A wireless channel multipath prediction method based on point cloud deep learning

By using a point cloud-based deep learning method, the multipath propagation characteristics of the channel are predicted in real time, which solves the problem of insufficient channel modeling in dynamic environments in existing technologies, realizes high-precision channel state tracking and prediction, and supports key technologies for 6G communication systems.

CN122293237APending Publication Date: 2026-06-26SOUTHEAST UNIV +1
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SOUTHEAST UNIV
Filing Date
2026-02-04
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Existing channel modeling methods struggle to capture the multipath propagation characteristics of channels in real time and efficiently under dynamic conditions, especially in scenarios such as autonomous driving, vehicle-to-everything (V2X) and mobile IoT, where they cannot accurately predict dynamic changes in channel state.

Method used

A point cloud-based deep learning approach is adopted. By acquiring and preprocessing 3D point cloud data, scene geometric features are constructed. The RayFormer model is trained and combined with ray tracing technology and physical confidence calculation to predict the multipath propagation characteristics of the channel in real time.

Benefits of technology

It achieves high-precision, real-time channel state tracking in complex dynamic environments, supports channel estimation and beamforming in 6G communication systems, and improves the performance of wireless communication networks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122293237A_ABST
    Figure CN122293237A_ABST
Patent Text Reader

Abstract

This invention discloses a wireless channel multipath prediction method based on point cloud deep learning, involving the cross-application technology of deep learning and wireless channel modeling. The method includes: 1) acquiring 3D point cloud data of a communication scene and preprocessing it to obtain standardized 3D point cloud data; 2) generating Gaussian scattering primitives (GSPs) representing the scene's geometric structure and scatterer distribution using the proposed construction and optimization algorithm, and saving the spatial location and shape information of the GSPs as GSP features; 3) combining the GSP features with the spatial location information of the transmitter (Tx) and receiver (Rx), inputting them into a RayFormer model for training, and optimizing the model to predict multipath information; 4) filtering the interaction point coordinates predicted by the RayFormer model using a physical confidence calculation algorithm, ultimately obtaining geometrically reasonable prediction points. This invention can mine environmental geometric information in 3D point clouds, achieving accurate prediction of multipath propagation paths in complex scenes.
Need to check novelty before this filing date? Find Prior Art