Reduced Distortion Wiener Space/Spatial-Frequency Filter

A new design methodology in the development of a Wiener space/spatial-frequency (S/SF) filter is proposed. This solution is based on creation of the filter’s region of support that is symmetrical around the detected local frequencies (LFs) of the estimated signal and is determined, to reach the best results, by auto-terms domains of Wigner distribution of the same signal. To this end, groups of S/SF points, that have equal distance and are symmetrically distributed around a detected LF, are considered one by one (ordered by distance regarding to a detected LF and starting with the group nearest to an LF) and each of these groups is included within the region of support creation if it meets criterion. The criterion, based on the spectral thresholding and related to the concentration of each signal component separately, makes our solution suitable for multicomponent signals, even if their components occupy ranges of frequencies of different widths. The proposition provides, with a very high probability, the filter’s region of support extension up to the spectral threshold-determined position, including this position and its immediate vicinity. In this way, our solution provides efficient estimation of non-stationary two-dimensional signals occupying limited, but not necessarily narrow ranges of frequencies, can seriously reduce (for the appropriate spectral threshold) distortion of the original signal, and hence results in significantly better estimates (42.34% better in the example observed here) in comparison to the state-of-the-art LF estimation-based solutions. Simultaneously, a way of performance evaluation regarding the noise influence and original signal distortion is proposed.