Cellular Network Performance Estimation With Graph Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network simulators are difficult to calibrate and require extensive real-world data sets for training, while AI-based models need costly data collection to predict network performance changes.
Innovation Solution
A performance estimator model using a combination of autoencoders and graph neural networks (GNN) to estimate network performance without requiring data sets of before-and-after measurements, leveraging network snapshots with varying configuration parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AI-based models are used to predict network performance changes, then prediction accuracy is improved, but data collection cost increases
Solution Approach 1:
The patent creates a digital twin (virtual copy) of the cellular network that replicates network behavior without requiring extensive real-world data collection. The graph neural network model learns from a synthetic training dataset generated from network specifications and propagation models, rather than requiring costly before-and-after measurement datasets from actual network modifications.
Solution Approach 2:
The system performs preliminary training of the AI model using synthetic data generated from network configurations and propagation models before deployment. This pre-training phase creates a robust model that can make accurate predictions without requiring extensive real-world data collection during operational use.
2Measurement precision
If network simulators are used to model network elements, then performance estimation capability is improved, but calibration difficulty increases
Solution Approach 1:
The patent replaces traditional mechanical/calibration-based network simulators with an AI-based graph neural network model. Instead of requiring manual calibration of simulator parameters to match real network behavior, the model learns network patterns directly from training data, eliminating the complex calibration process while maintaining accurate performance estimation.
Solution Approach 2:
The system transforms the calibration problem into a learning problem by using trainable parameters in the neural network that automatically adapt to network characteristics during training, rather than requiring manual adjustment of simulator parameters. The model learns optimal parameter values from data rather than requiring expert calibration.
3Measurement precision
If comprehensive training datasets with before-and-after measurements are collected, then model training accuracy is improved, but time consumption increases
Solution Approach 1:
The patent generates synthetic training data by creating virtual copies of network scenarios through propagation models and configuration variations. Instead of waiting for real network modifications and measurements, the system synthesizes training examples that capture the essential patterns of network behavior under different configurations, dramatically reducing data collection time while maintaining model training accuracy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
According to an aspect, there is provided a computer-implemented method for training a performance estimator model (22) for estimating the performance of a cellular network. The performance estimator model (22) comprises an encoder stage (24) comprising a plurality of encoders and a decoder stage (26) comprising a plurality of decoders. The network comprises a plurality of cells, and the cellular network has a plurality of configuration parameters and a configuration parameter of interest that are each configurable per cell. The method comprises: (I) obtaining (901) a training data set, the training data set comprising measurements of a plurality of performance parameters for the plurality of cells, wherein different values of the configuration parameters are being used among the plurality of cells, wherein the training data set further comprises respective values of the plurality of configuration parameters for the plurality of cells; (II) training (903) the encoders to encode the training data set into a respective representation for each cell, wherein each encoder receives, for a respective cell, the measurements of the plurality of performance parameters and corresponding values of the plurality of configuration parameters for that cell, and wherein a layer of each of the plurality of encoders are interconnected as a neural network representing the cellular network such that information on relationships between different pairs of cells in the cellular network is taken into account in the encoding; (ill) training (905) the decoders to decode a respective representation to determine a subset of the plurality of performance parameters for the respective cell associated with the configuration parameter of interest, wherein each decoder receives the respective representation and a current value of the configuration parameter of interest, wherein a layer of each of the plurality of decoders are interconnected as a neural network representing the cellular network such that information on relationships between different pairs of cells in the cellular network is taken into account in the decoding; (iv) determining (907) a value of a loss metric that is based on a difference between the input training data set and the output of the decoders; and (v) repeating (909) steps (II), (ill) and (iv) to retrain the encoders and decoders to obtain an improved value of the loss metric.