Gaussian Processes for Direction-of-Arrival Estimation With Random Arrays

Abstract

Gaussian process regression for direction-of-arrival (DOA) estimation in random linear arrays is formally introduced in this letter, a novel methodology to estimate or interpolate information in the complex domain is implemented to address the correlation between the real and complex variables of the signal model. The impinging waveform over the array aperture is sampled nonuniformly in the spatial domain using antenna elements scattered in an arbitrary manner. The output from the random array is subsequently mapped to the output of a virtual uniform array composed of an equal number of elements as the original array and a separation amounting to half a wavelength of the carrier frequency. The interpolated signal vector is finally the input to a standard root-MuSiC algorithm, which computes the DOA of incoming signals using a subspace methodology. The efficacy and robustness of the algorithm is substantiated through Monte Carlo simulations with respect to various scenarios, closely replicating real-time variations in the communication channel.