Authors:
Anirban Mukherjee, Praneeth Susarla , Pravallika Katragunta , S. S. Krishna Chaitanya Bulusu, Olli Silven , Markku Juntti, Dinesh Babu Jayagopi, and Miguel Bordallo Lopez
📄 Published in: 2025 IEEE 5th International Symposium on Joint Communications & Sensing (JC&S)
🔗 Paper Link: IEEE Xplore
This repository contains the source code and training scripts for the models presented in our research paper titled "Multi-domain CSI-based Indoor Localization with Deep Attention Networks (DAN) for MIMO JCAS system".
We propose a deep learning-based approach to indoor localization using Channel State Information (CSI) from multiple domains — Complex and Polar representations across different subcarrier groups (domains). Our final model utilizes a deep attention-based architecture (DAN) to effectively fuse and learn from multi-domain CSI data.
| Model No. | Input Domains | CSI Representation | Description |
|---|---|---|---|
| Model 1 | Domain 1 + 2 | Complex | |
| Model 2 | Domain 1 + 2 | Polar | |
| Model 3 | Domain 1 + 2 | Complex + Polar | |
| Model 4 | Domain 2 + 3 | Complex | |
| Model 5 | Domain 2 + 3 | Polar | |
| Model 6 | Domain 2 + 3 | Complex + Polar | |
| Model 7 | Domain 1 + 3 | Complex | |
| Model 8 | Domain 1 + 3 | Polar | |
| Model 9 | Domain 1 + 3 | Complex + Polar | |
| Model 10 | Domain 1 + 2 + 3 | Complex | |
| Model 11 | Domain 1 + 2 + 3 | Polar | |
| Model 12 | Domain 1 + 2 + 3 | Complex + Polar | |
| Model 13 | Domain 1 (C) + 2 (C+P) | Complex + Polar | Benchmark model |
| Model 14 | Domain 1 + 2 + 3 | Complex + Polar | DAN (Final proposed model) |
├── CSI_Localization_Multidomain_Data.ipynb # Main Jupyter notebook with model training and evaluation
└── README.md # Project documentation (this file)
-
tensorflow==2.15
To train and evaluate the models:
jupyter notebook CSI_Localization_Multidomain_Data.ipynb- The models are evaluated based on localization accuracy using multi-domain CSI features.
- Final performance comparisons between the 14 models are included in the paper.
- Introduced a deep attention network (DAN) to fuse multi-domain CSI data effectively.
- Benchmarked 13 different combinations of domain and representation to validate the effectiveness.
- Demonstrated significant improvements in indoor localization accuracy using our DAN model.
If you find this work helpful, please consider citing our paper:
IEEE Xplore: https://ieeexplore.ieee.org/document/10880643
DOI: 10.1109/ICASSP48485.2024.10880643
@inproceedings{mukherjee2025multi,
title={Multi-domain CSI-based Indoor Localization with Deep Attention Networks for MIMO JCAS system},
author={Mukherjee, Anirban and Susarla, Praneeth and Katragunta, Pravallika and Bulusu, SS Krishna Chaitanya and Silven, Olli and Juntti, Markku and Jayagopi, Dinesh Babu and Lopez, Miguel Bordallo},
booktitle={2025 IEEE 5th International Symposium on Joint Communications \& Sensing (JC\&S)},
pages={1--6},
year={2025},
organization={IEEE}
}