Block implementation of fault-tolerant LMS adaptive FIR filters

Abstract

Adaptive FIR filters are widely used in a variety of modern digital signal processing application, and with the advancement in VLSI technology, it is now feasible to build fairly complicated multiprocessor systems which can provide the necessary computational power required by some highly demanding real-time signal processing applications. The increased computational power of such systems also makes fault tolerance an even more important issue that needs to be addressed carefully, because a single hardware failure can easily render the whole compuational results useless. Algorithm-based fault tolerance (ABFT), recently developed as an effective high level fault-tolerant technique, employs, in addition to the normal outputs, parity numbers that are related to the outputs. These parities can be used to concurrently detect, and in some cases correct, errors caused by hardware failures. A highly efficient implementation of the encoding scheme based on the weighted checksum code is proposed. It is particularly suitable for block adaptive signal processing, and the computational efficiency of this method is compared with that of the more general weighted checksum code.