On the asymmetry in the PSD for 2-D linear prediction

The skewness in the peaks of the two-dimensional power spectral density (PSD) using the quadrant regions of support is investigated, In practice, limited data records or deviation from the assumed white noise hypothesis causes non-ideal approximations in the autoregressive process. If only a limited number of data samples is available, the correlation matrix is an estimate of the true correlation matrix. The autoregressive process is also based on additive white Gaussian noise, therefore any variation in noise will cause an approximation. This deviation from ideal is demonstrated to be the cause of the skewness in the power spectral density. (C) 1997 Elsevier Science B.V.