Does the ppk2 average the high sampling rate when using its low sampling rate?

[douglas.malnati]

Hi,

There is another thread with this exact question, but I don't understand the answer, and the thread is locked.

This is the PPK2 measuring at 1 sample per second.

I start up a high-consumption device, which starts basically instantly, and the graph shows multiple seconds of gradual increase.  You can see the individual data points, which should correspond to 1 data point per second.

Here is another shot, sampling at 10 samples per second, of the same event.  Now I can see 6 samples leading to peak consumption, which corresponds to 600ms, far less duration than the 1-sample-per-second graph above, for the same event.

This makes me believe the PPK is doing some rolling window averaging of values.  Is that how it works?

I know the device is capable of doing 100,000 samples/sec.  Does it do that all the time regardless of configuration, and feed that data into a rolling average?

How wide is the window of the rolling average?  Does it depend on the configuration settings?

And is that window centered on the data point to be drawn?  For example, in the first graph, it took multiple seconds of wall clock time for the graph to finally reach its peak value, despite the actual electronics changing consumption nearly instantly.  Makes me think there is a multi-second-wide window which is waiting for future data to arrive to calculate "now's" value.

Please if you can explain how this works.  Please let me know if my comments and terminology don't make sense I can try to be clearer. 

If there is a technical document describing the implementation that would be helpful too of course.

Thanks.

[Simonr]

Hi

Sorry about the late reply, I got an update from the devs on this yesterday, and we're using Simple Moving Average to calculate the samples on the fly in the Power Profiler: https://github.com/NordicSemiconductor/pc-nrfconnect-ppk/blob/7d175d9da2a43582de82ce370e75b65ea4236d6b/src/actions/deviceActions.ts#L274-L291 

For 1 sample/sec for example, we use Simple Moving Average on 100 000 samples before returning the averaged values, for 10 samples/sec we do it on 10 000 samples, and so on. So the difference could indeed be quite big.

Best regards,

Simon

[douglas.malnati]

Hi Simon, thanks for looking into that!

Not sure I understand how the described approach leads to the first graph I showed though.

Here, at 1 sample per second, I see it takes more than 3 seconds for my current measurement to rise from ~8mA to ~25mA.

However, I know, from a high-speed measurement, that the rise time is less than 1 second.

So that means, in the 1-sample-per-second graph:

• the first plotted rise point used some of the rise in its average
• the second plotted rise point used some of the rise in its average
• the third plotted rise point used some of the rise in its average

How can that be?

If each point takes into consideration 100,000 samples, and the device takes 100,000 samples per second, then each of the plotted points  couldn't have seen the rise in their window.

Or do the 1-sample-per-second plot points use overlapping data from one another?

And did you find out about how the rolling window is centered?

Thanks.

Doug

[Simonr]

Hi Doug

Just wanted to let you know that I've forwarded this question internally as well, but we're now in the Easter holiday here in Norway, so we're short on staff and delayed replies must be expected. I can't guarantee any updates on this before Tuesday April 11th. Thank you for your patience!

Best regards,

Simon

[douglas.malnati]

Ok thanks, happy holidays!