Improvements in detectors based upon colored noise

Abstract

Two techniques which improve the performance of partial response maximum likelihood systems (PRML) without increasing the number of states in the Viterbi algorithm (VA) are studied. Both techniques rely on the fact that the noise samples going into the VA are correlated. One technique, known as partial local feedback noise prediction, uses estimates of the past noise samples taken from the path memory of the VA to form a prediction of the current noise sample. The predicted noise is subtracted from the current equalizer output prior to updating the path metrics. In the other technique, the VA operates in a block mode simultaneously on X consecutive equaliser outputs. A branch metric appropriate for colored noise is used within each block of K samples. Simulation results for a Lorentzian channel equalized to a PR4 target show that block processing with 4 samples gives almost the same error probability as noise prediction using a 2-tap predictor. Results also show that the feedback in the noise prediction technique increases the length of error events.