Precoding-Based Blind Separation of MIMO FIR Mixtures

Abstract

This paper focuses on the problem of blind separation of sources mixed by multi-input multi-output finite impulse response channels, which is also called convolutive blind source separation (BSS) in short. This problem has been intensively studied in the context that the sources possess certain favourable properties, such as independence and sparsity. However, these properties may not exist in some practical applications. In this paper, we propose a precoding-based convolutive BSS method, which can deal with mutually correlated sources without requiring the sources to be sparse. It is also applicable to mutually independent sources. In the proposed method, the sources are preprocessed in transmitters prior to transmission by order-one precoders. At the receiving side, the second-order statistics of the sources and the $Z$ -domain features of the precoders are exploited to estimate the coded signals, from which the sources are recovered. Simulation results demonstrate the effectiveness of the new convolutive BSS method.