Programmable metasurface two-dimensional DOA estimation by combining matched filtering and U-Net

This paper presents a novel two-dimensional (2D) metasurface-based DOA estimation method by combining matched filtering and U-Net (MF-UNet). Specifically, this method transforms the DOA estimation problem into an image processing problem. The received signal is first converted to the spatial domain via matched filtering to generate a 2D spatial spectrum map. Then, the normalized spatial spectrum image is processed by the improved U-Net network for denoising and artifact removal. Finally, the DOA estimation result is obtained through peak finding operation. Simulation results demonstrate that MF-UNet can effectively balance accuracy and complexity while maintaining high estimation performance in low signal-to-noise-ratio (SNR) scenarios.