Resourse Allocation in the Wireless Internet-of-Things

Abstract

The Internet-of-Things (IoT) is widely regarded as a promising paradigm, marking a revolutionary shift in the way that technology interacts with the world. Despite the rising popularity and extensive integration of IoT across diverse domains, the development and deployment of these interconnected systems are met with considerable challenges. One notable challenge is the scarcity of spectrum resources, which poses a significant obstacle in accommodating the massive data transmission. In addition, small-sized mobile terminals are constrained by limited computation and energy resources. These limitations make the traditional standalone operation of devices increasingly unfeasible. In this thesis, we delve into comprehensive task offloading and resource allocation strategies to utilize of limited resources in wireless IoT. In the first part, we propose a full-dimensional task offloading scheme in the multi-layer computing network. On this basis, we formulate an mixed-integer nonlinear programming (MINLP) problem and develop an inverse reinforcement learning (IRL) based algorithm to solve this problem. Without sacrificing the global optimality, the algorithm can significantly accelerate the optimal branch-and-bound (B&B) algorithm. In the second part, we delve into strategies to optimize energy efficiency within the downlink cell-free massive MIMO systems. We develop a green energy scheme and formulate it as a non-convex MINLP problem. To solve this problem, we propose a novel optimization-embedded deep reinforcement learning (DRL) algorithm, which enjoys the benefits of directly inferring solutions for the formulated problem. In the last part, we develop a NOMA-based task offloading scheme in a multi-layer computing network. On this basis, we formulate the task offloading scheme as a non-convex mixed-integer optimization problem and propose a reincarnating DRL algorithm, where accumulated apriori information is incorporated for fast retraining.