Multi-Dimensional QoS and Collaborative MAC Layer Design for Dense, Diverse, and Dynamic IoT Network
Abstract
With the ubiquitous proliferation of Internet of Thing (IoT) devices, Access Points (APs) of future Wireless Fidelity (Wi-Fi) networks are expected to support dense Stations (STAs) with diverse Quality-of-Service (QoS) requirements under dynamic channel conditions. On account of high access collision and optimization problem complexity, the performance degradation brings new challenges to the existing Media Access Control (MAC) layer design in Wi-Fi. This thesis proposes novel technologies to enable low-latency high-performance MAC layer designs, which support dense access of STAs with real-time solutions of resource allocation and link adaptation problems. Both grouping and collaborative architectures are utilized based on game and graph theory to enable service provisioning and efficient network management.
When supporting hybrid and dense access of STAs with both guaranteed and non-guaranteed QoS requirements, the performance of Non-guaranteed STAs (NG-STAs) often suffers more due to the exacerbated random-access congestion and the increased random-access collision. To overcome these challenges, we propose a Joint Traffic and Access Management (JTAM) mechanism to reshape the access traffic of Guaranteed-STA (G-STAs) and arrange the access opportunities of NG-STAs. The reshaped traffic smooths the varying Resource Units (RUs) for random access, and the grouping strategy optimizes the access efficiency of NG-STAs. Both analytical and simulation results show that JTAM improves the average throughput and reduces the average access latency of NG-STAs in the dense access scenario without impacting the performance of G-STAs.
To precisely satisfy the diverse service requirements at the MAC layer, we propose an integrated evaluation scheme based on Deep Neural Networks (DNN) that takes into account performance parameters in end-to-end transmission along with access network multi-dimensional QoS performance. Based on users' different requirements and the characteristics of the end-to-end network, we optimize the resource allocation strategy in the access network. However, the complex functional forms and multi-dimensional variables contribute to an exceedingly high time complexity in the optimization problem of resource allocation. To implement resource allocation within limited processing time, a Distributed Optimization with Centralized Refining (DO-CR) mechanism is proposed to support effective and real-time resource allocation. Specifically, the new DO-CR mechanism utilizes the distributed processing capacity of each STA in the first stage, allowing them to optimize their resource allocation schemes. AP generates the graph based on individual optimization results to indicate the topology of RU trading among devices and utilizes the graph to find Pareto Optimal in the second stage. Consequently, the resource allocation problem at the AP is simplified based on Pareto Optimality with a smaller feasible region compared to conventional optimization approaches.
To meet diverse QoS metrics for future wireless applications, this thesis proposes a multi-dimensional QoS Provisioning Link Adaptation (QPLA) to enable more flexible Link Adaptation strategies. However, with multi-dimensional variables in the objective function, the overall optimization problem of power allocation and link adaptation becomes an NP-hard Mixed-Integer Nonlinear Programming (MINLP) problem. To enable the timely adjustment of Link Adaptation, this thesis also proposes a Collaborative Link Adaptation (CLA), which decomposes the optimization problem and utilizes distributed processing capacity to facilitate the convergence rate of optimization. Considering the differences between devices in processing capacity and potential overhead, CLA utilizes game theory to coordinate AP and STAs. By broadcasting the average optimization performance at the AP, CLA encourages STAs with advanced processing capabilities to process locally, while also ensuring that STAs with limited processing capabilities could be assisted by AP. The analysis proves that CLA can achieve a Pareto-optimal solution, and simulations demonstrate that our proposed CLA performs faster convergence rate and better performance than traditional centralized schemes under time-limited conditions.