Learning characteristics of adaptive fault tolerant filters in the presence of transient errors

Abstract

Adaptive digital filters can be designed with hardware redundancy so they can continue operating effectively after the occurrence of certain types of hardware failures, thereby achieving reliability through a mechanism known as adaptivefault tolerance. It is known that the extra degrees freedom provided by properly added hardware redundancy provides many equivalent optimal solutions within the coefficient parameter space. This paper investigates the convergence properties of an adaptive fault tolerant FIR filter when temporary noise-induced errors occur.

Fault Tolerant Signal Processing for Nano-scale VLSI Circuit Technology

“…In the single bit error case the filter re-learns through proper coefficient adjustments and the adaptive filter regains its pre-failure performance after experiencing a period of transient response. Figure 3 shows how an AFT FIR filter with one redundant coefficient responds to a transient (soft) coefficient error [9]. It is seen that a soft error temporarily forces the filter coefficients to a new non-converged state, from where the filter begins to re-adapt toward the global minimum error condition.…”

“…Adaptive Fault Tolerance (AFT) takes advantage of non-canonical adaptive filter architectures that use adaptive principles to achieve error masking. It has been shown that AFT methods are capable of masking and recovering from single and multiple fixed [7,8] and transient faults [9]. In this paper we explore the possibility of effectively using AFT principles, residue number system (RNS) arithmetic [10], and stochastic learning algorithms [12,13] to detect and correct soft errors in highly scaled adaptive VLSI processors.…”