A CNN-Based Radar Direction-of-Arrival Prediction Over Sea Surface

This article examines a direction-of-arrival (DOA) estimation approach realized by a convolutional neural network (CNN), at different radar operating frequencies. The advanced integral equation model (AIEM) is adopted as a descriptive model to simulate the polarized bistatic radar measurement datasets. Then the K-distributed speckle noise is concerned. A framework of convolutional neural network (CNN) is designed to use underlying statistical characteristics of speckle noise-influenced radar measurements to estimate the incidence directions of the received signal. According to experimental results, the average root-mean-squared error for the prediction of incident angle is approximately one degree, and for incident azimuth angle, the average root-mean-squared error falls between 3.2° and 3.5°. Results demonstrate the CNN-based approach achieves good performance in DOA estimation. Hence, it is fundamental to connect a descriptive model to a predictive model in radar DOA to attain good accuracy without involving a large-scale antenna array.