Re: [Gumstix-users] robostix clock rate

[Florian L. <gu...@fl...>]

On Saturday 21 May 2005 16:13, Dave Hylands wrote:
> Hi Florian,
>
> > This would mean, that reliable communications with the robostix are only
> > possible at 250k or higher...
> > (ATmega128 specs, page 197, table 85: Examples of Baud Rate Settings.)
> > At 4800, 19.2k, 38.4k and 76.8k one would "only" have 0.2% error-rate,
> > but afaik this is a "per bit" error-rating. When multiplying by 8 the
> > error-rate would be 1.6% bad bytes. Suppose we checksum 10bytes. In this
> > case we would loose 16% of our sent data. (In our specific case we
> > checksummed 13bytes, and lost more than 20%...).
>
> I think that your analysis is flawed ;)
>
> The error that's specified is the difference in clock frequencies. A
> 1.6% error in clock frequency won't cause any problems with
> communication. The real secret here is that async communication is
> occuring and everything gets resynced every character.
>
> The communication can handle upto to 1/2 bit of clock difference over
> 10 bits, which means that the clocks can differ by as much as 5% and
> the two devices can still communicate at any given baud rate.
Mea Culpa...
We nevertheless had *big* troubles with Atmels that used their internal 
oscillators. (20% bad checksums over 13 sent bytes). Perhaps we should have 
calibrated the oscillators (there seems to be a document on Atmel's site). 
With external oscillators everything worked fine. (we only tried the 0.0% 
error frequencies).
As long as the communication will work fine, I'm fine with 16MHz ;)
// florian

-- 
This function will terminate, if run infinitely.
  void f() { while (random() != 0); }