This repository contains an implementation of the graph neural network (GNN)-enhanced expectation propagation (EP) for MIMO turbo receivers [1], [2], which extends the GNN-enhanced EP detector (GEPNet) [3] into iterative detection and decoding (IDD).

• Python (>= 3.6)
• Tensorflow (>=2.3.0)
• numpy (>=1.18.5)
• scipy (>=1.4.1)

• Python folder: for training network models, which are saved in the 'model' subdirectory
• Matlab folder (preparing): for performing inference, i.e., iterative turbo receiving

0. Training and testing the uncoded GEPNet [3] may be helpful for getting started.
1. Run Python codes to train network models, including training APP-based GEPNet, generating extrinsic training LLR datasets (saved in the 'dataset' subdirectory), and training EXT-GEPNet (see [1] and main.py to learn the three-step training in detail).
2. Saved the models in the 'model' subdirectory.
3. Run Matlab codes to perform EXT-GEPNet-based iterative turbo receiving with the pre-trained weights in the 'model' subdirectory.

• Matlab folder to be added.
• More detailed comments in the source code and steps to carry out the simulations.

[1] X. Zhou, J. Zhang, C.-K. Wen, S. Jin, and S. Han, “Graph neural network-enhanced expectation propagation algorithm for MIMO turbo receivers,” to appear in IEEE Transactions on Signal Processing ([Online] Available: https://arxiv.org/abs/2308.11335).

[2] X. Zhou, J. Zhang, C.-K. Wen, and S. Jin, “Extrinsic graph neural network-aided expectation propagation for Turbo-MIMO receiver,” in Proc. 18th Int. Symp. Wireless Commun. Syst. (ISWCS), Hangzhou, China, Oct. 2022, pp. 1–6.

[3] A. Kosasih, V. Onasis, V. Miloslavskaya, W. Hardjawana, V. Andrean, and B. Vucetic, “Graph neural network aided MU-MIMO detectors,” IEEE J. Sel. Areas Commun., vol. 40, no. 9, pp. 2540–2555, Sep. 2022.

If you have any questions or comments about this work, please feel free to contact [email protected]