Left-Right Ambiguity Resolution Methods Based on Deep Neural Networks

Abstract

DOA using a single linear array with omnidirectional sensors results in an identical beam along the azimuth and its complementary angle due to the symmetry along the array axis. This is the Left/Right (LR) ambiguity problem. LR ambiguity resolution is generally achieved either by maneuvering own ship or having more than one linear array in parallel. This paper applies a Deep Neural Network (DNN) for LR resolution and analyzes its performance on a twin array. We address the left-right resolution problem on the basis of probability of estimated correct state. Defining $\hat{\zeta}$ as the estimated state and $\zeta$ as the correct state, the probability of error $p(e)$ in estimating the correct state is defined. The objective of the LR resolution algorithms is to minimize $p(e)$. Binary classification based on a deep neural network for twin array is used to resolve the LR ambiguity. Array perturbation in terms of amplitude and phase error is introduced to examine the performance.