Cumulant-Based 2-D Direction Estimation Using an Acoustic Vector Sensor Array

Abstract

This paper proposes a new acoustic vector sensor array geometry containing a sparse right triangular subarray with three guiding vector sensors plus an additional subarray with arbitrarily placed vector sensors at unknown locations and develops a new cumulant-based algorithm for two-dimensional direction estimation (CADE) of multiple non-Gaussian signals. In the CADE algorithm, three cumulant matrices are defined to form a third-order tensor, using which the direction cosines of the signal are recovered by utilizing parallel factor (PARAFAC) fitting. Furthermore, the PARAFAC model is extended to multiple cumulant matrices and a fourth-order tensor framework to enhance the identifiability issues and to improve the performance of the algorithm. Unlike most of the existing acoustic vector sensor direction estimation algorithms, the proposed algorithm imposes less geometric constraint on the array shape and requires no open-form two-dimensional searching and parameter pairing. Simulation results demonstrate the superiority of the CADE algorithm.