App_timer (RTC) measuring Timer periodic interrupts - unexpected results

[davege]

Hello,

While working with a bare metal (no SoftDevice or OS) test application to prove code functionalities, I ran out in a weird behaviour that I cannot explain.

I'm using a HW Timer (NRFX_TIMER) to trigger periodically some routines.

Since I wanted to prove the periodicity of the operation, I re-used an RTC that I init and start at the boot to count the ticks difference in a period.

We all know the max resolution of RTC is just 30.5us, so I expected some error, but I've actually found out that the difference I see is around 3ms every 100ms, which is impossible to happen with a working configuration.
I'm sure that this is a problem of how I'm configuring stuff, so I reproduced the problem on a dev kit (PCA10040).

I took as a base the timer peripheral application, which is using a Timer (https://infocenter.nordicsemi.com/topic/sdk_nrf5_v16.0.0/nrf_dev_timer_example.html) to blink LEDs.

I modified it in the below points:

Function timer_led_event_handler:

/**
 * @brief Handler for timer events.
 */
void timer_led_event_handler(nrf_timer_event_t event_type, void* p_context)
{
    static uint32_t prev = 0;
    uint32_t now = 0;
    uint32_t diff = 0;
    static uint32_t i;
    uint32_t led_to_invert = ((i++) % LEDS_NUMBER);

    switch (event_type)
    {
        case NRF_TIMER_EVENT_COMPARE0:
            now = app_timer_cnt_get();
            diff = app_timer_cnt_diff_compute(now, prev);
            NRF_LOG_INFO("Ticks Elapsed %d", diff);
            NRF_LOG_FLUSH();
            prev = now;

            bsp_board_led_invert(led_to_invert);

            break;

        default:
            //Do nothing.
            break;
    }
}

Then in the main:

 * @brief Function for main application entry.
 */
int main(void)
{
    uint32_t time_ms = 1000; //Time(in miliseconds) between consecutive compare events.
    uint32_t time_ticks;
    uint32_t err_code = NRF_SUCCESS;

    //Configure all leds on board.
    bsp_board_init(BSP_INIT_LEDS);

    err_code = NRF_LOG_INIT(NULL);

    /* Create logging task if no errors */
    if (NRF_SUCCESS == err_code)
    {
        NRF_LOG_DEFAULT_BACKENDS_INIT();
    }

    err_code = nrf_drv_clock_init();
    nrf_drv_clock_lfclk_request(NULL);
    //while(false == nrfx_clock_lfclk_is_running()){};
    err_code = app_timer_init();
    err_code = app_timer_create(&executionTimer, APP_TIMER_MODE_SINGLE_SHOT, AppTimerTimeout_Cback);
    if (NRF_SUCCESS == err_code)
    {
        err_code = app_timer_start(executionTimer, TEST_TIMEOUT, NULL);
    }

    NRF_LOG_INFO("Test started");
    NRF_LOG_FLUSH();
    
    ...
}

Of course I also included in the project all the dependencies needed to compile and modified sdk_config on purpose.

If I set time_ms to 100ms, I see as output:

00> <info> app: Ticks Elapsed 3230
00> 
00> <info> app: Ticks Elapsed 3230
00> 
00> <info> app: Ticks Elapsed 3230
00> 
00> <info> app: Ticks Elapsed 3229
00> 
00> <info> app: Ticks Elapsed 3230
00> 
00> <info> app: Ticks Elapsed 3229
00> 
00> <info> app: Ticks Elapsed 3229
00> 
00> <info> app: Ticks Elapsed 3230

So elapsed time is around 98.5ms.

With 500ms:

00> <info> app: Ticks Elapsed 16151
00> 
00> <info> app: Ticks Elapsed 16148
00> 
00> <info> app: Ticks Elapsed 16154
00> 
00> <info> app: Ticks Elapsed 16156
00> 
00> <info> app: Ticks Elapsed 16155
00> 
00> <info> app: Ticks Elapsed 16153
00> 
00> <info> app: Ticks Elapsed 16152
00> 
00> <info> app: Ticks Elapsed 16159
00> 
00> <info> app: Ticks Elapsed 16160

So elapsed time is around 493ms.

This is continuing linearly, since when time_ms is 1000 the elapsed time detected with RTC is around 986ms.

It may be something very stupid, but may I ask if someone understands where the problem may be?

Information about the system:

• running on nRF52 Dev Kit (PCA10040)
• RTC frequency is 32768 Hz
• CLOCK_CONFIG_LF_SRC/NRFX_CLOCK_CONFIG_LF_SRC is set to XTAL
• NRFX_TIMER_DEFAULT_CONFIG_FREQUENCY is 16MHZ, with NRFX_TIMER_DEFAULT_CONFIG_BIT_WIDTH of 16bit
• All interrupts are using priority 6 (default) - since this is the only thing happening in the system I doubt that this is affecting anything

Please let me know if you need any more information to investigate this fully.

Thanks!

D

[hmolesworth]

The 16MHz Timer may be using the RC high-frequency clock instead of the 32MHz crystal, as it is set separately to the 32kHz crystal; try this:

  // Start HFCLK from crystal oscillator, this will give the Timer higher accuracy
  NRF_CLOCK->TASKS_HFCLKSTART = 1;
  while (NRF_CLOCK->EVENTS_HFCLKSTARTED == 0);
  NRF_CLOCK->EVENTS_HFCLKSTARTED = 0;

There is an nrf library call to do this as well, though I forget offhand what that is.

[davege]

Hi hmolesworth,

thanks!

OK, this is something I already tried originally and it didn't work.

As a matter of fact, I then tried to run the same application on another dev kit (still PCA10040)...and with your suggestion things magically start to work as expected! I now have to prove this on my application...

Anyway I now start thinking that it could be a problem of this particular board - is there anyone that could suggest why I could see something like this? At pure HW level, can the XTAL malfunction is such a subtle way?

This would be a good lesson learned to detect similar problems in the future.

Thanks,

D