Delayless signal smoothing using a median and predictive filter hybrid
P. Taneli Harju, S.J. Ovaska, Vesa Välimäki · 2002
Median filters are widely applied in restoring signals that are corrupted by impulsive noise. Using a median filter of length N, impulses of length up to (N-1)/2 samples are filtered away from the signal, but at the same time the signal is delayed by (N-1)/2 samples. Furthermore, median filtering may distort the signal waveform. We present a novel hybrid scheme of using linear predictive filters in cascade with a median filter to compensate the delay caused by median filtering. In addition to canceling the delay, the proposed system relieves signal distortion due to the low-pass characteristics of the polynomial predictive filters used. The scheme is illustrated by an application in modeling of musical instruments where the proposed filter system successfully removes transient disturbances.