Deep Residual Learning for Channel Estimation in UAV mmWave Systems: A Model-Driven Approach

Accurate channel estimation is pivotal in satisfying the requirement of ultra-high data rates in future wireless networks. In this paper, we propose a novel model-driven channel estimation algorithm for an unmanned aerial vehicle (UAV) millimeter wave (mmWave) communication system, in which a UAV equipped with a large-scale antenna array performs the channel estimation based on the pilot signals from another UAV equipped with a single antenna. Considering the sparseness of the UAV mmWave channels, in the proposed algorithm, we exploit the use of deep residual learning to judiciously learn the key parameters of the alternating direction method of multipliers (ADMM) framework, reformulating the classical model-based channel estimation algorithm into deep learning model-based model-driven channel estimator. Compared to various benchmark schemes, we validate the efficiency of our suggested model-based channel estimation algorithm, and show how it can achieve good channel estimation accuracy even when the link quality is moderate and the pilot resources are limited.