A generalized data discrimination scheme using kernel machines in the presence of pattern-dependent noise

This paper describes an application of the nonlinear support vector machines (SVMs) on magnetic recording channels in order to reduce performance deterioration due to pattern-dependent noise. It has been confirmed that SVMs make simple and direct modification of log-likelihood ratios possible. Consequently, at a bit error rate of 10−4, this method can improve a bit error rate of the pattern-dependent noise by one order in cooperation with proposed iterative decoding.