Deep Learning Based DoA Estimation Using Compressed Sensing

Abstract

The direction of Arrival (DoA) is a widely used technique in array signal processing applications. Conventional DoA algorithms require many snapshots and a high SNR to perform well. However, in practical DoA applications, a large number of snapshots and a high SNR are not available in many scenarios. When the SNR is low, and the number of snapshots is small, the results estimated by the conventional DoA algorithm are not satisfactory. Therefore, this thesis proposes a scheme that combines autoencoder and OMP algorithms used for DoA estimation with a low SNR and single snapshot. This thesis performs performance analysis in terms of DoA estimation error, detection success rate, and processing time. The performance of the DoA algorithm is also evaluated by implementing a Software-Defined-Radio (SDR) platform for RF tests. The experimental results demonstrate that the proposed scheme significantly improves accuracy, and the detection success rate and processing time are better than other conventional DoA estimation algorithms.