Strange values from HAL_TIM_ReadCapturedValue

[iforce2d]

I'm using input capture with a 16 bit timer to read a 1.5ms pulse that occurs every 20ms. I can see on my scope that the input pulse conforms to this exactly. Here's the code for the interrupt callback:

uint16_t pwmVal0 = 0;
uint16_t pwmVal1 = 0;
uint16_t pwmLength = 0;
uint8_t whichEdge = 0;
 
void HAL_TIM_IC_CaptureCallback(TIM_HandleTypeDef *htim)
{
	if ( htim == &htim3 && htim->Channel == HAL_TIM_ACTIVE_CHANNEL_1 ) {
		if ( whichEdge == 0 ) {
			//HAL_GPIO_WritePin(GPIOD, GPIO_PIN_2, GPIO_PIN_SET);
 
			pwmVal0 = HAL_TIM_ReadCapturedValue( htim, TIM_CHANNEL_1 );
			__HAL_TIM_SET_CAPTUREPOLARITY( htim, TIM_CHANNEL_1, TIM_INPUTCHANNELPOLARITY_FALLING );
			whichEdge = 1;
		}
		else {
			//HAL_GPIO_WritePin(GPIOD, GPIO_PIN_2, GPIO_PIN_RESET);
 
			pwmVal1 = HAL_TIM_ReadCapturedValue( htim, TIM_CHANNEL_1 );
			__HAL_TIM_SET_CAPTUREPOLARITY( htim, TIM_CHANNEL_1, TIM_INPUTCHANNELPOLARITY_RISING );
			whichEdge = 0;
 
			pwmLength = pwmVal1 - pwmVal0;
		}
	}
}

Note the two commented lines. When I look at the pulses made by those on my scope, the detected waveform is exactly what I expect and is rock solid. So the interrupt is triggering just fine. The problem is that the values returned by HAL_TIM_ReadCapturedValue (and used to calculate my final goal value 'pwmLength') are only correct about 70% of the time. Here's the result:

Every value in this array should be around 1500, but there are consistently values around 62941 showing up, with a regular repetition.

What's weird is I've used this method many times in the past with no such problems. Is it possible that the hardware is at fault maybe?

[T J]

65535 - 62941 = 2594 ?

do you have the auto-reload set to 65535 - 2594 +1500 = 64441 = 0xFBB9 ?

[TDK]

It's really unlikely that the hardware is at fault. I can't see any issues with the code you posted. You could define those variables as volatile but I doubt that's it.

[waclawek.jan]

> do you have the auto-reload set to 65535 - 2594 +1500 = 64441 = 0xFBB9 ?

No, to 4096.

And the "wrong" result comes exactly in 1500/4096*100% of cases, cca 37%.

The solution is called modulo arithmetics. There's also no need to use anything but the maximum ARR.

JW

PS Always start with stating, which STM32 is you question related to.

[iforce2d]

Thanks for the tips, turns out you're correct.

Somehow despite spending almost 4 hours on this and double-checking everything a dozen times, I manage to miss this typo here every time I checked:

htim3.Init.Period = 0xfff-1;

It didn't help that I made the exact same typo in a second project I set up specifically to test the minimal case. I also noticed that when the pin was floating it would trigger interrupts at a similar rate to the 'wrong' values (50Hz possibly from mains AC?) which sent me down a pointless rabbit hole of suspecting hardware issues.

Anyway with the correct value of 0xffff-1 everything is fine. Maybe I shouldn't be coding so late into the night with a foggy brain...

[TDK]

The correct value is 0xFFFF, not 1 less than that.