Direction-of-arrival (DOA) estimation using deep-neural-network (DNN)

Abstract

Direction-of-arrival (DOA) estimation is a key area in the field of antenna array signal processing. It has great significance in digital communication, IoT applications and national security. Most studied DOA estimation methods such as MUSIC and ESPRIT involve the calculation of signal correlation matrixes and eigenvalue decompositions (or singular value decompositions). To formulate the correlation matrixes, however, we must extract both amplitude and phase information from the incident signals. As such, the synchronous operation of the antenna array can be ensured. In view of various practical challenges, a novel DOA estimation approach is proposed in this dissertation. The proposed method adopts neural networks to process the power of the received signals from the various antenna elements. The processed signals were classified into a one of eight discrete directions. In this dissertation, two network structures, namely SDAE-DNN and SDAE-CNN were built and tested independently. Both structures achieved relatively high prediction accuracies in the direction classification task.