LLM4CP

B. Liu, X. Liu, S. Gao, X. Cheng and L. Yang, "LLM4CP: Adapting Large Language Models for Channel Prediction," in Journal of Communications and Information Networks, vol. 9, no. 2, pp. 113-125, June 2024, doi: 10.23919/JCIN.2024.10582829. [paper]

Dependencies and Installation

• Python 3.8 (Recommend to use Anaconda)
• Pytorch 2.0.0
• NVIDIA GPU + CUDA
• Python packages: pip install -r requirements.txt

Dataset Preparation

The datasets used in this paper can be downloaded in the following links.
[Training Dataset] [Testing Dataset]

Dataset Generation

We generate dataset via QuaDRiGa. To assist researchers in the field of channel prediction, we have provided a runnable demo file in the data_generation folder. For more detailed information about the QuDRiGa generator, please refer to its user documentation uadriga_documentation_v2.8.1-0.pdf.

Get Started

Training and testing codes are in the current folder.

• The code for training is in train.py, while the code for test is in test_tdd_full.py and test_fdd_full.py. we also provide our pretrained model in [Weights].

• For full shot training, you need to set the file_path in the main function to match your training dataset. For example, if you want to try a full-shot experiment in a TDD scenario, you need to modify the train_TDD_r_path and train_TDD_t_path in train.py to the locations of your downloaded H_U_his_train.mat and H_U_pre_train.mat, respectively. Then, you can run train.py.

• For few shot training, you need to set the file_path in the main function to match your training dataset. Then, you can set is_few=1 when creating the training set in train.py like this: train_set = Dataset_Pro(train_TDD_r_path, train_TDD_t_path, is_few=1) and run train.py.

• For testing, you also need to set the file_path in the main function to match your testing dataset. Then, you can run test_tdd_full.py to obtain the results in Figure 7 of the paper, and you can run test_fdd_full.py to obtain the results in Figure 8 of the paper. You can also try loading the data under Testing Dataset/Umi to test the models' zero-shot performance.

Citation

If you find this repo helpful, please cite our paper.

@article{liu2024llm4cp,
  title={LLM4CP: Adapting Large Language Models for Channel Prediction},
  author={Liu, Boxun and Liu, Xuanyu and Gao, Shijian and Cheng, Xiang and Yang, Liuqing},
  journal={arXiv preprint arXiv:2406.14440},
  year={2024}