Comparison of Least-squares and Eigenfilter Techniques for Broadband Beamforming

In this paper two eigenfilter techniques are described for designing far-field broadband beamformers with an arbitrary spatial directivity pattern. For both techniques the resulting filter is the generalised eigenvector corresponding to the minimum generalised eigenvalue of two real, symmetric and positive definite matrices. In the conventional eigenfilter technique a reference point is needed, while in the eigenfilter technique based on a TLS (Total Least Squares) error criterion, this reference point is not needed. It is shown that linear constraints are easily incorporated into the design procedure. Both eigenfilter techniques are compared to the weighted least-squares filter design technique.