Stability-Guaranteed Delay Optimization for Secure Task Offloading in Multi-Tier UAV Networks

Abstract

The rapid evolution of low-altitude wireless networks (LAWNs) has driven massive IoT connectivity, generating computation-intensive traffic that exceeds the limited onboard capabilities of IoT devices and necessitates task offloading. In future 6G networks, the space-air-ground integrated network (SAGIN) will extensively deploy high-altitude platforms (HAPs) and UAVs as edge computing nodes. However, existing offloading schemes rarely consider tiered UAV architectures with heterogeneous computing capabilities or beam-dependent transmission effects. To address this gap, this paper considers a dual-tier UAV-assisted LAWN where lower-tier edge UAVs (E-UAVs) offload computation tasks to an uppertier HAP. Furthermore, we incorporate an encryption delay model that captures the security demand and data size of each task, adding the encryption delay to the total task completion time. The proposed framework features a cross-layer co-design of security (application layer) and resource allocation (PHY/MAC/NET layers). We formulate a long-term average delay minimization problem subject to queue stability and effective processing rate constraints. Using Lyapunov optimization, we transform the problem into a per-slot subproblem and propose a PPO (Proximal Policy Optimization) algorithm with a customized state-action-reward design. Numerical results demonstrate that the proposed scheme achieves fast convergence, low average delay, and queue stability while satisfying security demands.