5G Traffic Generation for Practical Simulations Using Open Datasets

Model Architecture

Generated Traffics

5GT-GAN

N-HiTS 5G

Traffic generated by 5GT-GAN and N-HiTS-5G models.

Getting Started

Environment

• Ubuntu 20.04 LTS
• Docker
• Pytorch & Pytorch Lightning

Prerequisites

• Recommend install CUDA, CUDNN

Installing

5GT-GAN

cd 5GT-GAN
pip install -r requirements.txt

N-HiTS-5G

cd N-HiTS-5G
pip install -r requirements.txt

Running the tests

5GT-GAN

1. Run main.py

   Options

   in config/model/GAN.yaml

   • seq_len: Enter the number of sequence length for input/output of model
   • latent_dim: Enter the number of random noise dimension for generator's input
   • cond_dim: Enter the number of data types in train dataset
   • hidden: Enter the number of output size of LSTM layer
   • n_layers: Enter the number of LSTM layers
   • sample_size: Enter the number of batch size of fixed_z if you want to use it
   • lr: Enter the value of learning rate

   in config/data/TrafficDataModule.yaml

   • seq_len: Enter the same number of seq_len in GAN.yaml file
   • data_path: Enter the path of data csv file
   • batch_size: Enter the number of batch size for train

   in config/config.yaml

   • VERSION: Enter the version of experiment
   • MODEL_NAME: Enter the model name for checkpoint file

2. Run inference.py

   Options

   in config/checkpoint/inference.yaml

   • path: Enter the path of checkpoint file
   • version: Enter the version of experiment for output file
   • epoch: Enter the epoch of checkpoint for output file
   • batch_size: Enter the number of batch size for generate each data types

N-HiTS-5G

1. Run model_train.py

   Options

   • dataset : Select one name for the dataset want to learn.
   • datatype : Select one name for the dataset type want to learn (ul or dl).
   • hyperopt_max_evals : Enter the maximum number of evaluations for hyperparameter tuning.
   • experiment_id : Enter a title for the current experiment.

   python3 model_train.py --dataset afreeca --datatype dl \
                          --hyperopt_max_evals 10 --experiment_id test_1
   

2. Run inference.py

   Option

   • dataset : Select one name for the dataset want to learn.
   • datatype : Select one name for the dataset type want to learn (ul or dl).
   • experiment_id : Enter a title for the current experiment.
   • size : Enter the size of the traffic you want to generate (output length = horizon * size).

   python3 inference.py --dataset afreeca --datatype dl --experiment_id test_1 --size 10
   

3. Run evaluation.py

   python3 evaluation.py
   

Built With

• DAEGYEOM KIM - Implementation of the N-HiTS-5G model
• MYEONGJIN KO - Implementation of the 5GT-GAN model

License

This project is licensed under the MIT License

Citations

@article{choi2023mlbased,
  title={ML-Based 5G Traffic Generation for Practical Simulations Using Open Datasets},
  author={Yong-Hoon Choi, Daegyeom Kim, Myeongjin Ko, Kyung-yul Cheon, Seungkeun Park, Yunbae Kim, Hyungoo Yoon},
  journal={IEEE Communications Magazine},
  year={2023},
  note={To appear}
}

Acknowledgments

• Date of submission October 31, 2022.
• This work was supported by the In-stitute for Information & communications Technology Promotion (IITP) grant funded by the Korea government (MSIT) (No. 2021-0-00092); and in part by the National Research Foundation of Korea (NRF) grant funded by the Korea government Ministry of Science and ICT (No. 2021R1F1A1064080).