Deep Learning-Based Angular Resolution Improvement in Planar Sensor Array

The high angular resolution, which leads to an accurate direction-of-arrival (DOA) estimation, is essential in the radar systems for target detection and localization. Therefore, we propose a generative adversarial network (GAN)-based method that improves the angular resolution in target detection images. In the proposed network, we use the U-Net and the patch discriminator as the generator and the discriminator, respectively. Then, we verify the performance of the proposed method through simulations. The mean-squared error between the image generated by the proposed deep learning network and the ground truth image is 0.004, indicating a high level of similarity. In addition, the peak signal-to-noise ratio of the image with the increased resolution is about 11 dB higher than that of the original low-resolution (LR) image. By enhancing the angular resolution through the proposed method, the accuracy of DOA estimation can be improved in radar systems.