UAV-Mounted SIM: A Hybrid Optical-Electronic Neural Network for DoA Estimation
Shining Lin, Jiancheng An, Lu Gan, Mérouane Debbah
Abstract
Unmanned aerial vehicle (UAV) communication plays a pivotal role in achieving ubiquitous connectivity for the sixth-generation (6G) networks. Accurate and real-time direction of arrival (DOA) estimation is crucial for beamforming in UAV communication systems. However, the existing high-precision DOA estimation algorithms encounter high computational complexity when being implemented on a UAV with the on-board signal processing constraints. To tackle this issue, a hybrid optical-electric neural network (HOENN) is utilized for DOA estimation, which is capable of generating angular spectrum based solely on amplitude observation. The proposed HOENN consists of two components: a stacked intelligent metasurfaces (SIM)-enabled diffractive neural network, which is mounted on UAV and can process signals in the wave domain at the speed of light with low energy consumption, and a fully connected layer for processing the received amplitude signal. Finally, the simulation results validate that the HOENN achieves significant performance gain compared to the conventional beamforming (CBF) method, albeit with its lower cost and RF-related power consumption.
BibTeX
@inproceedings{icassp2025_uavmountedsimahy,
title = {UAV-Mounted SIM: A Hybrid Optical-Electronic Neural Network for DoA Estimation},
author = {Shining Lin and Jiancheng An and Lu Gan and Mérouane Debbah},
booktitle = {ICASSP 2025},
year = {2025}
}