ON THE ASYMMETRY IN THE PSD FOR 2-D LINEAR PREDICTION

The skewness in the peaks of the two-dimensional power spectral densit y (PSD) using the quadrant regions of support is investigated, In prac tice, limited data records or deviation from the assumed white noise h ypothesis causes non-ideal approximations in the autoregressive proces s. If only a limited number of data samples is available, the correlat ion matrix is an estimate of the true correlation matrix. The autoregr essive process is also based on additive white Gaussian noise, therefo re any variation in noise will cause an approximation. This deviation from ideal is demonstrated to be the cause of the skewness in the powe r spectral density. (C) 1997 Elsevier Science B.V.