how to modify the wmulden function to make it shift-invariant?

Hi everyone!

I am trying to modify the wmulden function to make it shift-invariant. Let me explain better.

When I run wmulden with an array of 30 datasets of 4096 data each I get a particular denoising.

Obviously the data I have available is larger than 4096 so that I can run the moving window for all 4096 data.

But if I run wmulden in moving window and only take into account the last data to reconstruct the same

original matrix I get slightly different data. The difference between batch and online denoising is Gaussian

and this gives many problems because this self-produced Gaussian noise invalidates the denoising.

I analyzed the code and first of all I replaced the SWT flipped with HAAR to the DWT to eliminate the border distortion.

Theresolding heurshure depends only by data size.

Now the PCA and the FASTMCD still remain, both with the calculation of the eigenvector.

Do you have any suggestions for replacing or modifying outlier detection?

Do you know different robust causal denoising methods?

Thanks!