Robust Resource Allocation in IoT Networks Under Dynamic System Constraints

Abstract

The objective of applications and services in the Internet of Things (IoT) networks is to monitor resources, such as traffic and wireless channels and allocate them as per the varying quality of service (QoS), energy and computing conditions. This paper introduces the Robust Adaptive Sensitivity-Aware Differential Optimizer (RASA-DO), which combines and integrates the uncertainty of the physical system with the rate of resource optimization. Queue, energy, computing workload and channel states are jointly modelled, and adaptive primal and dual differential updates carry out the transmit power, bandwidth and computing resource allocation. An uncertainty-sensitivity variable determines the optimization rate, and adaptive damping helps to dampen the oscillations when the conditions are changing rapidly. Theoretical analysis is used to find the KKT points of a constrained equilibrium, a Lyapunov decay condition for disturbance-free dynamics and an ultimate error bound for bounded disturbances. RASA-DO is compared to the conventional optimizer, the standard primal-dual optimizer and a fixed-damping differential optimizer numerically. RASA-DO hits the convergence threshold at 93 iterations and has a normalized allocation error of 0.016, while the other methods have normalized allocation errors of 0.135, 0.060 and 0.050, respectively. It doesn't violate the constraint at all; its mean constraint-violation rate is 2.97%, it consumes energy 9.97 units/interval and has a delay of 22.54 ms. Even as the disturbances increase, it does not break the constraint and consumes less energy maintains a feasible resource allocation with robustness and stability across the various levels of disturbances, increasing the size of the IoT network.