AVR-Chat

I agree that it is a major pain in the neck having powers of 2 as 
the prescaler, but NOT powers of 2 as the usual crystal freq.

In my case I needed a 1 millisecond clock.  So, I used a 16 bit 
clock for my 16MHz crystal.  (8 bits would not do).

I uses CTC mode with a prescaler of zero, and triggered the 
interrupt after 16,000 clock cycles.  (16e6/16e3 = 1e3).  This 
worked out fine.  

The crystal's accuracy is rated for +/- 30 ppm.  I found the 
accuracy was 7ppm.  I also let the clock run for 2 days and then 
calculated a correction factor.  A "leap millisecond" is added or 
subtracted every so often.  It is very accurate now.  I need 
something that is better than a second per day (1 sec per day is 
11.7ppm).  I am sure I can get a better calibration if I were to run 
the clock for a week, but this is sufficient for my current needs.

I initially found that the error was MUCH higher, at 62ppm, but then 
I realized that this was about 1:16000.  I then realized that I had 
the counters interrupt trigger at the count of 16,000,  but this 
would occur on the 16,001th clock cycle.  So once I changed the 
timer to trigger on 15,999 and this was great.

For your 1mhz crystal, 16 bit timer, you would set the timer to 
trigger at 999 with a prescaler of zero.  You could also use a 
different prescaler and a smaller comparison value.  You have to 
check all the possibilities, but 1 mhz clock, prescale of 8 and a 
count of 125 works, on an 8 bit timer, and gives a millisecond 
resolution timer.

So, I have found that it is quite possible to get a reasonable 
millisecond resolution on the AVR timers.  I have also found that 
extremely good accuracy is possible with a simple calibration.

-Tony





--- In AVR-Chat@yahoogroups.com, "jeff_smith_8992" 
<jeff_smith_8992@y...> wrote:
> 
> When I use the AVR's internal oscillator or an external crystal
> oscillator, like 1.0 MHz, is the frequency 1,000,000 Hz (give or 
take
> a few Hz for accuracy, temp, etc)?  And not a power of 2, like 2 to
> the 20th = 1,048,576?
> 
> The reason I'm asking, is because the prescaler divisors are all
> powers of 2 (i.e., 2, 4, 8, 16, 32, etc), and this makes it 
difficult
> to use the prescaler to get whole number frequencies.  For 
instance,
> if I use a 1 MHz clock and the 1024 prescaler, it gives an 
interval of
> 0.001024 seconds instead of a nice, even number like 0.001 seconds.
> 
> I guess my real question is, "Given a particular input clock, how 
can
> I make a timer that counts in, say, one millisecond intervals"?  
And
> how can I keep it as accurate as possible over a long period of 
time?
> 
> Thanks for any help.
> 
> Jeff