nRF52840 external oscillator

[Nicolas Brunner]

Is it possible to put an external oscillator (e.g. TCXO) instead of a crystal for generating the 32 MHz of the nRF52840 ?

[bjorn-spockeli]

Hi Nicolas, 

I am assuming that you plan on sharing the external oscillator signal with other ICs in your design. Is this the case? Clock sharing is not recommended for the nRF52 series. The receiver is  very sensitive to noise on the reference clock and any noise here will affect the sensitivity of the receiver. Hence, we strongly recommend using a dedicated 32MHz crystal to ensure the performance stated in our product specifications.

Best regards

Bjørn 

[Nicolas Brunner]

I don't want to share the external oscillator. I just want to have a more accurate clock for the nRF52840, this is why i want to put a TCXO instead of a crystal.

[bjorn-spockeli]

The HFXO on the nRF52 is designed for crystal operation only due to phase noise requirements, hence using an external clock signal or a TCXO is not advise by us as we cannot guarantee the performance. You may try it out on your own risk, for all I know it might work, but this has not been characterized or verified by us.

[jjvicentenavarrete]

Has nordic any plan on verifying this? For wide-temperature applications, a mere crystal is not enough.

[bjorn-spockeli]

Hi Juan,

as far as I know there are no plans for developing a reference design with the nRF52 family that uses a TCXO. The nRF52 family has been designed with an XO as the clock source for the radio and crystals with a 10ppm frequency tolerance and 10ppm frequency stability over the nRF52 familiy's recommended operating temperature range (-40 to +85 C) are available, e.g. this and this. This should give a lot of headroom for the +/- 40ppm requirement for BLE. 

Would such a crystal not meet your requirements?

Bjørn 

[Mickael]

Hi Bjorn, 

Ths topic is totally relevant to me too. Because of our application, I also need tighther timings than the BLE requirements at the HFCLK level. 

So what is not clear to me is: what are the reasons why it might not work? 

Surely feeding a more accurate clock will not be a problem - given that it is well selected, I do also believe TCXO can be as good or better than XO for radio purposes. 

So are you worried about how feeding directly a cliiped-sine wave or HCMOS level into the XO circuit might cause issues?  

Thanks for your help,

Mickael

[bjorn-spockeli]

Hi Mickael, 

I am not saying that its not going to work, im just saying that this is not something that we recommend as we have not done any testing or verification with TCXOs as the nRF52 familiy is designed with XOs in mind. 

The clock accuracy is not the issue, I completely agree with you that a more accurate clock is always preferred. However, the external clock source must meet the following requirement with regards to phase noise:

100k offset : -80dBc/Hz 
300k offset : -100dBc/Hz
900k offset : -110dBc/Hz

 

Many of the TCXOs use PLLs which have phase noise above the limits stated above. 

Best regards

Bjørn 

[Mickael]

Hi Bjorn, 

we are going to try and this might be interesting to more people here!

Could you please specifiy if we should feed XC1 or XC2 pin (as in which is the input?) when doing this? 

Thank you,

Mickael

[bjorn-spockeli]

HI Mickael, 

You feed XC1 and XC2 should then, depending on what kind of signal your feeding into XC1, be one of the below:

• If low-swing sine: XC2 pin shall be grounded.

• If rail-to-rail: XC2 can be left open or grounded. Voltage requirement from GND to VDD_NRF.

[Mickael]

very clear! Thank you Bjorn

[abkirchhoff]

Mickael,

How did your attempt go at using a TCXO to feed XC1 on the nRF52840? Did it work reliably? If so, would you be willing to share the part number for the TCXO you used?

Thanks, Austin

[Mickael]

Hi Austin, I am getting boards back next week so will be able to test and give you an answer then. 

Dont hesitate to ping me if I forget.