Joint Resource Allocation and Trajectory Optimization for Multi-UAV Assisted Semantic Communication Systems

Semantic communication, shifting from accurate bitlevel signal delivery to the semantic meaning convey, is an emerging paradigm that has been regarded as a breakthrough beyond the Shannon boundary. In this paper, an unmanned aerial vehicle (UAV)-assisted semantic communication system is considered for data collection, where multiple UAVs are dispatched to gather the observed data generated by ground users (GUs). To combat the possible cross interference of GUs and wireless fading, GUs can convey relevant semantic meaning rather than bit-level raw data, thus the semantic representation, UAVs' service association with GUs, GUs' transmission power as well as UAVs' trajectory should be properly scheduled to maximize semantic spectrum efficiency (S-SE) while completing the semantic data collection within the limited flight time. By leveraging the powerful non-linear approximation nature, semantic meaning generally can be extracted by encoding and decoding neural networks. However, this results in a lack of closed-formed expression for semantic analysis. Therefore, the paper proposes a matching combined with successive convex approximation (SCA) and multi-agent reinforcement learning approach (MADRL) to maximize the S-SE. Numerical simulation results reveal the effectiveness and superiority of the proposed scheme over conventional bit communication in terms of S-SE and energy consumption.