Generation of synthetic 5G network dataset using Generative Adversarial Network (GAN)
While the Fifth Generation (5G) network is actively being deployed in most countries to create new possibilities for better lifestyle and economic development, it is a technology that is currently being a focal point for researchers across the world along with 6G. Starting from 3GPP Release-18, Arti...
| Main Authors: | , , , , |
|---|---|
| Format: | Conference or Workshop Item |
| Language: | English |
| Published: |
Institute of Electrical and Electronics Engineers
2024
|
| Online Access: | http://psasir.upm.edu.my/id/eprint/116516/ http://psasir.upm.edu.my/id/eprint/116516/1/116516.pdf |
| Summary: | While the Fifth Generation (5G) network is actively being deployed in most countries to create new possibilities for better lifestyle and economic development, it is a technology that is currently being a focal point for researchers across the world along with 6G. Starting from 3GPP Release-18, Artificial Intelligent (AI) and Machine Learning (ML) are identified as enabler towards intelligent network in 5G and beyond. Nevertheless, the models based on AI/ML need a sufficient amount of data for learning patterns and relationships, enabling them to provide precise predictions for unfamiliar data and situations. The existence of Generative Adversarial Network (GAN) helps solve the issue by generating fake data from an existing dataset to resemble real-world data to be used in training and testing of different algorithms. In this paper, the process of generating synthetic data of 5G network was demonstrated from an extensive test drive results that will encourage innovation in mobile communication. Generation of data use two types of GAN which are the Conditional Tabular GAN (CTGAN) and Topological Variational Autoencoder (TVAE). The two algorithms were compared based on statistical analysis such as the distribution and Pearson Correlation analysis. TVAE showed a better overall performance score (94.14%) over CTGAN (89.66%) when compared with the original data, but the CTGAN produced more similar distribution for certain individual columns. |
|---|