Can't find old topic on clock compensation

Maybe it's just cheap Chinese junk...

Or it is being mis-used.

Well, perhaps. Atmel distributes "firmware" with note AVR4001 that theoretically outputs the watch-crystal signal—but it didn't have anything for the 328P, only the 168 or 169. Maybe I can't actually output that clock signal. That could well be it.

randomizer:
note AVR4001

Got a link to that? Striking out searching Atmel's site...

it didn't have anything for the 328P

The purpose of an application note is to show you a concept so that you can understand it and take it to your particular application / target.

dhenry:
The purpose of an application note is to show you a concept so that you can understand it and take it to your particular application / target.

The firmware (attached) includes sample code for various Atmel processors, including the 324 and 329, but not the 328. I did look through that and the datasheet to do the same thing on the 328, but I'm thinking it might not be an option there. (I now see chapter 8.9 says the system clock is output on CLKO. So. I was wrong to try.)

Note AVR4001

I'm thinking it might not be an option there.

What made you think so?

randomizer:
The firmware (attached) includes sample code for various Atmel processors, including the 324 and 329, but not the 328. I did look through that and the datasheet to do the same thing on the 328, but I'm thinking it might not be an option there. (I now see chapter 8.9 says the system clock is output on CLKO. So. I was wrong to try.)

Note AVR4001

Thanks, I sure couldn't find it. And I think I'd seen it before, I remember the pics of the crazy crystal attachment techniques. Might have to look again.

Anyway, I was thinking along these lines as well:

5.2 MegaAVR
An asynchronous timer overflow is used to toggle an I/O pin, and hence the clock
signal will be divided by 2 (nominal frequency of 16384Hz). All megaAVR® devices
are supported, but a device family define needs to be set (see list of defines in the .c
file).

Did you mean to attach something? (Maybe I'm not seeing that either, LOL!) It'd be easy enough to code up the timer as described above. In fact, now I'm curious about it, and I've just cobbled together this frequency counter...

Funny, I thought I did! Here's the download right from Atmel: http://www.atmel.com/Images/AVR4100.zip

EDIT: Whoa. I just now noticed that it's 4100, not 4001. I've been looking at that doc for a couple weeks now, and always thought it was 4001. Sorry for any unnecessary hassle!

randomizer:

[quote author=Jack Christensen link=topic=145772.msg1096626#msg1096626 date=1359575605]
Did you mean to attach something?

Funny, I thought I did! Here's the download right from Atmel: http://www.atmel.com/Images/AVR4100.zip

EDIT: Whoa. I just now noticed that it's 4100, not 4001. I've been looking at that doc for a couple weeks now, and always thought it was 4001. Sorry for any unnecessary hassle!
[/quote]

Ha! No worries, I didn't quite make a career out of searching for it!

Got it! I realized I was being too lame—expecting the Atmel code to work perfectly. This is how I set it up (with an interrupt cobbled from various people online):

void setup() { 
delay(1000); //give watch crystal time to stabilize

// initialize Timer2
    cli();          // disable global interrupts
    
    ASSR = (1<<AS2);
    
    TCCR2A = 0;     // set entire TCCR2A register to 0 
    TCCR2B = 0;     // same for TCCR2B
    TCCR2A |= (1 << WGM21); // Configure timer 2 for CTC mode
    TIMSK2 |= (1 << OCIE2A); // Enable CTC interrupt
    TCCR2B = (0<<WGM22)|(0<<CS22)|(0<<CS21)|(1<<CS20); //no prescaler
    OCR2A = 0;           //trigger interrupt every clock cycle
    
    
    // enable global interrupts:
    sei();

pinMode(9, OUTPUT);
}


ISR(TIMER2_COMPA_vect){
    digitalWrite(9,HIGH);
    digitalWrite(9,LOW);
}

void loop() {}

I get 16.38 kHz on my DMM. Assuming that it's not trying to say 16.384, that's ~120ppm (thanks Wolfram|Alpha!). So I don't see why my clock should have been so far off. Perhaps any inaccuracies in the chip get amplified with the prescaler? For the clock, I was prescaling by 1024 to get a 1Hz interrupt...

EDIT: Hmmmmmm. Setting OCR2A to 33 instead of 0 (this is really useful), I get 952.5 Hz instead of 1000 Hz. Or I was, until the DMM decided to quit on me. Aaaaaarg.

    OCR2A = 0;           //trigger interrupt every clock cycle

You can't usefully do that with no prescaler. Servicing an interrupt takes at least 19 clock cycles (see: http://www.gammon.com.au/interrupts), then your digital writes will take time, plus the time to leave the ISR.

expecting the Atmel code to work perfectly.

It will work perfectly, on its intended target.

All you need to do is to understand what the code is doing, read the datasheet of your own target and replicate the same on your target: aka what an application note is designed to do.

19 system clock cycles, yes? I'm clocking TIMER2 asynchronously. That's how the Atmel code did it:

#ifdef ATMEGA48_88_168
int main( void )
{
	// Crystal clock will be output on PORTB3 (divided by 2)
	DDRB |= (1<<PORTB3);

	// Enable CTC mode with toggle on compare match. Precale 1 and top at 0
	OCR2A  =  0;
	TCCR2A = (0<<COM2A1)|(1<<COM2A0)|(1<<WGM21)|(0<<WGM20);
	TCCR2B = (0<<WGM22)|(0<<CS22)|(0<<CS21)|(1<<CS20);
	ASSR   = (1<<AS2);

	while(1){}
}
#endif

Not using timer interrupts, just using output compare to make its frequency measurable on a pin. Well, half it's frequency is the best that can be done.

randomizer:
19 system clock cycles, yes? I'm clocking TIMER2 asynchronously. That's how the Atmel code did it:

Oh, OK. Sure.

19 system clock cycles, yes?

No. The latency varies, depending on many factors, starting from as low as 2.

OK, I'm calling you on this one, dhenry. From the datasheet, which you keep recommending people read:

Since we do indeed use an interrupt vector, that is 7 cycles minimum. Add onto that the code in the ISR to, at the very least, save the status register (SREG) and that is another 3 cycles. On top of that you have to push any registers you are planning to use, unless your ISR doesn't do anything.

So, no, it is not "as low as 2".

I was wrong about the 19 cycles. That is the time take, from looking at actual code generated, to leave the ISR. It is 23 cycles to enter it. That would probably vary depending on what the ISR does. The more things it does, the more registers the compiler has to save (and restore later).

Figures, disassembly, etc. on this page:

It's interesting—with OCR2A set to 0 or 1, I get, as expected, half of 32768—1638(4). But when I up OCR2A to ~31,32,33 to get a 1kHz frequency, I get a 1kHz frequency (or close to it)—not 500Hz. I think that's because in the interrupt, I have

digitalWrite(9,HIGH);
    digitalWrite(9,LOW);

instead of

if (toggle2){
    digitalWrite(9,HIGH);
    toggle2 = 0;
  }
  else{
    digitalWrite(9,LOW);
    toggle2 = 1;
  }

which is what it was originally. At the relatively high frequencies, the minuscule gap between HIGH and LOW is... something, and at lower frequencies it... isn't. Not really sure what I'm saying exactly, but I know what I mean. :slight_smile:

At OCR = 0, you get 16khz;
OCR = 1, you get 8khz;
OCR = 31, you get 0.5khz.
...

Well, with the original code (using the toggle) I find that OCR=1 generates 1/2 of 16384Hz (interestingly, so does OCR=0). At least, it should. I'm reading 8190, which is ~250ppm off—way more than the ~40 expected from the temperature. That could well be errors in the DMM—I don't really see what else, unless the crystal is picking up parasitic capacitance through the air from the breadboard rails underneath it.

That translates to 21 seconds per day slow, if I define a second as 8292 interrupts. (Or I could just code it as 8190, and assume that my meter isn't off.) That's not ideal, but it's not the second-every-five-minutes loss I was having before.

One thing I've determined: The calculator I linked to earlier rounds the answer it spits out—when I say I want 4 kHz, for example, it tells me I can use no prescaler and OCR=8 and have 0% error, but the fact is I will have error because everything's in factors of two. Nominal 8 kHz is actually 8192 Hz. This is possibly why the clock was so slow earlier.