Abstract:In this paper, the problem of vehicle service mode selection (sensing, communication, or both) and vehicle connections within terahertz (THz) enabled joint sensing and communications over vehicular networks is studied. The considered network consists of several service provider vehicles (SPVs) that can provide: 1) only sensing service, 2) only communication service, and 3) both services, sensing service request vehicles, and communication service request vehicles. Based on the vehicle network topology and their service accessibility, SPVs strategically select service request vehicles to provide sensing, communication, or both services. This problem is formulated as an optimization problem, aiming to maximize the number of successfully served vehicles by jointly determining the service mode of each SPV and its associated vehicles. To solve this problem, we propose a dynamic graph neural network (GNN) model that selects appropriate graph information aggregation functions according to the vehicle network topology, thus extracting more vehicle network information compared to traditional static GNNs that use fixed aggregation functions for different vehicle network topologies. Using the extracted vehicle network information, the service mode of each SPV and its served service request vehicles will be determined. Simulation results show that the proposed dynamic GNN based method can improve the number of successfully served vehicles by up to 17% and 28% compared to a GNN based algorithm with a fixed neural network model and a conventional optimization algorithm without using GNNs.
What problem does this paper attempt to address?
This paper attempts to address the problem of how to optimize the selection of sensing and communication service modes and vehicle connections in vehicular networks under the Terahertz (THz) band to maximize the number of successfully served vehicles. Specifically, it investigates how service provider vehicles (SPVs) choose between providing sensing only, communication only, or both services, and determines which requesting vehicles will be served. The problem is formulated as an optimization problem aimed at maximizing the number of successfully served vehicles by jointly deciding the service mode of each SPV and its associated requesting vehicles. To achieve this goal, the authors propose a dynamic graph neural network (GNN) model that can select appropriate graph information aggregation functions based on the vehicular network topology, thereby extracting more vehicular network information and outperforming traditional static GNN models.
### Main Contributions
1. **Practical Solution**: A practical solution is proposed for utilizing the THz band in vehicular networks to achieve joint sensing and communication, where SPVs can provide sensing, communication, or both services.
2. **Optimization Problem**: An optimization problem is defined to maximize the number of successfully served vehicles by jointly deciding the service mode of each SPV and the requesting vehicles it serves. The problem considers severe THz path loss, blockage of sensing/communication links, and the dynamic environment of vehicle mobility and service demands.
3. **Dynamic GNN Model**: A dynamic GNN model is proposed that can select appropriate graph information aggregation functions based on the vehicular network topology, thereby extracting more vehicular network information. Compared to traditional static GNN models, this model jointly optimizes the aggregation function selection parameters and GNN model parameters during training.
### Simulation Results
Simulation results show that the proposed dynamic GNN-based method can increase the number of successfully served vehicles by up to 17% and 28% compared to fixed neural network model GNN algorithms and traditional optimization algorithms without GNN, respectively.
### System Model and Problem Formulation
- **System Model**: A vehicular network consisting of a set of communication service requesting vehicles, a set of sensing service requesting vehicles, and a set of service provider vehicles is considered. Each SPV can use the same THz band to provide sensing, communication, or both services.
- **Vehicular Network Dynamics**: Includes vehicle mobility, vehicle blockage, and the unknown working status of each SPV (i.e., providing service or unable to provide service).
- **Optimization Problem**: The goal is to optimize the service mode of each SPV and its associated vehicles to maximize the number of successfully served vehicles. Specifically formulated as:
\[
\max_{\alpha, \beta} |B| + |O|
\]
where the constraints include:
- Each communication service requesting vehicle can only be served by one SPV.
- Each sensing service requesting vehicle can only be served by one SPV.
- The sets of successfully served communication vehicles \( B \) and successfully served sensing vehicles \( O \) must meet specific delay and SINR requirements.
### Dynamic GNN Solution
- **Graph Representation**: The vehicular network is represented using a graph, where each vehicle is a node, and each connection link (such as sensing links, communication links, and interference links) is an edge.
- **Dynamic GNN Model**: Unlike traditional static GNN models, the dynamic GNN model can select appropriate functions from multiple node aggregation functions and layer aggregation functions to extract more vehicular topology features.
- **Training Process**: By training the dynamic GNN model, it can extract features of newly arrived vehicles without retraining, thereby reducing training complexity.
In summary, this paper effectively addresses the problem of optimizing the selection of sensing and communication service modes and vehicle connections in vehicular networks under the THz band by proposing a dynamic GNN model, thereby increasing the number of successfully served vehicles.