Source separation and localisation via tensor decomposition for distributed arrays

Abstract

This study focuses on the problem of power spectra separation and localisation of multiple sources using distributed arrays. First, the array structure and signal model are discussed. By cross-correlating the multi-channel received signals in time-domain, a third tensor is constructed. Then, utilising the multi-dimensional characteristic, the tensor is decomposed to separate the array manifold matrix and the power spectra matrix through alternating least square (ALS) method. Finally, the sources are located using the relative x-y plane relationship between the distributed arrays and the direction of arrival (DOA), which can be estimated by spectrum analysis of each column of the array manifold matrix. The effectiveness and superiority of the proposed method is demonstrated by simulation results.

Keywords

• array signal processing
• matrix algebra
• direction-of-arrival estimation
• tensors
• spectral analysis
• least squares approximations
• correlation methods
• source separation
• multidimensional signal processing
• tensor decomposition
• distributed arrays
• power spectra separation
• array structure
• signal model
• multichannel received signals
• array manifold matrix
• power spectra matrix
• multidimensional characteristics
• source separation
• source localisation
• cross-correlation
• alternating least square method
• als method
• direction of arrival
• doa
• spectrum analysis