Comparative performance of forward/backward and symmetric linear predictions

Three linear prediction schemes are studied based on different combinations of the forward and backward estimates. One of the schemes is shown to be a two-sided prediction method with symmetric weights, which has been widely used in the detection of wideband data embedded in narrowband interference and noise. To make performance comparison possible, the variances of data estimation errors are analysed. According to the analytical results, the symmetric estimation outperforms the forward/backward estimation when the number of prediction weights is small.