Self-Orthogonalizing Overlap-Save GSC

Abstract

This paper discusses a computationally inexpensive and fast converging approach to broadband beamforming. Exemplarily utilising the generalised sidelobe canceller (GSC), accurate low-cost implementations in the DFT domain based on overlap-save techniques have been previously suggested, which however suffer from slow convergence when used in combination with the least mean squares algorithm. To overcome this limitation, the beamformer proposed here exploits decorrelation of the input signal within the overlap-save architecture. By inclusion of a self-orthogonalizing component into the adaptive algorithm, the eigenvalue spread of the covariance matrix of the input signal is reduced, thereby increasing the adaptation rate. Simulation results indicate that the convergence speed of the proposed beamformer is comparable to a time domain realisation, albeit at a much reduced cost.