On Wed, 21 Apr 1999, Dwayne Reid wrote:

> G'day to all.
>
> I'm about to start a project where I need serial receive on a pic without
> onboard UART and was wondering if anyone had favourite routines they would
> be willing to share.  I'm looking at 9600 baud or slower and have to check
> for reception at any time - this means I'll be either calling the routine
> repeatedly or interspersing the 'check start bit' tests throughout my code.
> I've looked at the app note routines but still want to glean what I can from
> others that have solved real world problems.
>
> John Payson wrote about a neat concept some while back - sampling at 3X data
> rate and using a long shift register to accumulate the samples.  Sounds
> neat, John - how well does it handle speed errors?  When testing for the end
> of the byte (100 at the start, xx1 at the end), what do you do when the
> start sample has shifted to the test point in the SR chain but the ending
> 'xx1' isn't?  Do you check for 'x10' or 'xxx 1' and adjust the data bit
> sample locations accordingly?
>
> I'm also comtemplating a sampling routine similar to what Scott Dattalo
> helped me with a year back or so - it would actually be my first choice if I
> was trying to achieve 38K or higher data rates.
>
> This little critter also has to generate continous PWM.  The approach I'm
> leaning towards is to inline the RX routine, TX routine, PWM into one large
> loop that runs as the main task.  Of course, since most of my previous
> projects put all the house keeping stuff into the background routine, I
> don't have any working 'templates' to start from.  This project is starting
> with from a clean piece of paper.
>
> I'm open to any suggestions or ideas.  Thanks!

Dwayne,

Why do you get all of the fun projects?

One of the difficulties that I can see right off hand is that you have two
periodic tasks that are not harmonically related. One's the UART the other
is the PWM. (I'm assuming that the PWM is an all software one - If it
isn't and you're not concerned about cost then you'd be better off with a
PIC that supports both in hardware.  Unless there's something else I don't
know about.) It would probably make most sense to make the pwm synchronous
with the PIC's cpu clock (i.e. run it as fast as possible) and adjust the
timing for the UART using phase accumulators. I don't have a complete
solution available, however you may wish to investigate my all software
pwm routine:

http://www.interstice.com/~sdattalo/technical/software/pic/pwm256.txt

This routine works on a 12-bit or 14-bit core and has single instruction
cycle resolution. There are 256 pwm steps and the shortest step can be one
instruction cycle. On a pic with a 20Mhz clock, the pwm frequency is

20e6 / 4 / 256 = 19531.25 Hz

This is about twice the frequency of 9600 baud.

There's roughly 145 cycles out of the 256 available for other things (like
the UART). It also possible to sample the UART twice per pass through the
PWM loop and have an effect 4x sample rate. In either case there will be
some jitter in sampling time, although I don't think that this would be a
problem (certainly not on the receive).

Scott