What does confidence mean in DOA_value.html?

I see in 12. Appendix · krakenrf/krakensdr_docs Wiki · GitHub the

3. Confidence Value : (0-99 float, higher indicates a better quality reading),

My question is what does confidence mean? How is it calculated?
Why in the output format for DF Aggregator (xml), is this value greater than 200?
Thank you.

You can see how confidence is calculated in the function calculate_doa_papr in krakensdr_doa/krakenSDR_signal_processor.py at main · krakenrf/krakensdr_doa · GitHub

It’s just a metric looking at the max DOA power divided by the mean of the DOA power. Higher and shaper DOA peaks will give a higher confidence value.

It was used in the older KerberosSDR app to discard low confidence readings, but it’s not really needed in the new Kraken app.

1 Like

How can one discard low-confidence readings now? You mentioned it is not needed in the new Kraken app. Could you please explain how the discarding works in the application? I was considering discarding signals by the confidence value myself, but is there a better way?

The same settings for confidence and power filtering are still there in the app.

But I found that discarding low confidence values has almost no effect on the final result. Especially when you use the new way the grid is calculated in the Kraken app, which uses the full 360 DOA array. In the new algorithm, bad DOA readings naturally contribute less to the result anyway, so no need to do discarding.

@krakenrf_carl if one would like to use confidence value for signal filtering to display only real ones, but not discard weak signals or signals behind natural obstacles by accident, what would be the good threshold value for no signal/signal? I made some measurements, and the “no signal” confidence value was around 0.8-2.2, and the “signal” - was around 6.0-11.0

Weak signals and low confidence signals are likely to be very correlated.

The threshold is hard to say as it depends on the particular data you got, which is another reason why the confidence value isn’t really a good solution. It’s probably most useful in post processing so you will need to look into the data and see what confidence values you got, and determine some sort of threshold that way.

1 Like