Learning-Based Collaboration for Secure Transmission Effectiveness Maximization in Low Altitude MEC Systems

Low altitude mobile edge computing (MEC) offers promising prospects through unmanned aerial vehicle (UAV) services. However, the scarcity of resources, coupled with the growing personalized demands of devices, complicates the effective balance in UAV-MEC systems. Furthermore, the increasing threat of active aerial eavesdropping (AAE) poses severe risks of eavesdropping and attacking, significantly affecting the security of offloaded computation. To overcome these challenges, we propose a learning-based secure transmission scheme against AAE for UAV-MEC systems. Specifically, a customized secure transmission effectiveness is designed to effectively guide the optimization of limited resources and trajectory to meet individualized secure requirements. Then, multi-dimensional resources, including offloading decision, transmit power, computation frequency and UAV trajectory, are jointly optimized to maximize the secure transmission effectiveness with satisfying the individualized computing demands of devices. To address the multi-variables-coupling and closed-form-lacking problem, a deep reinforcement learning-based collaboration of multi-networks trajectory optimization and resource allocation (DRCTORA) scheme is proposed, where deep neural network ordering preserving (DNOP) is employed to generate offloading decisions, and double deep Q-network (DDQN) is connected in DNOP to obtain the solutions of the other variables. Simulation results show that the proposed DRCTORA enhances the secure transmission performance compared with benchmarks.