Direction-of-arrival estimation using Gaussian process interpolation

Gaussian processes (GP) have been shown to achieve superior performance for acoustic field prediction by interpolating sparse field observations. The superior performance of GP's for acoustic field reconstruction can likely be attributed to the ability to choose an appropriate field spatial coherence (kernel) as well as its ability to denoise measurements. GP interpolation can be extended for direction of arrival (DOA) estimation, which is an important problem in various domains ranging from radar systems to wireless networks. GP based DOA estimation has not been fully explored before. In this work, we propose a GP interpolation based method for DOA estimation. The "denoised" acoustic field is obtained using a GP which is then used for DOA estimation. We demonstrate that our method outperforms conventional beamforming method when the discretely sampled acoustic field suffers from spatial aliasing.