Deep Learning Based Energy-Efficient Hybird RSMA for UAV-Assisted mmWave Communications

This paper investigates hybrid rate-splitting multiple access (RSMA) in unmanned aerial vehicle (UAV) assisted millimeter-wave (mmWave) communication network (RSMA-UAV-MMWCN), where a UAV transmits messages to multiple ground user equipment under the influence of an external jammer. We formulate a non-convex joint optimization problem involving hybrid RSMA matrices and a common rate allocation vector, with the objective of maximizing energy efficiency while approaching the performance of ideal hybrid RSMA. Departing from traditional non-convex problem-solving methods, we introduce a hybrid RSMA optimization scheme based on deep residual networks to enhance the feasibility of hybrid precoding and decoding. Initially, due to the absence of standardized and universal datasets, we propose a dataset generation algorithm to create training and testing datasets for subsequent communication model training. Subsequently, we construct a loss function that integrates the objective function with the constraints of the optimization problem. Lastly, to ensure that the optimization variables strictly comply with the constraints, we design a mandatory constraint module comprising modulus, power, and rate constraint sub-modules. Simulation results demonstrate that the proposed algorithm surpasses traditional optimization methods, and RSMA shows significant advantages over conventional multiple access (MA) schemes.