Stable adaptive filter and method

Abstract

A Fast Affine Projection (FAP) adaptive filter and method of adaptive filtering are disclosed, which reduce instability associated with FAP filters caused by error accumulation in the process of inversion of an autocorrelation matrix. The method provides updating of the adaptive filter coefficients by solving at least one system of linear equations whose coefficients are the autocorrelation matrix coefficients, by using a descending iterative method with intrinsic feedback. The results of the solution are used to update the adaptive filter coefficients. The approach is applicable for a normalized step size ranging from zero to unity, and allows either direct determination of updated filter coefficients without determining an inverse autocorrelation matrix, or, determining the inverse autocorrelation matrix by a descending iterative method. In some embodiments, a normalized step size is set close to unity, and the system of linear equations is solved by steepest descent or conjugate gradients methods. In other embodiments, a normalized step size is substantially less than unity, e.g. less than about 0.7. Accumulation of inevitable numerical errors is avoided and the stable adaptive filter and method are suitable for various DSP platforms, e.g. 16 and 24 bit, fixed-point and floating-point platforms.

Dependencies are inferred from the claim text.