Removing isolated high amplitude noise from array

I have an array of sound speed, SS, with noise in it. The noise can be seen as a blurred patch between -0.04 and -0.07 on the y axis, across the whole profile (Additionally there is a thinner band at ~ -0.1 s).

I would like to remove the noise without affecting signal.

The noise appears quite clearly in second derivatives (called FXX and FYY in code below).

I have tried several noise removal methods, with limited success - not helped by a lack of signal processing knowledge. Currently I am using the second derivative with respect to x (FXX, distance, km) to simply replace values over a threshold with an average.

[FX,FY] = gradient(SS);
[FXy,FYY] = gradient(FY);
[FXX,FYx] = gradient(FX);
for i=1:nx
    for j=1:ny2
        if abs(FXX(j,i))>3;
            SS(j,i)=average_SS_val_for_noisy_region;
        end
    end
end

This is not working very well as you can see below

Could someone please help me sort out an effective way to filter out the noise?

File attached for better viewing

Answer 1

[Star Strider]

If your figures are a spectrogram with frequency on the y-axis, you can easily use a lowpass or bandpass filter to pass everything below the noise frequencies, or a bandstop filter to specifically remove the noise band.

[K G]

The figures above are not spectograms, simply sound speed as a function of TWTT (essentially depth) and distance along the profile. Would it be possible to convert them into a spectogram domain?

[Star Strider]

I would convert them to the frequency domain with the fft (link) function to see if the noise is band-limited. If it is, you can use a frequency-selective filter to eliminate it.

Your other option is a wavelet transform to identify the noise location in the signal and eliminate it. Wavelets are good for de-noising signals, and there are examples in the Wavelet Toolbox on how to do this. (I do not have enough experience with wavelets at present to help with that.)