This repository contains the implementation for our paper "ChannelMAE: Self-Supervised Learning Assisted Online Adaptation of Neural Channel Estimators" submitted to INFOCOM 2026.
ChannelMAE/
├── cebed/ # Main package
│ ├── models/ # Base model implementations
│ │ └── mae_random_mask.py # Random mask MAE
│ ├── models_with_ssl/ # Self-supervised models
│ │ └── recon_net.py # Main ChannelMAE implementation
│ ├── datasets_with_ssl/ # Dataset implementations
│ ├── online_ttt_v3.py # Online adaptation scripts
│ ├── online_adapt_*.py # Baseline adaptation scripts
│ └── ...
├── hyperparams/ # Model hyperparameter configs
├── scripts/ # Utility scripts
│ └── online_ttt.py # Online TTT execution script
├── runs/ # Training and evaluation scripts
└── ray_tracing_data/ # Ray tracing simulation data
The experimental results and figures referenced in the paper are in: plot_figs/
- Pretraining ChannelMAE:
cd runs
bash pretrain.sh- Online adaptation and evaluation:
bash ttt.sh- Baseline evaluation:
bash eval_baselines.shThis project is licensed under the same terms as the original CeBed project. Please refer to the CeBed repository for license details.
This codebase is built upon the excellent foundation provided by the CeBed project. We thank the original authors for their valuable contribution to the wireless communication community.