Hybrid Convolutional Beamspace for DOA Estimation of Millimeter Wave Sources

Sensor arrays designed for millimeter waves (mmWaves) have gained popularity recently for their potential to offer more bandwidth. To reduce hardware cost, mmWave processing is divided into analog and digital domains. The analog combiner serves as a beamspace processor, reducing the number of required RF chains. Convolutional beamspace (CBS) is a recently proposed beamspace method. It achieves lower computational complexity, higher DOA resolution, and smaller MSE for correlated sources. In this paper, hybrid analog and digital CBS is proposed for DOA estimation of mmWave sources using a receiver array. Constant-modulus constraints imposed on the analog combiner are tackled by the fact that any complex vector is a linear combination of two vectors with unit-modulus entries. The required number of RF chains equals the dimension of the CBS output after decimation. Besides traditional CBS with uniform decimation, a new form with nonuniform decimation is presented. The retained samples correspond to sensor locations of a virtual dilated sparse array. The coarray method then enables the estimation of
$O(R^{2})$
sources, where
$R$
is the number of RF chains. The dilation results in larger coarray aperture and smaller estimation errors. Numerical examples are given to show the effectiveness of hybrid CBS.