might not be the best choice.
>
> Resistors are often offered in different precision grades, which is
achieved
> by selecting the closest-tolerance parts from a production run.  The ones
> that make 1% or 5% go into the higher-priced bin, and the rest go into the
> cheap bin.
>
> The bunch in the cheap bin are unlikely to have a normal distribution
about
> the center value.  If the process is well-tuned and the ones coming off
the
> line have a nice distribution about the target value, then the cheapie
parts
> will have a hole in the distribution where the good ones were taken out.
> They may all be within 20%, but NONE will be within 5%!
>
> Getting your random variation from a part like a cap that isn't sold in
> varying precision grades might give a more dependably random distribution
of
> times.
>
> Maybe you could buy good resistors of varying values and mix your own
random
> selection by dumping them in a barrel and shaking them up.
>
> > -----Original Message-----
> > From: Wollenberg, Frank [mailto:F.Wollenberg@GSP-BERLIN.DE]
> > Sent: Tuesday, August 21, 2001 9:21 AM
> > To: PICLIST@MITVMA.MIT.EDU
> > Subject: Re: [PIC]: Generating a "truly" random number on the
> > application of power
> >
> >
> > Tony Pan wrote:
> >
> > > Our products are used in motel environments to control ACs.
> > > The reason we
> > > need to stagger the startup is that if we don't, in case of
> > > power loss and
> > > resume, all the AC will start at the same time and that will cause a
> > > problem. As you know each board is situated in its room and working
> > > independently so there is no way one board knows what the
> > > other's doing.
> >
> > AC will start on power-resume AND power-on..
> >
> > > In my hardware we use RC oscillator. Now I use the initial
> > > TMR0 value upon
> > > powerup to get a random number. I also add the initial values
> > > of FSR, ADRES,
> > > PORTB and some ADC reads of the board into the number. I
> > > guess that would
> > > give me nearly "unique" number for each board. I then use the
> > > last few bits
> > > of this number to determine the startup time.
> >
> > Using TMR0 is ok when using RC oscillator. Don't use TMR0 in crystal
> > operation.
> > FSR and ADRES are uninit on powerup, but i think they won't differ,
> > especially if the PIC's are from the same silizium.
> > PORTB must be input, but because of the identical hardware, the values
> > should be identical.
> > Reading a voltage divider with ADC is OK. Use carbon
> > resistors (20%). Try to
> > not meet the acquisition time, the Chold is varying from chip
> > to chip. Also
> > you can use an external reference voltage from a voltage
> > divider (carbon!).
> >
> > Compute this value ONLY on first powerup and use it as a
> > initial seed value
> > of a random generator (hint: in C: persistent). Compute a
> > random number as a
> > delay time after power becomes OK.
> >
> > > I haven't got the chance to test it yet. For we have other
> > features to
> > > implement at the moment. If you see any potential problems with this
> > > approach, please let me know.
> >
> > It is possible to compute identical delay times, but the
> > possibility is very
> > small.
> > Another approach is to use a "digital ID chip" from DALLAS
> > (DS2401 i think).
> > This number IS unique !!!
> >
> > Frank
> > --------------------------------------
> > GSP Sprachtechnologie GmbH
> > Frank Wollenberg
> > HW-Entwicklung
> > Tel.:   +49 (0)30 769929-78
> > Fax:    +49 (0)30 769929-12
> > eMail:  f.wollenberg@gsp-berlin.de
> >
> >
> > --
> > GSP Sprachtechnologie GmbH
> > Teltowkanalstr.1, D-12247 Berlin
> > Tel.:  +49 (0)30 769929-0
> > Fax:   +49 (0)30 769929-12
> > eMail: Info@gsp-berlin.de
> > Web:   http://www.gsp-berlin.de
> >
> > --
> > http://www.piclist.com hint: The list server can filter out subtopics
> > (like ads or off topics) for you. See http://www.piclist.com/#topics
> >
> >
>
> --
> http://www.piclist.com hint: The list server can filter out subtopics
> (like ads or off topics) for you. See http://www.piclist.com/#topics
>
>
>

--
http://www.piclist.com hint: The list server can filter out subtopics
(like ads or off topics) for you. See http://www.piclist.com/#topics