DEVELOPER PORTAL
Knowledge Base
HOME » KNOWLEDGE BASE » FORUM

## SX1276 FRF Register Frequency Step

Hi,

I just wanted to clarify the frequency step for the FRF registers.

Is Fstep definitely 61.03515625?

On page 109 of rev 6 of the SX1276 datasheet in the RegFrfLsb description it says "Resolution is 61.035 Hz".

On page 36 in the Frequency Settings section it says "Fstep = Fosc/2^19". Therefore Fstep = 32000000/2^19 = 61.03515625.

I'm guessing that 61.03515625 is the correct value and the description on page 109 _is just a rounding down_! The calculation on page 36 seems to make sense.

From the LoRa-net drivers "#define FREQ_STEP 61.03515625".

The background to my question is that I started logging the value of the frequency error registers and found that Rx frequency error on the 4 devices I tried was consistently high. If I use an Fstep of 61.035 (rounded down to the nearest value representable in single-precision) the error becomes less high. In other words, as far as I can tell 61.035 works better than 61.03515625.

Any thoughts greatly appreciated,
Matt
Hi Matt,

This is just a rounding, so the formula given is the right answer. By accident the rounding gives a better result on that board, because your reference frequency (32 MHz), must be a couple of ppm off

Rgds
> By accident the rounding gives a better result on that board,
> because your reference frequency (32 MHz), must be a couple of ppm off

Yes, agreed.

I've looked at the indicated frequency error on a few (3 or 4) of our our boards and the FEI looks similar. I'm seeing frequency errors between 10 kHz and 20 kHz at room temperature, roughly speaking.

I guess _errors_, i.e. component tolerances, in our board design account for the similar behaviour of different boards.

I think this is a related question...

Unsurprisingly the frequency error increases if I decrease the temperature.

Currently _image calibration_ is only performed when the device boots, _auto image cal_ is off.

Should I be performing _image calibration_ more often?

Our device is a temperature sensor and could be moved from room temperature to a freezer.

Thanks,
Matt
Hello,

A hand-waving estimation would be about 10dB degradation on the image attenuation across the full temperature range. Depending on energy constraints and ability to measure temp, you may add a runtime calibration (is it a metal box freezer, a fridge? If so, the probably of having high interferer in the image channel may nnot be that high?)

Rgds
> is it a metal box

Yes, that's a common use case.

Another common use case is having many devices in close proximity.

> If so, the probably of having high interferer in the image channel may nnot be that high?

I don't understand the technical details of your comment which is probably because I don't understand what the _image channel_ is. Perhaps I should read the documentation more closely??

But I think you're saying that, there are many other factors that are likely to be more dominant and performing the image channel calibration is unlikely to have much effect.

Thanks,
Matt
> A hand-waving estimation would be about 10dB degradation on the image attenuation
> across the full temperature range.

What does that specifically mean in terms of receive sensitivty ?

If you carry out image calibratuion at startup, and room temperature, is it the case that you can loose significant amounts of receive sensitivity if the receiver then goes to very low temperature, even if it stays within the normal 25% of bandwidth capture range ?
hello,

this has nothing to do with sensitivity (in the absence of interference at the image frequency, of course)

Sebastien