Smoothing of power spectral densities

Power spectral estimates are smoothed using a Kalman filtering approach. The filter is used to segment the power spectral density by separating signal-dominated regions from noise-dominated regions. In doing so, it tends to preserve the fidelity for signal-related spectral peaks while smoothing the segments dominated by the noise. Relative to standard windowing, noise contributions are reduced, while the resolution of an unwindowed spectral estimate is essentially preserved.>