On Fri, 27 Feb 2004 23:12:26 +0200, Peter L. Peres wrote on
[PIC]: 1-Wire bus noise sensitivity:

>Try to express your error rate as a bit error rate on the bus. Eg if for
>each read value you need to send and receive 20 bits then you actually
>have one error in 20*2000 bits which is respectable for a noisy wire in a
>smpsu I think (it is a S/N of 66db if I am not mistaken). You can probably
>improve it by adding RLC filtering at both ends and by rewiring the
>harness so the adjacent wires carry signal ground (and not some other
>ground). The math behind this is that of the Nyquist/Shannon bandwidth vs.
>bit rate equation. You want to limit the bandwidth as much as possible
>using the RLC groups.

Thanks for this answer; I have installed a spy in my
software, and my current estimate is that about 10 out of
2000 temperature reads are in error, where one temperature
read implies reading 9 bytes; if I suppose that in a false
reading only one bit is wrong, this means I have 1 errors in
(2000 * 72 bits) / 10, about one third your figure above.

Since my first message, I have implemented error detection
by CRC and error correction, which improves things a lot;
also I have lowered the impedance of the 1-wire bus to half
the suggested value of 4k7.

There is no way I could rewire the harness: my switching
power supply feeds a CCD camera through a commercial 25 wire
shielded cable; there is no independent shield for the bus
which travels along with the Peltier module power wires.


* Xrobert.soubie@free.frX (veuillez supprimer les "X")
* http://www.astrosurf.com/soubie
* Au royaume des aveugles, les borgnes sont mal vus... - P.Dac

--
http://www.piclist.com hint: The PICList is archived three different
ways.  See http://www.piclist.com/#archives for details.