HAL_TIM_Base_Stop() and HAL_TIM_Base_Start() adds time? (trying to have a nanoseconds (100ns) control delay using a BASE TIMER)

[Luis Ber]

Hello,

I need to control signals in the order of nanoseconds magnitude (~ 100 ns). For this purpose, I have programmed a hardware timer in order to have an interrupt with 100 ns delay.

I'm using a STM32F746VE microcontroller at 216 MHz. I'm using the TIM6 as a Base_TIM and I'm programming it with the following configuration:

timConfig.Prescaler = 1;

timConfig.CounterMode = TIM_COUNTERMODE_UP;

timConfig.ClockDivision = TIM_CLOCKDIVISION_DIV1;

timConfig.RepetitionCounter = 0;

timConfig.Period = 5;

timConfig.AutoReloadPreload = TIM_AUTORELOAD_PRELOAD_ENABLE;

Supposing that APB frequency of the bus is 108 MHz, this is a 111 ns timer (approx.).

Then in order, to measure the real delay of this timer I toggle a GPIO pin in its interruption handler, and measuring the time with an oscilloscope. The interruption handler is the following:

void TIM6_DAC_IRQHandler(void)
{
	HAL_TIM_Base_Stop_IT(HAL_TIM6);
	
	if ( i == 0 )
	{
		GPIOE->BSRR = GPIO_PIN_3;
		i = 1;
	}
	else
	{
		GPIOE->BSRR = (uint32_t)GPIO_PIN_3 << 16;
		i = 0;	
	}
	HAL_TIM_Base_Start_IT(HAL_TIM6);
}

Then, measuring the delay of the interruption is 1.4 us. And if I comment the Stop() and Start() functions I achieve a delay of 235 ns. But not the theoretical 111ns calculated previously with the formula ( T = (1/APB_TIM_CLK) * (PRESCALER_Value + 1) * (PERIOD_Value + 1) ).

Where is the problem here? What happens with HAL_TIME_Base_Start_IT() and HAL_TIME_BASE_Stop_IT()? These funcions add a lot of time.

Thank you in advance.

[Tesla DeLorean]

You can't expect to interrupt at 9 MHz !!

Period and Prescaler are N-1

What are the APB clock dividers here?

Set the period to be maximal, and observe elapsed ticks via TIM6->CNT

>>What happens with HAL_TIME_Base_Start_IT() and HAL_TIME_BASE_Stop_IT()? These funcions add a lot of time.

You could use DWT_CYCCNT to measure the elapsed time in 216 MHz ticks.

You could review the listing file and see how much code is actually involved.

[waclawek.jan]

> signals in the order of nanoseconds magnitude (~ 100 ns)

That's IMO a case for legitimate use of cycle-waste delays. Not that those are trivial on an 'F7.

Still, generating signals in hardware is the first choice in the 32-bitters.

JW

[Luis Ber]

Hello,

I have set the period to the maximum value, and then I have used TIM6->CNT in order to perform the nanosecond delay. I have achieved to toggle a pin with 150ns delay. This means a 3.33 MHz frequency. Is possible to improve this value? I suppose that the thing limiting this time, is the time that needs the microcontroller to move the GPIO?

I have used DWT_CYCCNT to measure the Start() and Stop() functions. Each function needs 700ns (this means 1.4us), that is a lot of time...

Thank you a lot!

[berendi]

HAL functions are excruciatingly slow, avoid even looking at their general direction when dealing with intervals shorter than 100 microseconds.

You have DWT->CYCCNT, use that to measure 100 ns.

Your system is running on 216 MHz, so 100 ns is 21.6 cycles. It is barely enough to start executing the interrupt handler.