Obtaining high resolution surface analytical spectra can be complicated by the degrading
inclusion of noise due to instrumental effects, i.e. detector noise, or even sample
condition, i.e. surface roughness. Over the years, many different mathematical techniques
have been developed to deal with the removal of convolved noise in signals. These techniques
can be divided into three main classes: the curve fitting methods; the digital filters; and
newer digital signal processing techniques for example wavelet denoising.
To determine the effectiveness of these different methods, simulated noisy spectra, where the
added noise characteristics are known, were denoised. The change in the signal-to-noise
ratios for the smoothed spectra, and also the resultant change in peak areas were used as
benchmark parameters. As a practical demonstration, the methods were applied to some Auger electron spectra, but are also applicable to other spectroscopies.
The benchmark shows that novel techniques like wavelet thresholding can be used effectively to remove noise, even though artifacts can be formed if the threshold conditions are not chosen correctly. Simple digital filters like the Savitzky-Golay polynomials perform just as well provided that the conditions for optimal smoothing are met.
|