Accuracy of the 32.768kHz XTAL oscillator over temperature

[Peter]

Hello,

as I understand the accuracy of the LF Crystal oscillator for the nRF52832 development board DK_PCA10040 ist set within the sdk_config.h in the following way:

#define NRF_SDH_CLOCK_LF_SRC 1

#define NRF_SDH_CLOCK_LF_RC_CTIV 0

#define NRF_SDH_CLOCK_LF_RC_TEMP_CTIV 0

#define NRF_SDH_CLOCK_LF_ACCURACY 7   // => NRF_CLOCK_LF_ACCURACY_20_PPM

An Epson FC-135 crystal with an initial frequency tolerance of +/-20ppm @25°C is used.

My question is this:

is the significant frequency drift over temperature taken into account anyway (e.g. internal correction by temperature measuerent ) or must it be considered within the NRF_SDH_CLOCK_LF_ACCURACY define depending on the real application temperature? For example an ambient temperature of +70°C will lead to a frequency error of -81ppm in addition to the +/-20ppm initial tolerance and the aging.

Best regards

Peter   

[AmbystomaLabs]

You did not specify the full part number for your crystal.  Epson makes many versions of that product with different tolerance and temp stability specs.

Crystals are always tolerance plus stability plus aging plus loading/warping.  Capacitive loading can move the nominal frequency, but if you followed the reference design your nominal is probably close to spec. Assuming yours is +/-20ppm tolerance and +/-20ppm stability then the final error is +/-40ppm.

There is no temp comp in the nRF.  The field associated with NRF_CLOCK_LF_ACCURACY is just you telling the SD what the error of your solution is so it can adjust its internal timing to compensate.

The BLE spec allows for up to +/-250ppm for the LF clock.

[Andreas]

Hi,

 

As AmbystomaLabs says, frequency drift due to temperature is not taken into account in the NRF_CLOCK_LF_ACCURACY. If you expect large temerature deviation, you might want to adjust the loading, the drift is indicated by the 'Parabolic coefficient' parameter in the datasheet.

 

Best regards,

Andreas

[Peter]

Yes, that is also my understandig. A tuning fork crystal like the FC 135 type will always have a temperature drift in the range of (0,4*(T-T25)^2) other than AT cut crystals. So we will set the NRF_CLOCK_LF_ACCURACY to the worst case value (<250ppm) depending on our max/min application tempearture.

Thank you, 

Peter