Fourier-ratio deconvolution before EELS signal quantifications?

[rotefuchs3]

I understand based on exspy documentation: "The core-loss should have the background removed before Fourier-ratio deconvolution." So, shall I remove background before each element on core-loss data, then do deconvolution for each element in a specific energy range (perhaps same to the edge fitting range)? When I do this, the Fourier-ratio deconvolution have issue either:

1. when extrapolate coreloss=False, Deconvolution produces: mean=nan, all NaN values; is that because without extrapolation, deconvolution can not handle edge regions?
2. when extrapotate coreloss=True, Deconvolution fails with power-law errors: No inflexion point could be found in some positions that have been marked with nans.
   WARNING | Hyperspy | Power-law parameter estimation failed because of a "divide-by-zero" error. (hyperspy._components. power_law:166); cannot unpack non-iterable bool object

My goal is to achieve similar Fourier-ratio deconvolution function in DM that for each element in core-loss data the background was firstly fitted then signal being deconvoluted, and then signal being fitted for quantification.

[ericpre]

It seems that some computation errors occur when extrapolating the edge. Would you be able to share a minimal reproducible example or the data?

An alternative to deconvolution is to do model fitting with low-loss convolution, which should give a better signal-to-noise because the deconvolution step will increase the noise. The disadvantage of the model fitting approach is that it will be slower.