Extension of the signal-subspace projection method to multi-dimension using CCA

The traditional signal-subspace projection (SSP) method combats the problem of array manifold uncertainty to gain the robustness by means of projecting the nominal manifold vector onto the signal subspace so as to eliminate the errors lying in the noise subspace. The main contribution of this paper is to extent the SSP approach from one dimension to multi-dimension. We assume that the actual manifold vector of the desired signal can be expressed as a product of a known matrix and an unknown coordinate vector. Then it is shown that the SSP method can be derived from the perspective of a problem of canonical correlation analysis (CCA) where the dimension of one subspace is one. When the dimension of the subspace (which the actual manifold of the desired signal belongs to) increases to multi-dimension, a novel projection method is developed, which can be viewed as the extension of the SSP method from one dimension to multi-dimension. Numerical results demonstrate the superiority of the proposed beamformer relatively to the conventional SSP method.