STM32L4 Timer: External clock count both edges

[PKulh]

Hello,

I need to count the number of (positive and negative) edges in a signal within certain time period. For that I use a timer on STM32L452. The AN4776 says that External Clock Mode 1 is able to update the counter on both edges (for the price of not being able to trigger). I tried this (see the code below) and figured out that with a 50% duty cycle square test signal of 100mHz the CNT register is incremented every 10 seconds by 2. I would expect the CNT to increment by 1 every 5 seconds.

What am I doing wrong?

Thanks

Petr

Below is my code using TIM2 CH1 for clock and CH3 for capture (without clock, DMA and interrupt initialization):

// ---------- initialize GPIO ------------
    GPIO_InitTypeDef GpioInitStruct;
 
    // initialize the bitstream pin
    GpioInitStruct.Pin = GPIO_PIN_5;
    GpioInitStruct.Speed = GPIO_SPEED_FREQ_HIGH;
    GpioInitStruct.Mode = GPIO_MODE_AF_PP;
    GpioInitStruct.Pull = GPIO_NOPULL;
    GpioInitStruct.Alternate = GPIO_AF1_TIM2;
    HAL_GPIO_Init(GPIOA, &GpioInitStruct);
 
    // initialize the sync pin
    GpioInitStruct.Pin = GPIO_PIN_2;
    GpioInitStruct.Speed = GPIO_SPEED_FREQ_HIGH;
    GpioInitStruct.Mode = GPIO_MODE_AF_PP;
    GpioInitStruct.Pull = GPIO_NOPULL;
    GpioInitStruct.Alternate = GPIO_AF1_TIM2;
    HAL_GPIO_Init(GPIOA, &GpioInitStruct);
 
    // ------------ initialize timer -------------
    // basic initialization of the structure
    TIM_HandleTypeDef *pTmrHandle;
    pTmrHandle->Instance = TIM2;
    pTmrHandle->Init.Prescaler = 0;
    pTmrHandle->Init.CounterMode = TIM_COUNTERMODE_UP;
    pTmrHandle->Init.Period = 0xffff;
    pTmrHandle->Init.ClockDivision = TIM_CLOCKDIVISION_DIV1;
    pTmrHandle->Init.RepetitionCounter = 0;
    pTmrHandle->Init.AutoReloadPreload = TIM_AUTORELOAD_PRELOAD_DISABLE;
 
    // configure the clock source
    TIM_ClockConfigTypeDef ClockSourceConfig;
    ClockSourceConfig.ClockFilter = 0;                           // no filter
    ClockSourceConfig.ClockPrescaler = TIM_CLOCKPRESCALER_DIV1;  // no prescaler
    ClockSourceConfig.ClockPolarity = TIM_CLOCKPOLARITY_BOTHEDGE;   // trigger on both edges
    ClockSourceConfig.ClockSource = TIM_CLOCKSOURCE_TI1ED;          // both edges
 
    // select the clock using the above clock configuration
    if (HAL_TIM_ConfigClockSource(pTmrHandle, &ClockSourceConfig) != HAL_OK) {
        return false;
    }
 
    // initialize input capture time base
    if (HAL_TIM_IC_Init(pTmrHandle) != HAL_OK) {
        return false;
    }
 
    // configure the input-capture channel (triggered by the sync signal) to trigger on both edges
    TIM_IC_InitTypeDef InputCaptureConfig = {
            .ICPolarity = TIM_INPUTCHANNELPOLARITY_RISING,          // trigger on the rising edge
            .ICSelection = TIM_ICSELECTION_DIRECTTI,                // direct mapping of timer inputs
            .ICPrescaler = TIM_ICPSC_DIV1,                          // no divider - we want to count all edges
            .ICFilter = 0,                                          // no filter
    };
    // trigger on the rising edge of CH3 (sync)
    if (HAL_TIM_IC_ConfigChannel(pTmrHandle, &InputCaptureConfig, TIM_CHANNEL_3) != HAL_OK) {
        return false;
    }
 
    // configure both edges on CH1 (clock)
    InputCaptureConfig.ICPolarity = TIM_INPUTCHANNELPOLARITY_BOTHEDGE;
    if (HAL_TIM_IC_ConfigChannel(pTmrHandle, &InputCaptureConfig, TIM_CHANNEL_1) != HAL_OK) {
        return false;
    }
 
 
// to start reception:
    if (HAL_TIM_IC_Start_DMA(pTmrHandle, TIM_CHANNEL_3,  (uint32_t*)dmaBuffer_, 2 * dmaBlockSize_) != HAL_OK) {
        return false;
    }

[PKulh]

The system clock (or the TIM internal clock) is 80MHz

One more thing that I noticed is the following odd behaviour:

If I apply a 10Hz signal on channel 3 (which is the input-capture) the CNT counts up for 5 seconds (up to the value 202), then does not count for 5 seconds, then counts again, etc. It seems the trigger (coming from CH3) generates pulses, which are gated by the supposed clock signal. Can you understand why is that?

[waclawek.jan]

You messed up the two pins?

What happens if you remove the DMA code?

JW

[PKulh]

Indeed, I measure the signals on PA2 and PA5 they are exactly what the signal generator is supposed to generate. And they are definitely not mixed up.

[waclawek.jan]

> I measure the signals on PA2 and PA5 they are exactly what the signal generator is supposed to generate.

Did you measure *directly* on the pins, i.e. probe physically touching pins on the mcu's package?

With the 10Hz signal on CH3, DMA's NDTR counts while the CNT does not count?

I'm aiming at a short between the two signals, behaving as an AND.

JW

[PKulh]

Without DMA (using HAL_TIM_IC_Start() ) is the same.

The pins are not physically messed up, the oscilloscope shows a clean signal on both, not mixed up.

By the way, it doesn't make any difference if I take TI1F_ED as clock input (SMCR:TS=4) or TI1FP1 (SMCR:TS=5) - the result is always the same, which is odd.

Is there a better schematics of the STM32L timers somewhere? The pictures (Figure 268) in the RM0394 do not correspond with the description. The description says both-edge detection is selected by setting both CCxP and CCxNP bits in CCER register to 1, whereas the picture shows only 2 values in the corresponding mux.

[PKulh]

Also when measured directly on the package the signal is clean.

With the 10Hz on CH3 the CNT counts if CH1 is high.

[waclawek.jan]

I mean, whether NDTR changes when CNT counts, and whether NDTR stops when CNT stops counting.

Also, try minimal code without DMA/CH3 setup.

Also, try different pins, perhaps also different TIM.

I don't have a 'L452. On a 'F4, I've set up TIM9 with PE5=CH1 and basically your setup (without CH3) and it counted as expected.

JW

[waclawek.jan]

Looking for whether PA5 is somehow special, I noticed it's not 5V tolerant - is your input signal amplitude 3V?

[PKulh]

Yes, the signal is 3V :)

[waclawek.jan]

> By the way, it doesn't make any difference if I take TI1F_ED as clock input (SMCR:TS=4) or TI1FP1 (SMCR:TS=5) - the result is always the same, which is odd.

This is expected - if you'd change the setting in CCxP/CCxNP, TI1F_ED should still count on each edge, whereas TI1FP1 should not.

> Is there a better schematics of the STM32L timers somewhere?

I don't think so. Think of it more as a rough diagram rather than schematics; it has more defficiencies when it comes to the minute details.

JW