| The magnetic recording channel is a communication channel where extremely low bit-error-rate detection (reading) is required. As a communication channel, the magnetic recording channel suffers from long intersymbol interference (ISI) and a special form of channel noise known as the medium noise. Unlike the electronics noise, the statistical characteristics of this medium noise depend on the transmitted symbols (signal-dependent noise). Two major topics of the noisy channel detection problem, the hard-decision detection and the soft-decision detection, are studied in this thesis for the signal-dependent noise ISI channels (hard-decision detection) and the general ISI channels (soft-decision detection).; For the hard-decision detection, the well-known Viterbi algorithm is optimal for the channels with finite ISI length and additive white Gaussian noise. My research in this thesis focuses on the maximum likelihood sequence detection (MLSD) under the signal-dependent noise described above. We derive an (asymptotically) optimal MLSD by using the noise prediction approach, where the predictors are also dependent on the local bit pattern (pattern-dependent noise prediction; PDNP). This PDNP technique can be applied to many other detectors such as the BCJR algorithm (also known as optimal MAP algorithm) as well as the MLSD achieving 1∼3 dB SNR gain depending on the channel. Complexity-reduction techniques for the PDNP will also be studied.; For the soft-decision detection, we derive an alternative approach toward optimal soft-output detection. While the BCJR algorithm is optimal, our new approach which we call a new MAP algorithm is also optimal and has the benefits of simpler detector structure, easy derivation, less complexity, and significant saving in memory. To make the new MAP algorithm more practical, sub-optimal finite-window mode MAP algorithm will also be developed. In the finite-window mode, the new MAP algorithm can generate soft-decisions from the channel outputs within a fixed amount of decision delay. Other problems such as large amount of memory usage and long decision delay are also cleared in the finite-window mode MAP detection. To verify the quality of the soft-decisions, an outer low-density parity check code is used with the massage passing decoding algorithm. |
