Overview

This is the PyTorch implementation of the ICC'23 paper JPTS:Enhancing Deep Learning Performance of Massive MIMO CSI Feedback. If you feel this repo helpful, please cite our paper:

@article{ji2022enhancing,
  title={Enhancing Deep Learning Performance of Massive MIMO CSI Feedback},
  author={Ji, Sijie and Li, Mo},
  journal={arXiv preprint arXiv:2208.11333},
  year={2022}
}

Requirements

To use this project, you need to ensure the following requirements are installed.

• Python >= 3.7
• PyTorch >= 1.2 also compatible with PyTorch >= 1.7 with fft fix
• thop

Project Preparation

A. Data Preparation

The channel state information (CSI) matrix is generated from COST2100 model. Chao-Kai Wen and Shi Jin group provides a pre-processed version of COST2100 dataset in Google Drive, which is easier to use for the CSI feedback task; You can also download it from Baidu Netdisk.

You can generate your own dataset according to the open source library of COST2100 as well. The details of data pre-processing can be found in our paper.

B. Project Tree Arrangement

We recommend you to arrange the project tree as follows.

home
├── JPTS  # The cloned CLNet repository
│   ├── dataset
│   ├── models
│   ├── utils
│   ├── main.py
├── COST2100  # The data folder
│   ├── DATA_Htestin.mat
│   ├── ...
├── Experiments
│   ├── checkpoints  # The checkpoints folder
│   │     ├── in_04.pth
│   │     ├── ...
│   ├── run.sh  # The bash script
...

Adpoting JPTS to SOTA methods

This repo provides three SOTA methods: CSINet, CRNet and CLNet. You can simply configure crnet.py to change the models and train it from the scratch. An example of run.sh is listed below. Simply use it with sh run.sh. Change scenario using --scenario and change compression ratio with --cr.

python /home/JPTS/main.py \
  --data-dir '/home/COST2100' \
  --scenario 'in' \
  --epochs 1000 \
  --batch-size 200 \
  --workers 8 \
  --cr 4 \
  --scheduler cosine \
  --gpu 0 \
  2>&1 | tee log.out

Results and Reproduction

The NMSE result reported in the paper as follow:

JPTS-CSINet:

Scenario	Compression Ratio	NMSE	Checkpoints
indoor	1/4	-24.19	CSINET_IN_4.pth
indoor	1/8	-15.20	CSINET_IN_8.pth
indoor	1/16	-10.65	CSINET_IN_16.pth
indoor	1/32	-8.59	CSINET_IN_32.pth
indoor	1/64	-6.26	CSINET_IN_64.pth
outdoor	1/4	-12.20	CSINET_OUT_4.pth
outdoor	1/8	-7.97	CSINET_OUT_8.pth
outdoor	1/16	-5.22	CSINET_OUT_16.pth
outdoor	1/32	-3.12	CSINET_OUT_32.pth
outdoor	1/64	-2.17	CSINET_OUT_64.pth

JPTS-CRNet:

Scenario	Compression Ratio	NMSE	Checkpoints
indoor	1/4	-26.84	CRNET_IN_4.pth
indoor	1/8	-16.32	CRNET_IN_8.pth
indoor	1/16	-11.55	CRNET_IN_16.pth
indoor	1/32	-8.98	CRNET_IN_32.pth
indoor	1/64	-6.50	CRNET_IN_64.pth
outdoor	1/4	-12.72	CRNET_OUT_4.pth
outdoor	1/8	-8.01	CRNET_OUT_8.pth
outdoor	1/16	-5.41	CRNET_OUT_16.pth
outdoor	1/32	-3.38	CRNET_OUT_32.pth
outdoor	1/64	-2.21	CRNET_OUT_64.pth

JPTS-CLNet:

Scenario	Compression Ratio	NMSE	Checkpoints
indoor	1/4	-28.38	CLNET_IN_4.pth
indoor	1/8	-16.03	CLNET_IN_8.pth
indoor	1/16	-12.16	CLNET_IN_16.pth
indoor	1/32	-9.00	CLNET_IN_32.pth
indoor	1/64	-6.86	CLNET_IN_64.pth
outdoor	1/4	-12.90	CLNET_OUT_4.pth
outdoor	1/8	-8.40	CLNET_OUT_8.pth
outdoor	1/16	-5.61	CLNET_OUT_16.pth
outdoor	1/32	-3.61	CLNET_OUT_32.pth
outdoor	1/64	-2.30	CLNET_OUT_64.pth

If you want to reproduce our result, you can directly download the corresponding checkpoints from OneDrive

To reproduce all these results, simple add --evaluate to run.sh and pick the corresponding pre-trained model with --pretrained. An example is shown as follows.

python /home/JPTS/main.py \
  --data-dir '/home/COST2100' \
  --scenario 'in' \
  --pretrained './checkpoints/in4.pth' \
  --evaluate \
  --batch-size 200 \
  --workers 0 \
  --cr 4 \
  --cpu \
  2>&1 | tee test_log.out

Acknowledgment

This repository is modified from the CLNet open source code & CRNet open source code. Thanks Chao-Kai Wen and Shi Jin group for providing the pre-processed COST2100 dataset, you can find their related work named CsiNet in Github-Python_CsiNet