DCentNet (Decentralized Signal Classification using Early Exit)

Experimental Code

1. Inference and Sender (BLE)

• Deep Sleep Mode:
  Code: DeepSleep(idle_mode).cpp

• Inference Mode:
  Code: bk_vtx001_noBLE_Infer_noidle.cpp

• Inference Idle Mode:
  Code: bk_vtx001_noBLE_Infer_idle.cpp

• Connected Mode:
  Code: bk_vtx001a_connected_noinfer_idle_keepconnected.cpp

• Broadcast Mode:
  Code: bk_v009a_2DenseOutput_Broadcast_realclass.cpp

2. Receiver (BLE)

• Connected Mode:
  Code: rec.cpp

• Broadcast Mode:
  Code: REC_Broadcast.cpp

---

Deploy in Embedded System

• Trained Model:
  Download: 2Layerswith96%Acc.zip

Steps for Deployment:

1. Launch Arduino IDE (Version used: 2.3.3).
2. Navigate to Sketch -> Include Library -> Add .ZIP Library -> (Add the above-trained model).
3. Go to Tools -> Manage Libraries -> Search for "ArduinoBLE" -> Install (Version used: 1.3.7).
4. Include the inference file in your code (see the code example under Experimental Code - Inference Mode).

Alternative Trained Models:

• EEPS-2LayerCNN.zip
• EEPS-3LayerCNN.zip
• EEPS-4LayerCNN.zip
• EEPS-5LayerCNN.zip

---

Training Code

• Simple Version:
  Code: eeps-train.ipynb

Instructions:

• Register on Edge Impulse and obtain an API key (free for personal use).
• Replace the placeholder in the training code with your API key.
• You can customize and replace the network for your specific use case.

Enjoy training your model!

---

Replicate Guide

• Support:
  If you have any questions, feel free to email me at bh.huang@ieee.org. I will address your questions and update the guide as needed.

---

Citation

• Support:

• If you find our article is useful to your research, please cite following papers:
  

@article{LI2025107468,
  title = {DCentNet: Decentralized multistage biomedical signal classification using early exits},
  journal = {Biomedical Signal Processing and Control},
  volume = {104},
  pages = {107468},
  year = {2025},
  issn = {1746-8094},
  doi = {https://doi.org/10.1016/j.bspc.2024.107468},
  url = {https://www.sciencedirect.com/science/article/pii/S174680942401526X},
  author = {Xiaolin Li and Binhua Huang and Barry Cardiff and Deepu John},
}