An adaptive NR filter for spectrum analysis

Abstract

An adaptive non-recursive (NR) filter with application to spectrum analysis is presented. The adaptive nature is implemented with a weight adjustment algorithm (LMS) on the filter that performs a steepest descent minimization of mean square error. The difference between an “interference” or coherent representative of it and the raw signal under analysis via its spectral content is defined as the error. A priori knowledge of the disturbances consists of the spectral content of the disturbances without regard to amplitude or phase information. The algorithm is manifest with a digital arithmetic unit using high speed logic in a robust 512 weight digital filter fabricated with current integrated circuit technology. The experimental configuration is described along with improvements currently attainable with advancements in technology. Applications to spectrum analysis, whereby unknown disturbances degrade interpretation of spectrum signatures in bearing studies, acoustic phenomena, and sonar signal analysis, is described. Adaptive filter behavior with stationary as well as non-stationary input signals are presented. It is shown that for some narrowband as well as broadband disturbances the filter tracks and adaptively rejects disturbance in real-time so as to present uncluttered renditions of the spectrum signatures or spectrograms for analysis.