Watchdog parity channels for digital filter protection

Abstract

A novel method is presented for protecting digital filters from both hard and soft errors emanating in their physical realization by using the error-detecting, systematic, real number, convolutional codes. Normal filter operation is augmented by parallel parity channels, and errors in the filter's operation are detected by comparing the parity values with parity samples recalculated from the filters' output. The parity values are calculated in parallel at reduced rate, and no speed degradation is incurred. The rate reduction results because in an (n, k) convolutional code the parity channels operate decimated by a factor of k. It is shown that modifying the parity channels to cancel filter poles leads to a significant reduction in the complexity without sacrificing the detecting power of the convolutional code. Finite-precision arithmetic in the realization introduces noise in error decision variables even in the case of fault-free operation. A statistical analysis is given establishing a bound on the comparators' thresholds above which an error is decided.<>