An Intelligent Direction Finding Method with Deep Neural Network

Abstract

Different from the model-driven direction finding (DF) methods, the data-driven DF methods have many advantages, such as not relying on the array geometry, no need for a special channel calibration module, and better adaptability to the DF system error. In this paper, an intelligent DF method based on deep neural network (DNN) is developed. It is composed of auto-encoder network and residual neural network. Numerical simulation experiments have demonstrated the superiority of the proposed method in direction of arrival (DOA) estimation precision especially when the signal-to-noise ratio (SNR) is low.