Fault-tolerant FIR adaptive filter using the conjugate gradient algorithm

In several recent papers we have introduced the notion of Adaptive Fault Tolerance (AFT), where the adaptive process is used to automatically compensate for faults occurring in the adaptive coefficients. Several adaptive filtering structures have been proposed which use AFT to provide fault tolerance with very low hardware overhead, while maintaining reasonable convergence behavior for certain classes of input signals. The previous structures were adapted with the LMS algorithm which is known to have poor convergence properties when the input signal is highly colored. In this paper we propose using the Conjugate Gradient (CG) algorithm to update the filter coefficients. The CG algorithm, which allows rapid convergence of the fault tolerant adaptive filter regardless of the input noise statistics, is known not to suffer from the same performance drawbacks as the LMS algorithm. Simulations will be presented to demonstrate the performance of the CG algorithm and comparisons with previous work will be made. Implementation issues will also be discussed.