Candidate: Mingcheng He
Date: November 29, 2024
Time: 10:00 AM
Location: Online - contact the candidate for more information.
Supervisor: Shen, Sherman
All are welcome!
Abstract:
Connected autonomous vehicles (CAVs), which enable information exchange and content delivery through vehicular networks to support multifarious vehicle-to-everything (V2X) applications, are expected to play a very important role in the safety and efficiency of future vehicular transportation systems. However, V2X applications necessitate stringent requirements on low delay, high reliability, and global connectivity, which pose challenges to existing terrestrial networks due to the geographic-constrained fixed infrastructure deployment and scarce resources. In this thesis, we consider satellite-terrestrial integrated vehicular network (STIVN) for supporting V2X applications with reliable, flexible, and ubiquitous service provisioning by jointly managing communications, computing, and caching resources from both terrestrial and low Earth orbit (LEO) satellite networks. Particularly, we will investigate the following three research issues: 1) how to manage resources with low complexity for the coupling decision-making in STIVN in terms of diverse communication modes with different resource costs; 2) how to flexibly coordinate resources from satellite and terrestrial networks with distinct and time-varying coverage to find the global optimal resource utilization across large-scale regions; and 3) how to adaptively schedule resources in STIVN to respond to dynamic network environments resulting from the mobility of satellites and CAVs, as well as the spatiotemporal fluctuations in service demand.
In this thesis, we propose a novel network slicing-based resource management framework to address the three issues. For the first issue, we investigate reservation-stage resource management in STIVN by jointly optimizing coupling decisions to support high-definition (HD) map distribution, where both satellites and terrestrial base stations (BSs) can deliver HD map files to CAVs. By reserving both communication and caching resources from both terrestrial and satellite networks, the goal is to minimize the overall cost of STIVN in terms of resource usage, delay, and slice reconfiguration cost while accommodating more service demand. Considering both multicast and unicast communication modes can support the HD map distribution in STIVN with different characteristics, we design a resource slicing architecture by establishing two slices accordingly to jointly support the HD map distribution application and formulate the resource management problem as a sequential decision-making problem. To solve the problem, we propose a two-layer resource slicing approach, which is a data-model co-driven approach, to find the optimal resource reservation and slicing window length decisions with low complexity. Specifically, considering the time-varying available resources and dynamic service demand, we design a reinforcement learning-based algorithm to decide the communication resources reserved in each network slice in each access point and the slicing window length to find the long-term optimal performance. Then, according to the reserved communication resources in each slice, a swap-based matching algorithm is derived to determine the optimal multicast and caching decisions in each slicing window. To address the second issue, we study reservation-stage resource management across multiple cells in STIVN to coordinate satellite and terrestrial networks. The cell has the same coverage of satellite beams, and the controller for each cell establishes network slices for each application and manages resources. Considering the distinct and time-varying coverage of satellite and terrestrial networks, it is difficult to achieve the global optimum, where each cell has different available resources, and resources from each satellite can be shared across different cells. To reserve resources for each slice, we formulate a resource slicing problem with cross-cell coordination in STIVN to minimize the overall cost related to resource usage and delay performance and propose a distributed resource slicing scheme for flexible resource management to solve the problem. To alleviate the competition of resources among different cells, an asynchronous multi-agent proximal policy optimization-based algorithm is presented to select satellites serving each cell in each slicing window, and then a distributed optimization algorithm is employed to determine the resource reservation decisions for each slice. To address the third issue, we investigate operation-stage resource management in STIVN by adaptively scheduling resources in real-time to cope with dynamic network environments. Specifically, we focus on the infrastructure-assisted environment sensing application where CAVs in the terrestrial cell require fresh environment data from satellites and roadside units (RSUs) to increase the trust level of the observation. To support satellite- and RSU-assisted environment sensing, we formulate a long-term resource scheduling problem in STIVN to satisfy sensing data freshness and reliability requirements with efficient resource usage. We propose a cooperative satellite-terrestrial resource scheduling scheme that jointly optimizes the sensing interval and resource allocation in STIVN. Leveraging the multicast capabilities of satellites to simultaneously serve multiple CAVs in a group, CAV partitioning and sensing interval determination methods are designed to minimize the cost of computing and communication resources. Based on the CAV grouping and sensing interval, we then develop a reinforcement learning-based algorithm to make real-time decisions on computing and communication resource allocation for different CAV groups to cope with highly dynamic environments. In summary, the proposed slicing-based resource management scheme can support V2X applications with reliable, flexible, and ubiquitous service provisioning and provide useful guidelines for network resource management in future vehicular transportation systems.