Implementation of Gold deconvolution for enhanced energy resolution in EEL spectra

Gold’s iterative deconvolution algorithm has been applied to one-dimensional EEL spectra from hexagonal BN. The experimental resolution was varied from 1.1 to 2.25 eV and Gold’s algorithm was able to restore low-loss and core-loss spectra overall well. To estimate the instrument response function, the most convenient method was to extract the zero-loss peak from the low-loss spectrum. By instead using low-loss spectra as kernel, as suggested by Egerton, enhanced energy resolution could also be obtained with plural scattering simultaneously removed. It is further shown how the FWHM of the π⁎ peak in the boron K-edge of hexagonal BN is reduced from 1.4 to 0.7 eV with almost no noise amplification after 500 iterations while resolving the σ⁎ doublet. The result was almost identical after a stunning 5000–10,000 iterations, implying that Gold’s method converges and can be stable for a large number of iterations. However, for lower-intensity spectra the number of iterations is limited. The results close to the intense zero-loss peak were uncertain and further testing with better experimental resolution is recommended. It is also found that to improve the resolution and not just sharpen the spectra, a large number of iterations is required.

Research Highlights
► Gold's method is systematically studied and strengths and weaknesses are highlighted.
► Under favorable conditions, the energy resolution was improved by a factor of ∼1.4.
► Improved energy resolution (not just sharpening) requires a large number of iterations.
► A large number of iterations was performed without significant noise amplification.
► The energy resolution in iteratively deconvolved spectra is not constant everywhere.