Intelligent resource allocation for UAV-IRS assisted semantic task offloading

Semantic communication provides a promising scheme to enhance spectral efficiency and system reliability by transmitting only critical semantics. In this paper, the research focus is directed towards the unmanned aerial vehicle (UAV)-intelligent reflective surface (IRS) assisted semantic mobile edge computing (SMEC) network, given computationally intensive nature of semantic tasks. Specifically, the UAV can offload semantic task data to edge computing servers to alleviate the task implement pressure on the cost of power and energy so as to improve communication flexibility. Meanwhile, the IRS is designed to tackle the challenges posed by unfavorable communication environments. To find an optimal solution to realize the above-mentioned ideal, we frame the problem of joint optimization of UAV’s trajectory, transmission power, IRS reflection coefficient and the number of semantic symbols. Considering the existing hybrid discrete-continuous action space and the non-convex nature of the optimization problem, a deep reinforcement learning (DRL) scheme is proposed. In particular, the DDPG algorithm is employed to control the IRS reflection coefficients and transmission power, while D3QN algorithm is used to control the UAV trajectory and the number of semantic symbols. The results of simulation indicate that the proposed scheme greatly improves the transmission rate by 67% in comparison with the traditional communication methods.