NeuroComm-MSNN

Neuromorphic Wireless Split Computing with Multi-Level Spikes

This repository implements Neuromorphic Wireless Split Computing using Multi-Level Spikes. The code is based on SNNCutoff.

---

Dataset Preprocessing: DVS128 Gesture

1. Download the dataset from Dropbox.
2. Use the provided preprocessing code to prepare the dataset for training.

---

Getting Started

1. Install PyTorch and other dependencies:

   pip install -r requirements.txt

---

Training

Noiseless Training

Run the following script for noiseless training:

sh scripts/DirectTraining/tet/training_neurocomm_noiseless.sh

End-to-End Training with Sampled Channels

For end-to-end training using the sampled training channel (channel.npz), run:

sh scripts/DirectTraining/tet/training_neurocomm_emu.sh

Note: To reduce the file size, the channel.npz file represents a subset of the channel used in the original paper.

Evaluation

Run the following script for evaluation (digital):

sh scripts/DirectTraining/tet/graded_spike_evaluation_power.sh

---

Citation

For more details, please refer to the paper.

@article{wu2025neuromorphic,
  title={Neuromorphic Wireless Split Computing with Multi-Level Spikes},
  author={Wu, Dengyu and Chen, Jiechen and Rajendran, Bipin and Poor, H Vincent and Simeone, Osvaldo},
  journal={IEEE Transactions on Machine Learning in Communications and Networking},
  year={2025},
  publisher={IEEE}
}