Projects/STM32 EMG acquisition

EMG Acquisition & Monitoring System

STM32 firmware and a Python GUI that turn raw forearm EMG into stable RMS/MAV features at a nominal 4 kHz.

Role: Firmware & GUI
Platform: STM32F411CEU6
Sensor: DFRobot SEN0240
Stack: C · Python / Tkinter

EMG test setup: SEN0240 electrode band on a forearm wired to an STM32 board

01 / Problem

Problem

Our team built a rehabilitation-hand system for an undergraduate thesis. It needed a dependable way to acquire surface EMG (the electrical activity of muscles), turn it into stable features, and show the whole system's state to an operator.

My part was the EMG acquisition and signal-processing subsystem, plus the full monitoring interface. Aldin focused on the stimulator and Zain on fuzzy-control design.

02 / Approach

Approach

SEN0240sEMG sensor, analog 0–3.0 V

STM32 ADC12-bit, timer-triggered DMA, nominal 4 kHz

Digital filters20 Hz HPF · 500 Hz LPF · 50 Hz notch

FeaturesRMS & MAV over 400 samples (~100 ms)

UART → GUI115200 baud · Python/Tkinter · CSV

Firmware processes ADC samples in half/full DMA buffers, filters them and computes RMS/MAV per window, then streams telemetry over UART. The Tkinter GUI parses the serial stream, plots EMG features, monitors status and logs to CSV. Coding agents assisted the implementation, and I reviewed and tested what they produced.

03 / Build gallery

Build gallery

SEN0240 EMG electrode band on a forearm, wired to the acquisition board — Test setup: SEN0240 electrodes on the forearm, wired to the STM32 acquisition board.

Wiring diagram: SEN0240 VCC to 3V3, analog output to PA0 ADC1_IN0, ground to GND — Wiring: SEN0240 analog output to PA0 (ADC1_IN0) on the STM32F411CEU6.

Block diagram of the EMG acquisition and processing subsystem — Acquisition and processing pipeline from the thesis.

Frequency response plots of the cascaded filters and the 50 Hz notch — Designed filter response, including the 50 Hz notch.

Time-series plot exported from the EMG GUI showing RMS, EMA and baseline — Session data exported from the GUI.

04 / Results

Results

4 kHzNominal sampling (configured)

12-bitADC resolution

400Samples per feature window

10 / 10GUI self-test check groups passed

Line chart of filtered MAV and RMS during a 10-minute test — Filtered MAV and RMS during a 10-minute test.

RMS of EMG across four grip conditions, from relaxed to strong — RMS across four grip conditions.

Bar chart comparing MAV and RMS across three test repetitions — MAV and RMS across three test repetitions.

Welch PSD showing 50 Hz mains interference before filtering — Power spectrum showing 50 Hz mains interference, which the notch filter targets.

Scatter plot of filtered MAV versus RMS with a linear fit — Relationship between filtered MAV and RMS.

The plots are from my thesis testing. Sampling rate and window size are firmware configuration. The GUI self-test checks the parser, safety logic and logger. It does not demonstrate clinical validity, rehabilitation effectiveness or full device safety.

05 / Lessons

Lessons

Breaking a sensor-to-screen problem into acquisition, processing, transport and display made each part testable.
Mains interference at 50 Hz is visible in the spectrum, so the notch filter earns its place.
Program configuration and measured results are separate claims, and I keep them separate.
AI assistance is useful when I read and check what it produces. I disclose it.

Original contribution notes (Bahasa Indonesia)