Comparing efficient broadband beamforming architectures and their performance trade-offs

In this paper, we evaluate efficient implementations of a broadband beamforming structure, that permits to project the data onto subspaces defined by the principle components of the array data. This optimum but computationally expensive approach is approximated in the frequency domain by processing in independent frequency bins. The latter is computationally optimal, but suffers from spectral leakage. We show that this problem persists even if the frequency resolution is increased, and that the worst case performance depends on the available degrees for freedoms only. Further, an oversampled subband scheme is proposed, which sacrifices some computational complexity but has a considerably improved and controllable worst case performance.