Gaussian process regression for array interpolation

Abstract:

This paper introduces the Gaussian process regression formulation for estimating the bearings of signals with randomly spaced antenna arrays. A generalized methodology to estimate complex variables using a multivariate Gaussian is formulated. The incoming signals are sampled nonuniformly in the spatial domain using antenna elements at random locations on a linear axis. The nonuniformly sampled outputs of the array are subsequently interpolated to map the outputs of a virtual uniform array using a multivariate Gaussian process regression. The estimated signal vectors are the input to a root-MuSiC algorithm for computing the directions of arrival of signals. The proposed algorithm is subjected to Monte Carlo simulations and preliminary findings are compared to the output of a standalone root-MuSiC algorithm to validate the efficacy of the approach.

Año de publicación:

2019

Keywords:

• Direction of arrival estimation
• unsupervised learning
• Machine learning
• Multivariate Gaussian processes
• Complex Gaussian process
• Adaptive arrays
• Array synthesis

Fuente:

scopus