Numerical Methods of Noise Reduction for Frequency Domain SEM

This paper addresses the numerical problem of performing a pole residue expansion on noisy data. A method of combining dissimilar data sets to achieve an effective increase in signal-to-noise ratio is tested on computer generated data simulating scattering from a sphere and from a thin wire. The data is corrupted with white Gaussian noise and the algorithm is tested for signal-to-noise ratios ranging from -10 dB to 60 dB. An iterative version of the algorithm is also tested. Finally, the problem of filtering noise from relatively quiet data is discussed and a novel filtering algorithm is presented.