Is there a way to scan through a signal and remove the parts that don't look "normal". This would mean using more advanced then a simple pass band filter, but something that can look at a period in the signal and remove it if that period didn't fit into a "template".
4
-
$\begingroup$ What do the red circles represent? If they are not based on a-priori knowledge, it seems they could be useful to determine the positions of the noise parts... $\endgroup$– applesoupMar 1, 2019 at 14:42
-
$\begingroup$ Those are at the start of each rising signal. If I could figure out how to remove signals that didn't fit into a specific "mold" then I would have a solution, but I don't know if that's possible. $\endgroup$– sgmmMar 1, 2019 at 14:52
-
$\begingroup$ Do you have some sort of mathematical model for the "useful" signal (the one you want to preserve) and/or the noise? Think about their properties, and which of them we can use to distinguish the two. We can see that the noise has a much higher amplitude, for example. We can also see that the noise and signal overlap in time, but only at a limited number of samples. You could look at the frequency domain, and see whether or not signal and noise overlap there. As a simple rule: the more "overlap" there is, and the less we know about the form of the signals, the harder it is to remove the noise. $\endgroup$– mateCMar 1, 2019 at 15:11
-
$\begingroup$ To add a more practical suggestion: if the amount of samples affected by noise is small, it might be enough to detect those samples (e.g., by comparing the energy in a local window to the mean energy of the signal) and simply discard them. $\endgroup$– mateCMar 1, 2019 at 15:14
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
2
The short answer is no, unless you know exactly what the noise is, you can not in any practical sense remove the noise. There are things you can do to maximize SNR but no technique exists to remove all noise.
-
$\begingroup$ Isn't there a way to analyze the signal and look for sharp peaks and irregularities? I know you can't remove ALL noise but can't we remove the noise that is clearly visible? $\endgroup$– sgmmMar 1, 2019 at 14:39
-
$\begingroup$ @CrystalPritzker You can do it, but first you need to translate from the English "clearly visible" to a computer algorithm. That's the hard part. $\endgroup$– MBazMar 1, 2019 at 15:14
$\begingroup$
$\endgroup$
You have to analyse your signal and noise spectral distribution. If by chance your signal and noise are not distributed over the same frequencies, you can design a filter ( high pass, low pass or bandpass) depending on those frequencies. If noise occupies the same frequencies than your signal, it will be complicated. If you have a periodic signal, you can always average on multiple acquisitions which will enhance the SNR. In your case, it can be noticed that your noise is pulsed. You can blank some sequences of your signal corresponding to high signal power level (threshold to be defined) in order to remove some noise.
