| Iterative message passing algorithm (iMPA) is a very powerful technique for approximating maximum likelihood decoding in modern communication systems. In this research, we investigate problems related to hardware implementation of iMPAs.;In the first part of this dissertation, iMPA is applied to solve the pseudo-noise acquisition problem in an ultra wideband (UWB) system. We propose a novel iMPA based on a standard graphical model augmented with multiple redundant models. Simulation results show that our new algorithm operates at lower signal to noise ratio than earlier works using similar formulations. We also demonstrate an efficient hardware architecture for implementing the new algorithm. Specifically, the redundant models can be combined together so that substantial memory usage associated with redundancy can be reduced. Our prototype achieves the combination of fast acquisition and low hardware complexity unattainable by traditional approaches.;In the second part, we investigate the effects of manufacturing defects on a typical modern error correcting code decoder based on the iMPA. Specifically, we analyze the performance degradation caused by single stuck at faults on a high speed repeat accumulate code (RA) decoder. Results show that the majority of the single stuck at fault patterns in the data paths cause degradation less than 0.5 dB. We estimate the performance degradation caused by these faults using a combination of EXIT chart and Gaussian SNR evolution. We then construct an error-rate and error-significance acceptance criteria for the faulty chips that can reject 100% of unacceptable faults with a false alarm rate less than 30%. This criterion can be directly applied to generate test vectors that cover 100% of the unacceptable faults. |
