Accurate sample rate with ncs

[RonanB96]

Are there any examples of sampling something like an I2C bus at a constant sample rate? I have seen issues myself with the k_sleep and k_busy_wait not giving accurate delays. Tested with a salae logic analyser. I haven't tested the timings of the k_timer yet.

I have seen this https://github.com/zephyrproject-rtos/zephyr/issues/6498 when looking into the inaccuracy of the delay functions above

[Håkon Alseth]

What are your requirements in terms of constant? What delay (us/ms/s?) are you looking for?

Have you tried running a work queue or a kernel timer instance?

Note that the K_TIMER runs in interrupt context, so taking semaphores shouldn't be done directly in the callback.

 

Kind regards,

Håkon

[RonanB96]

We require a sample rate of 100hz for a Max of 60s. 

Is there issues with the accuracy of the k_timers on the nrf91 similar to the zpehyr issue above?

When I was checking  k_delay and k_busywait with a delay of 1ms, I was measuring a delay between  2-8ms, I was toggling an led and measuring with a logic analyser. There was no other tasks running and this was being called in a loop

[Håkon Alseth]

Hi,

 

I can confirm that k_* timed functions in < 10 ms are very jittery. Even a k_timer instance will not be matching 100 Hz directly (measured to 12 ms at my end).

What I can recommend is to use either a high speed TIMER, or use RTC0 (RTC1 used by the kernel), to generate this event. RTC0 will be able to give 32768/328 Hz = ~99,9 Hz, but would require the use of either nrfx driver or bare metal implementation. If you want to look at selecting the nrfx driver, I have a project here where other nrfx components are manually selected in the application kconfig (required at this time, due to entries having no prompt):

https://github.com/Rallare/fw-nrfconnect-nrf/blob/nrf9160_samples/samples/nrf9160/nrfx_timed_signal/Kconfig

 

Kind regards,

Håkon

[RonanB96]

Hi Hakon,

I tried implementing the timer with RTC0 but it is crashing when I try writing to the Prescaler register within nrfx_rtc_init. 

__STATIC_INLINE void nrf_rtc_prescaler_set(NRF_RTC_Type * p_reg, uint32_t val)
{
    NRFX_ASSERT(val <= (RTC_PRESCALER_PRESCALER_Msk >> RTC_PRESCALER_PRESCALER_Pos));
    p_reg->PRESCALER = val;
}

The timer instance and configuration seem ok

__func__                      "nrfx_rtc_init"    0x00030aec    14    const char[14]                
handler                       0x0002a9f7         r2            4     void(nrfx_rtc_int_type_t)*    
    *                         rtc_handler        0x0002a9f7    2     void(nrfx_rtc_int_type_t)     
p_config                      0x20024be0         r1            6     const nrfx_rtc_config_t*      
    *                         <struct>           0x20024be0    6     struct                        
        prescaler             327                0x20024be0    2     short unsigned int            
        interrupt_priority    7                  0x20024be2    1     unsigned char                 
        tick_latency          65                 0x20024be3    1     unsigned char                 
        reliable              false              0x20024be4    0     _Bool                         
p_instance                    0x00030a88         r0            8     const nrfx_rtc_t*             
    *                         <struct>           0x00030a88    8     struct                        
        p_reg                 0x40014000         0x00030a88    4     NRF_RTC_Type*                 
        irq                   RTC0_IRQn          0x00030a8c    1     enum                          
        instance_id           0x00               0x00030a8d    1     unsigned char                 
        cc_channel_count      0x04               0x00030a8e    1     unsigned char                 

And I made this change in spm to move it to the non-secure domain

diff --git a/subsys/spm/spm.c b/subsys/spm/spm.c
index ec00037..23b7ecf 100644
--- a/subsys/spm/spm.c
+++ b/subsys/spm/spm.c
@@ -287,6 +287,7 @@ static void spm_config_peripherals(void)
        static const struct periph_cfg periph[] = {
                PERIPH("NRF_P0", NRF_P0, CONFIG_SPM_NRF_P0_NS),
                PERIPH("NRF_CLOCK", NRF_CLOCK, CONFIG_SPM_NRF_CLOCK_NS),
+               PERIPH("NRF_RTC0", NRF_RTC0, CONFIG_SPM_NRF_RTC0_NS),
                PERIPH("NRF_RTC1", NRF_RTC1, CONFIG_SPM_NRF_RTC1_NS),
                PERIPH("NRF_NVMC", NRF_NVMC, CONFIG_SPM_NRF_NVMC_NS),
                PERIPH("NRF_UARTE1", NRF_UARTE1, CONFIG_SPM_NRF_UARTE1_NS),

See what the issue is here?

Also I added "CONFIG_COMPILER_OPT="-DNRFX_RTC_ENABLED=1 -DNRFX_RTC0_ENABLED=1"" to get it to compile

[Håkon Alseth]

Hi,

 

I took the rtc sample from nRF5 SDK and did a port of it and put it here:

https://github.com/Rallare/fw-nrfconnect-nrf/tree/nrf9160_samples/samples/nrf9160/nrfx/rtc

 

Could you try it and see if it works at your end?

 

Kind regards,

Håkon

[RonanB96]

Ok got it working now, didn't know I needed to add this to ncs/nrf/subsys/spm/Kconfig as well make the change in spm.c

config SPM_NRF_RTC0_NS
	bool "RTC0 is Non-Secure"
	default y

Is there a reason there is manual changes needed in the SDK to enable RTC0 in the non-secure domain?

[Håkon Alseth]

Glad to hear that you got it working.

RonanB96 said:

Is there a reason there is manual changes needed in the SDK to enable RTC0 in the non-secure domain?

I do not have a good answer here, other than the inherited behavior from Cortex M33 core, as all peripherals reset value is secure mode. Since the modem requires you to be in non-secure mode in order to communicate with it, it is then normal that the application would also reside in this defined area. I'll input this to the developers as a feature request, ie: set most peripherals to non-secure.

 

Kind regards,

Håkon