Iterative processing: From applications to parallel implementations

Over the last ten years, there has been a considerable interest in the area of iterative processing from the application perspective, and more recently from the theoretical perspective as well. Iterative processing indirectly exploits the global structure of a system by decomposing it into simpler local structures, and iteratively exchanging extrinsic soft information between them. This decomposition leads to a linear instead of an exponential increase in complexity.; This thesis presents a new low-complexity iterative algorithm for joint channel estimation and decoding in coded orthogonal frequency-division multiplexing (OFDM) systems. The proposed iterative algorithm exploits all the information available to the receiver. It can be used to blindly estimate the channel within a single OFDM symbol or to significantly improve the performance of semi-blind systems that use a minimum amount of training to estimate and track the channel. We also propose the use of an optional outer low-density parity-check (LDPC) code in serial concatenation to offer a trade-off between latency and performance without affecting the complexity of the core iterative algorithm.; The decomposition associated with iterative processing also facilitates the parallel and distributed implementation of iterative processing algorithms. This thesis investigates the implementation of iterative LDPC decoding on programmable parallel architectures, which is found to be communication-limited. We present general methodologies for optimizing the parallel decoding of LDPC codes on programmable shared-memory and message-passing architectures.