Secure Multihop Partial Task Offloading and Resource Allocation in Cache-Assisted OFDMA- and NOMA-Enabled Ultradense IoT Networks

Abstract

With the exponential growth in Internet of Things (IoT) devices (IDs), ultradense IoT networks (UINs) have emerged as a promising architecture to accommodate massive computation and service demands. However, they confront critical challenges, including severe co-channel interference, excessive energy consumption, and potential communication security vulnerabilities. To address these issues, we present a secure caching-aided multihop computation-offloading framework for UINs integrated with orthogonal frequency-division multiple access (OFDMA) and nonorthogonal multiple access (NOMA). Specifically, we formulate an optimization problem to minimize the total energy consumption of IDs under fair resource allocation, delay bounds, and security cost thresholds, with joint optimization of device association, wireless channel allocation, security service distribution, band partitioning, local power/capability, and offloading/caching decisions. An improved gradient-based optimizer (IGBO) is then proposed to solve this nonlinear mixed-integer problem, and we systematically analyze its convergence, complexity, and parallel implementation feasibility. This algorithm enhances the traditional gradient-based optimizer (GBO) through substantial structural modifications and the integration of a best neighborhood search (BNS) strategy. Simulation results reveal that under a high satisfaction rate of delay constraints, the IGBO achieves lower energy consumption and security costs than baseline algorithms. Compared with the GBO, the IGBO reduces the total cost by up to 7.29% and improves the satisfaction rate of cost constraints by up to 18.88%, while its total energy consumption is reduced by up to 82.21%. Moreover, the IGBO achieves substantial performance improvements, for example, 29.58%–43.97% faster convergence at the 99% threshold, with only about 1.25% additional computational overhead. Furthermore, the reduction in total energy consumption grows gradually as ID density and task-data size increase.