Correlation matrix estimation by an optimally controlled recursive average method and its application to blind source separation

Abstract

In this paper we describe a fast and precise method of estimating a correlation matrix and its application. The estimation of correlation matrices is widely used in array signal processing. The estimation is commonly carried out by averaging input signals using a fixed-length time window. To achieve high performance, the window length should be set at the optimum value depending on the acoustical environment, such as the signal-to-noise ratio. However, in dynamically changing environments it is difficult to set a fixed window length because the optimum value also changes dynamically. To solve this problem, we propose an optimally controlled recursive average (OCRA) method that can control the window length adaptively. To evaluate our OCRA method, we applied it to geometric source separation (GSS), which is a sound source separation method suitable for real-time systems. Experimental results showed that the proposed method improved sound source separation.