Security-Aware Adaptive Offloading for Resource-Constrained UAV Systems Using Homomorphic Encryption

Abstract

Unmanned Aerial Vehicles (UAVs) increasingly rely on edge-assisted computation for applications such as smart delivery, surveillance, emergency response, and airspace compliance. However, transmitting sensitive information, including UAV coordinates and mission policies, to third-party or partially trusted edge servers may expose private or mission-critical data. Homomorphic Encryption (HE) protects such data by enabling computation over ciphertext but introduces latency, ciphertext expansion, and computational overhead. Existing studies demonstrate the use of HE for UAV inference, secure federated learning, and privacy-preserving navigation, but generally assume encrypted computation remains operationally practical once deployed. This paper investigates the operational feasibility of HE-based secure offloading in resource-constrained UAV systems by analysing the relationship between encrypted computation latency, policy complexity, and mission latency budgets. We propose a security-aware adaptive offloading framework using CKKSbased encrypted per-edge geofence arithmetic as a representative mission-policy workload. The framework estimates HE latency from policy complexity and selects secure edge offloading or local processing according to the UAV's latency budget. The decrypted CKKS edge scores achieved a mean absolute error (MAE) of $8.62 \times 10^{-6}$, a maximum absolute error of $6.35 \times 10^{-5}$, and 100% agreement with plaintext geofence decisions. Experimental results show that HE latency increases with policy complexity; secure offloading remained feasible for all evaluated polygon complexities (3-30 edges) under latency budgets of 200 ms and above, while a 100 ms budget supported policies containing up to 10 polygon edges. These results provide a practical foundation for latency-aware, privacy-preserving computation placement in UAV edge systems.