Fault tolerant adaptive filters based on the block LMS algorithm

The author propose combining residue number system (RNS) arithmetic with a block least-mean-square (BLMS) adaptive algorithm to provide fault tolerance in high speed VLSI adaptive filters. An RNS architecture with double modular redundancy provides an efficient high speed network structure that is well suited for VLSI implementation with fault protection. It is shown that the BLMS algorithm is compatible with an RNS implementation because most of the computation is easily done with integer arithmetic, while requiring the more complicated operation of scaling only one time for each block of input data. The combination of the BLMS adaptive algorithm and an RNS implementation with double modular redundancy provides an excellent hybrid design for a VLSI adaptive filter which must perform reliably in real-time environments