Semi-blind Technique for Non-stationary Noise Cancellation in System Identification

Abstract

This paper proposes a method for signal recovery from noisy observation in system identification. From an ICA (Independent Component Analysis) point of view, this method estimates the noise model by resorting the problem to what is called semi-blind deconvolution, and then cancels the noise effect, thereby enabling us to identify the system more accurately. An adaptive (on-line) semi-blind method is described with focus upon the cost function. Numerical example for a spring-mass-damper system illustrates the result.