EMG

EMG Deep Learning BCI Controller

Biomedical EMG/ECG Acquisition Circuit, Keras Deep Learning Classifier, and Wireless Robotic Actuation

Overview

EMG Deep Learning BCI Controller is an end-to-end biomedical engineering and Human-Machine Interface (HMI / BCI) system. It acquires biological biopotential signals (Electromyography - EMG and Electrocardiography - ECG), processes and extracts spectral harmonic features, performs real-time neural network classification using TensorFlow / Keras, and wirelessly triggers robotic actuators and mobile vehicles.

The system encompasses custom analog hardware design in KiCad (instrumentation amplification, bandpass filtering, 60 Hz notch filtering, and envelope detection), ESP32 firmware streaming digitized ADC data via WebSockets and Zeroconf (mDNS), and a sleek CustomTkinter dark-mode operator console.

graph TD
    A["Surface Electrodes (Ag/AgCl)"] --> B["Analog Front-End PCB (INA + Active Filters + Notch 60Hz + Envelope)"]
    B --> C["ESP32 Dual-Channel ADC (12-bit, DMA/ISR Sampling)"]
    C -->|"WebSockets / mDNS Zeroconf (_esp32emgdual._tcp.local)"| D["Python Desktop Host (CustomTkinter)"]
    D --> E["Feature Extraction (FFT Fundamental + H2/H3/H4 Harmonics + Envelope)"]
    E --> F["TensorFlow / Keras Neural Network (.keras Model)"]
    F -->|"Class: Inactive / Forearm Active / Gastrocnemius Active"| G{"Intention Decision Engine"}
    G -->|"Actuation Packets over Wi-Fi"| H["ESP32 Motor Driver Node / Robotic Car"]
    D --> I["Real-time Signal Visualization & CSV Data Logger"]

Key Features

• Custom Biomedical Hardware (KiCad):
  • High CMRR instrumentation amplifier stage (INA128 / AD620 architecture).
  • Multi-stage active bandpass filter (20 Hz - 500 Hz EMG bandwidth) and 60 Hz twin-T notch filter.
  • Precision precision half-wave rectifier and low-pass envelope detector.
  • Separate dedicated PCB designs for EMG and ECG (Electrocardiography).
• Embedded Telemetry & Network Discovery:
  • High-frequency dual-channel ADC sampling on ESP32.
  • Zero-configuration networking (mDNS Zeroconf) and low-latency WebSocket communication.
• Deep Learning Classification (Keras / TensorFlow):
  • Feature extraction in both time domain (RMS, peak envelope) and frequency domain (FFT fundamental frequency, 2nd, 3rd, and 4th harmonic magnitudes).
  • Trained Deep Neural Network (modelo_emg.keras, modelo_emg_con_pin.keras) classifying muscle activation states in 50 ms sliding inference windows.
• Modern Desktop UI & Robotic Control:
  • Dark-mode operator dashboard built with CustomTkinter.
  • Real-time voltage waveform plots, connection status monitor, and live prediction display.
  • Wireless control of differential mobile robot chassis (ESP32-Carrito).

Technology Stack

Repository Structure

emg-deeplearning-bci-controller/
 src/
    gui/                        # Desktop interfaces
       emg_controller_motor_app.py # Main CustomTkinter UI + TF inference + motor control
       emg_gui_app.py          # Standalone monitoring GUI
       emg_gui_con_pines.py    # Pin-specific acquisition GUI
    deep_learning/              # Neural network models and inference
       inference_keras_model.py
       threshold_classifier.py
       modelo_emg.keras
       modelo_emg_con_pin.keras
    signal_processing/          # Data logging and exploration
        emg_signal_logger.py
        emg_data_exploration.ipynb
 firmware/
    esp32_emg_acquisition/      # ESP32 EMG sampling firmware (Single/Dual channel)
    esp32_robotic_car/          # ESP32 motor receiver firmware
 hardware/
    kicad_emg/                  # Complete KiCad project for EMG analog front-end
    kicad_ecg/                  # Complete KiCad project for ECG front-end
    schematics/                 # PDF electrical schematics (V2.pdf)
 docs/                           # Lab presentation and biomedical documentation
 requirements.txt
 .gitignore
 LICENSE
 README.md

Installation & Usage

1. Setup Environment

git clone https://github.com/JuanIDK201/emg-deeplearning-bci-controller.git
cd emg-deeplearning-bci-controller

python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt

2. Launching the EMG Controller & Motor Interface

python src/gui/emg_controller_motor_app.py

Author

Juan Garzón
Biomedical & Embedded Systems Engineer

• GitHub: @JuanIDK201

License

This project is licensed under the MIT License - see the LICENSE file for details.