Graph Neural Network Enabled Propagation Graph Method for Channel Modeling

Xiping Wang

doi:10.1109/TVT.2024.3382650

Graph Neural Network Enabled Propagation Graph Method for Channel Modeling

2024-09-01

25

PRE

AI

DOI：10.1109/TVT.2024.3382650

Original

Original request for help

Save

Abstract

En 中文

Channel modeling is considered as a fundamental step in the design, deployment, and optimization of vehicular wireless communication systems. For typical vehicular communication scenarios in urban areas, dense multipath may exist in the wireless channels. The propagation graph (PG) method is an efficient approach to simulate multipath radio propagation. In this paper, we extend the PG method into a Graph Neural Network (GNN) enabled data-driven method for calculating channel transfer function (CTF) and channel impulse response (CIR) in a given space. ChebNet, a classical GNN, is utilized for estimating the scattering coefficients of the edge gains in the PG method. The proposed GNN-enabled method performs better than baseline algorithms, such as multilayer perceptron (MLP), simulated annealing (SA) algorithm, and genetic algorithm (GA) in effectively estimating a large number of scattering coefficients in PG. Mean absolute errors of the proposed method are provided and evaluated in this paper. Additionally, the potential future research directions of the GNN-enabled PG method for channel modeling are discussed.

Keywords:

Wireless communication

Scattering

Channel estimation

Data models

Graph neural networks

Genetic algorithms

Channel impulse response

Channel modeling

graph neural network (GNN)

propagation graph (PG)

ray-tracing (RT)

Graph Neural Network Enabled Propagation Graph Method for Channel Modeling

Abstract

Journal

Organization