Wireless Precoding GNN for Hypergraph Edge Correlation Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current graph neural networks (GNNs) exhibit poor generalization performance and are unable to effectively learn optimal precoding strategies in dynamic wireless communication environments, particularly in scenarios involving hypergraphs, and struggle with resolving hybrid digital-analog precoding issues.
Innovation Solution
A dimension-scalable graph neural network (GNN) is employed, utilizing two non-linear units to calculate and aggregate edge features, allowing for the consideration of mutual impacts between edges, and enabling the GNN to operate in hypergraphs and solve hybrid digital-analog precoding problems by using edge features instead of specific matrices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a current graph neural network is used for wireless strategy optimization, then the system can process dynamic changes in wireless environment, but the generalization performance is poor and it cannot resolve hypergraph-related problems
Solution Approach 1:
The patent extends the traditional graph neural network from operating on standard graphs to operating on hypergraphs, which is a dimensional generalization. This allows the model to handle higher-order relationships and complex wireless scenarios while maintaining adaptability to dynamic environments and improving generalization performance through dimension-scalable architecture
2Productivity
If a graph neural network uses specific matrix structures for processing, then the computation is efficient, but the model cannot be applied to hypergraphs and hybrid digital-analog precoding problems
Solution Approach 1:
The patent designs a universal graph neural network architecture that can process both traditional graphs and hypergraphs, as well as handle various precoding problems (digital and hybrid digital-analog). The dimension-scalable design allows the same model to be applied across different problem types and dimensions, achieving multi-functionality without sacrificing computational efficiency
3Device complexity
If the GNN does not consider mutual impact between different edges, then the computation is simpler, but the optimal precoding strategy cannot be obtained
Solution Approach 1:
The patent merges the features of multiple edges by considering their mutual impacts through aggregation operations in the graph neural network. This allows the model to capture the interdependencies between different edges while maintaining computational tractability through efficient aggregation functions that combine multiple edge contributions
Data Source
Figure 1~3
Figure 4~6
Figure 7~8
AI summary
An information processing method includes: obtaining channel state information between at least one network device and at least one terminal device; and processing the channel state information by using a GNN, to obtain precoding between the at least one network device and the at least one terminal device, where the GNN includes a first non-linear unit and a second non-linear unit. The first non-linear unit is configured to calculate a correlation between an edge feature of a target edge in graph data used by the GNN and an edge feature of each edge on each target node, to obtain at least one correlation result, and the target node includes a node connected to the target edge and a node of a same type as the node connected to the target edge. The second non-linear unit is configured to aggregate a result obtained by summarizing the edge feature of the target edge and the at least one correlation result, to update the edge feature of the target edge. In this method, an optimal precoding strategy can be learned, and a problem of hybrid digital-analog precoding can be resolved.