A Study on a Viterbi Decoding Method with a Noise Whitening Function by Noise Prediction using its Survived Paths

Abstract

This paper describes a method of eliminating the auto correlation of equalized noises to have the optimum performance of a Viterbi decoder designed for white noise. This elimination is performed by subtracting predicted noises from an arriving equalized noise. This predicted noise is generated by using both coefficients calculated from auto correlation of the equalized noises and noises corresponding to survived paths. A correctly predicted noise reduces an arriving equalized noise on average, so the error rate becomes lower. A Viterbi decoder with a noise whitening function reduced the number of errors to about 1/6 of those of a conventional Viterbi decoder's for an experimental digital video cassette recorder. The conditions of the experiments were : relative tap e-to-head speed of 7.5 m/s, recording rate of 40.5 Mb/s, and 8/12 code whose minimum run length was limited to 2 (in NRZL notation) combined with EPR 4, and four coefficients for noise prediction.