A new technique to separate the non stationary component of a signal and its application in biomedical engineering

Several algorithms to separate nonstati- onary part of a signal are presented in this work. In the autoregressive model both end prediction (prediction using N/2 pastValues) and midprediction filters are used. If the nonstationary part consists of random impulsive waves of low occurance rate it may be separated as error signal corrupt with 'carried over' errors. Two schemes, signal inversion and time-reversal, are proposed to remove this corruption. Impulses with slowly rising and falling edges are recovered better by the mid-prediction filter because it intro duces a pair of complex conjugate zeros on the frequency axix.