Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mobile Network Configuration Recommendation using Deep Generative Graph Neural Network

Shirwan Piroti, Ashima Chawla, Tahar Zanouda

TL;DR

This paper addresses the complexity of configuring a large set of RAN parameters and misconfigurations. It proposes a Deep Generative Graph Neural Network that encodes the RAN as a graph, constructs subgraphs around each cell, and uses a Siamese GNN to learn embeddings in $\mathbb{R}^d$ to drive parameter recommendations. Key contributions include a three-stage methodology (graph representation, subgraph-based inductive learning, and parameter value generation), a one-shot meta-learning framework, and misconfiguration detection via anomaly scores. On real-world datasets, the method achieves higher configuration accuracy and demonstrates generalization and robustness to concept drift, outperforming baselines. The approach holds practical value for operators and opens paths toward O-RAN deployment and explainability.

Abstract

There are vast number of configurable parameters in a Radio Access Telecom Network. A significant amount of these parameters is configured by Radio Node or cell based on their deployment setting. Traditional methods rely on domain knowledge for individual parameter configuration, often leading to sub-optimal results. To improve this, a framework using a Deep Generative Graph Neural Network (GNN) is proposed. It encodes the network into a graph, extracts subgraphs for each RAN node, and employs a Siamese GNN (S-GNN) to learn embeddings. The framework recommends configuration parameters for a multitude of parameters and detects misconfigurations, handling both network expansion and existing cell reconfiguration. Tested on real-world data, the model surpasses baselines, demonstrating accuracy, generalizability, and robustness against concept drift.

Mobile Network Configuration Recommendation using Deep Generative Graph Neural Network

TL;DR

This paper addresses the complexity of configuring a large set of RAN parameters and misconfigurations. It proposes a Deep Generative Graph Neural Network that encodes the RAN as a graph, constructs subgraphs around each cell, and uses a Siamese GNN to learn embeddings in to drive parameter recommendations. Key contributions include a three-stage methodology (graph representation, subgraph-based inductive learning, and parameter value generation), a one-shot meta-learning framework, and misconfiguration detection via anomaly scores. On real-world datasets, the method achieves higher configuration accuracy and demonstrates generalization and robustness to concept drift, outperforming baselines. The approach holds practical value for operators and opens paths toward O-RAN deployment and explainability.

Abstract

There are vast number of configurable parameters in a Radio Access Telecom Network. A significant amount of these parameters is configured by Radio Node or cell based on their deployment setting. Traditional methods rely on domain knowledge for individual parameter configuration, often leading to sub-optimal results. To improve this, a framework using a Deep Generative Graph Neural Network (GNN) is proposed. It encodes the network into a graph, extracts subgraphs for each RAN node, and employs a Siamese GNN (S-GNN) to learn embeddings. The framework recommends configuration parameters for a multitude of parameters and detects misconfigurations, handling both network expansion and existing cell reconfiguration. Tested on real-world data, the model surpasses baselines, demonstrating accuracy, generalizability, and robustness against concept drift.
Paper Structure (16 sections, 18 equations, 4 figures, 1 table)

This paper contains 16 sections, 18 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Data Description
  4. Feature Values and Pre-processing
  5. RAN Graph
  6. Datasets for inductive Graph Neural Networks
  7. Model Definition & Architecture
  8. Architecture & Training
  9. Inference
  10. Using closest semantic subgraph neighbor
  11. Using Majority voting
  12. Evaluation Metrics
  13. Configuration Accuracy Score
  14. Misconfiguration Detection Score
  15. Results
  16. ...and 1 more sections

Figures (4)

  • Figure 1: Overview of model training and configuration generation for diverse use cases.
  • Figure 2: End-to-end model architecture for configuration management framework.
  • Figure 3: 2-dimensional representation of the embeddings produced by three distinct models. Dimension of embeddings are redcued using PCA.
  • Figure 4: S-GNN's 2D embeddings represent novelty, with highlighted circles (embedding) for highest anomaly scores.