2
$\begingroup$

I am trying to detect occupied bandwidth by looking at the spectrum of the signal in environment. The scenario is mostly blind and the dominant noise is the white Gaussian noise of the environment. Although, I don't have any knowledge of the incoming signals and the channel between us could be something complex. What I do is capture the signal using a SDR (it delivers to me the I/Q of the signal) and then I estimate the spectrum of the signal using Welch method. I start sweeping the signal data and use Hamming window and 8 times averaging (the windows have 75% overlap) and the FFT length is 1024. So far I found out that if occupied bands are narrowband, the moving median as an outlier detector can be a very good indicator of the noise level and using moving median and subtracting the spectrum from the median I can detect where a signal exist. however, when the present signal is wideband and have a relatively smooth spectrum, I can't find it! Cause the median tracks it very well and the subtraction yields nothing. For example the WiFi signal that I have captured below Captured WiFi in Spectrum and some single tone in it's vicinity

I can find the single tone adjacent to it but I can detect the WiFi itself. Is there a better noise floor estimator that allow me to find smooth signals like WiFi? Should I change my method, instead of floor estimation, look for something else? Is there an optimal noise floor estimation method?

Improve this question
$\endgroup$
1
  • $\begingroup$ Based on the plot, consider using the 10% quantile of the spectrum bins as the noise floor estimate. $\endgroup$ Commented Aug 23 at 2:25

1 Answer 1

Reset to default
1
$\begingroup$

Your observations are correct. Essentially, you're trying to build a spectral energy detector over a wide bandwidth. The median filter works well for narrowband signals that only occupy a few spectral bins, likely shorter than your median filter length. But, of course, those OFDM WIFI channels occupy many more spectral bins than the length of your median filter, so your median filter will only produce output that is the value of one of the WIFI spectral bins rather than any noise floor bin.

You're going to run into a couple challenges. Are these WIFI channels highly active? e.g. Are they turning on and off quickly? Or are they used sparsely, with a lot of dead-air time between bursts? If sparse, then you have a shot of measuring the noise floor then, but it will still be difficult and error prone.

One way to do it would be a heuristic combination of moving median, moving min, and moving max filters over the FFT frames. Aggregating the output of these filters can help you derive a noise floor estimate based on the output of these filters. You can set a dB threshold above this floor where, if the magnitude of each bins is greater than the threshold, you'd consider there to be a signal there. Finally, apply a leaky integrator filter to this threshold over time, which will make it resistant to short, powerful bursts, like WIFI bursts. You now have a decent noise floor estimate from FFT frames. Note that is this also dependent on $NFFT$ and thus your FFT frame time. Make those frames shorter or longer based on detection needs.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Thank you for the guide, but I was curious, there is no optimal solution (like optimal filtering approach) for such scenario. am I right? Or can we describe this problem in a constrained type optimization scenario or something and find the solution, like what happens with matched filtering and other similar problems? Is there such problem or reference out there solved? $\endgroup$ Commented Aug 22 at 6:28
  • $\begingroup$ "Optimal" for the median filter, or min/max filter, will depend on your NFFT, input sampling rate, and other features. in practice, the order of those filters generally is 13 or less. You'll have to tailor this filter length proportionally to the ratio between NFFT and the approximate bandwidth of signals you are trying to detect. The Wifi signals in your spectrum are "wide", so a smaller NFFT is usable. Compare that to narrowband signals, and you'll need a higher NFFT, which will require tweaking the filters as well. it's all a tradeoff -- no universal solution. $\endgroup$ Commented Aug 27 at 1:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.