Fault tolerant adaptive filters based on arithmetic coding techniques

Previous work reported the hybrid combination of a Walsh-Hadamard Transform architecture and Modulus Replication RNS arithmetic to produce a fault tolerant adaptive filter structure (WHT-MRRNS). This paper further analyzes fault tolerant capabilities of the WHT-MRRNS architecture that uses MRRNS coding to provide both RNS (digit) redundancy and PRNS (polynomial) redundancy. Interactions between these two types of redundancy are characterized and strategies are proposed that enable error detection and correction while updating filter coefficients and computing the filter outputs. The proposed fault tolerant architecture relies on MRRNS error correction to eliminate erroneous transient behavior caused by both soft and hard errors.