Intelligent and Customized Resource Management and Network Cooperation in Beyond 5G and 6G Networks
Abstract
In recent years, a plethora of new wireless applications have emerged, each with its specific quality of service (QoS) requirements and quality of experience (QoE) perception mechanism. These developments pose significant challenges to resource management and network cooperation in future wireless networks. To overcome these challenges, this thesis focuses on the development of several intelligent and customized resource management and network cooperation techniques for beyond 5G and 6G networks.
First, for reducing power consumption in cloud radio access network (CRAN), a traffic prediction-enabled energy-efficient dynamic computing resource allocation scheme is proposed. Specifically, a novel wireless traffic prediction method based on two-dimensional convolution neural network (CNN) long short-term memory (LSTM) model with temporal aggregation is proposed. Moreover, a multi-start simulated annealing (MSA) algorithm is proposed for energy-efficient dynamic computing resource allocation in CRAN. Simulation results demonstrate that compared with fixed computing resource allocation, the proposed scheme could reduce the total power consumption by 52.18%.
Second, categorized QoS provisioning is a natural idea to satisfy the diverse QoS requirements of different categories of applications through a two-stage resource allocation scheme. To optimize the key QoS indicators of three different application categories and mitigate backhaul network congestion simultaneously, a decoupling-based iterative optimization (DBIO) algorithm is proposed for the first-stage resource allocation from radio access network (RAN) to tenants to achieve categorized QoS provisioning.
Third, in downlink 6G heterogeneous networks (HetNets), for achieving tailored QoS provisioning to ensure fairness among users with different QoS requirements, an interactive optimization algorithm (IOA) is proposed for joint user association and resource allocation.
Last, for achieving application-specific QoE enhancement, multi-user multi-application packet scheduling in downlink 6G RAN is investigated, which is formulated as a Markov decision process (MDP) problem. For solving this problem, a deep deterministic policy gradient (DDPG)-based solution is proposed. Moreover, to avoid resource waste and improve the performance of DDPG-based packet schedulers, a knowledge embedding method is proposed. Simulation results demonstrate that utilizing the proposed knowledge embedding method, the performance of the DDPG-based packet schedulers could be improved by at least 8.07% on the test dataset in terms of average reward per time slot.