Support vector regression based DOA estimation in heavy tailed noise environment

In this paper, Direction Of Arrival (DOA) estimation of an incoming signal in the presence of heavy tailed noise environment is considered. Heavy tailed noise is modeled using Laplace distribution. Support vector regression (SVR) based approach is considered for DOA estimation. The computational burden of SVR due to the larger dimensionality of feature vectors/ signal vector is reduced by considering forward-backward (FB) averaging based covariance matrix of the received signal. The proposed approach is validated by several numerical results and it is found that, SVR based DOA estimation results in better performance than the MUSIC and ESPRIT in terms of estimation accuracy and computational time.

I. Introduction

Recently, the requirement of DOA estimation increases, in order to meet the stringent demands in fifth generation wireless systems (5G) [1]. Generally, DOA estimation involves in determining the direction of incoming electromagnetic or acoustic signals that impinges on the antenna or sensor array. Commonly, the DOA estimation techniques can be divided into three categories, namely, classical, subspace based and maximum likelihood (ML) techniques [2]. Several methods have been proposed under each category. Multiple signal classification (MUSIC) and estimation of signal parameter via rotational invariance technique (ESPRIT) [3], [4] are the most widely used subspace based methods among the others. But subspace methods are computationally intensive because of the Eigen value decomposition involved in it. Recently, the problem of DOA estimation has been performed by one of the supervised learning methods called Support vector regression (SVR) [5].