Iterative detection for partial response magnetic recording channels-a graphical view

In this paper, we compare detection algorithms for additive white Gaussian noise (AWGN) channels and correlated/signal-dependent noise channels for a serially concatenated recording system from the viewpoint of probability propagation. It has been proven that the classical turbo decoding algorithm is an instance of Pearl's belief propagation algorithm, and the decoding algorithm for serially concatenated convolutional codes also can be derived from a belief propagation viewpoint. This result is also valid for correlated/signal-dependent noise channels, such as magnetic recording channels. Using the relationship of probability dependency, Bayesian networks and their corresponding probability propagation schemes for a rate-16/17 serially concatenated recording system are obtained for both AWGN and correlated/signal-dependent noise channels. Noise predictive turbo systems (NPTS) are also discussed from a graphical perspective.