Adaptive noise cancelling by accelerated gradients techniques

Demonstrates the application of the method of steepest descent with exact linear search (SDELS) and the conjugate gradients method (CGM) to adaptive noise cancelling (ANC). An adaptive algorithm based on the method of SEDLS is proposed and its computational complexity and numerical stability are discussed. The performance of the algorithms in terms of convergence speed, tracking data nonstationarities, and the effect of fixed-point roundoff errors is illustrated by simulating a single channel ANC scheme using the standard recursive least-squares algorithm and the proposed algorithm. These results illustrate the close performance of both algorithms and verify the lower numerical sensitivity of the proposed algorithm to quantization errors in a finite-precision implementation. Furthermore, the authors demonstrate how block data adaptation and parallel-processing techniques facilitate the adaptive implementation of the CGM (for stationary data).<>