Skip to content
Research Prototype

Parkinson's Tremor-Suppression EMG Cuff

Wearable system for real-time detection of Parkinsonian tremor using forearm EMG signals. Built Python simulation to validate signal processing approach and suppression triggers before hardware integration.

Role Lead developer, signal processing
Timeline Sep – Dec 2024
Type Course project / Research prototype
Standards IEC 60601-1-2, ISO 10993

The clinical need

Parkinson's disease affects approximately 1 million people in the U.S., with resting tremor (4–8 Hz) being one of the most disabling symptoms. Tremor impairs fine motor tasks: writing, eating, grasping objects. This severely reduces quality of life. Current treatment options are limited. Pharmacological interventions have variable efficacy and side effects, while deep brain stimulation (DBS) is invasive, costly, and not accessible to all patients.

Non-invasive wearable solutions that can detect tremor onset in real time and trigger suppression mechanisms remain underdeveloped. A wearable EMG cuff capable of identifying tremor timing could enable counter-stimulation or mechanical dampening strategies, offering patients a low-risk intervention option.

The Challenge

Detection complexity

Parkinsonian tremor occurs at 4–8 Hz in forearm musculature. Detecting this narrow frequency band in real time from EMG requires robust signal processing. Noise, motion artifacts, and physiological variation complicate detection. Hardware-first approaches risk wasting time on non-viable signal chains.

The Approach

Simulation-driven validation

Build a complete Python signal processing pipeline modeling the EMG flow through bandpass filter, feature extraction, and detection algorithm. Simulate tremor dynamics at known frequencies. Validate detection sensitivity and latency before any hardware fabrication. De-risk the hardware build.

System architecture

The wearable system follows a classical real-time biomedical signal processing flow. EMG sensors on the forearm capture muscle electrical activity, which is then processed through a tight filtering and feature extraction pipeline to identify tremor-specific signatures.

Signal Processing Pipeline
EMG Sensors Forearm surface electrodes
Bandpass Filter 4–8 Hz tremor band
Feature Extraction RMS, spectral power, coherence
Tremor Detection Threshold + coherence check
Suppression Trigger Counter-stim / dampening

Design decisions

Bandpass filter centered at 6 Hz. Parkinsonian tremor peaks around 4–8 Hz. A Butterworth bandpass filter (6th order, 4–8 Hz) isolates tremor signatures while rejecting DC drift, low-frequency movement, and high-frequency noise. Real-time latency is critical. We target less than 100 ms end-to-end detection.

Multi-feature detection, not single threshold. RMS energy alone generates false positives from voluntary muscle activation. We combine RMS, spectral power concentration, and coherence across electrode channels. A tremor detection event requires all three features to align, raising specificity without sacrificing sensitivity.

Python simulation before hardware. We modeled tremor dynamics (sinusoidal + noise) and stress-tested the pipeline against realistic EMG corruption. Only after validating detection accuracy on simulated data did we move toward hardware fabrication. This iterative de-risking prevented costly false starts.

Interactive: See the filter in action

This visualization shows a live EMG signal being processed through a 6th-order Butterworth bandpass filter. Watch how the filter isolates the 5.8 Hz tremor signal from power line interference and motion artifacts.

This is a simplified simulation. The real device uses a 6th-order Butterworth bandpass filter at 4 to 8 Hz.

N/A
Detected Frequency
1.2
Signal-to-Noise Ratio
Too Noisy
Classification

Regulatory and safety framework

This wearable medical device concept must adhere to international standards for medical device safety, biocompatibility, and electromagnetic compatibility. Understanding regulatory pathways early shapes architecture choices and validation protocols.

IEC 60601-1-2

Electromagnetic compatibility (EMC) for medical electrical equipment. Critical for wearables operating near body and in varying EM environments.

ISO 10993 Series

Biocompatibility of medical devices. Electrode materials and skin-contact components must meet cytocompatibility and irritation thresholds.

IEC 60601-1 General

Safety requirements for medical electrical equipment. Risk analysis, electrical safety, and design controls.

FDA De Novo Pathway

For novel wearable neurotech devices without predicate. Requires comprehensive design history file and clinical validation.

Signal processing targets

Validation metrics measure both the accuracy of our tremor detection algorithm and the real-time feasibility of the system.

4–8Hz
Tremor band targeted
<100ms
Detection latency target
95%
Simulated sensitivity
6
Filter order (Butterworth)

Implementation stack

Python scipy.signal numpy matplotlib EMG Processing Signal Filtering Feature Extraction Real-time Simulation

Technology rationale

scipy.signal: Production-grade filtering (Butterworth, Chebyshev, Bessel). We compared multiple designs for phase distortion and attenuation characteristics before settling on 6th-order Butterworth, offering excellent rolloff without excessive phase lag.

numpy for vector operations: EMG processing is array-heavy. numpy's C-backend ensures real-time throughput even with 1+ kHz sampling rates across multiple channels.

matplotlib for validation: Time-domain waveforms, FFT spectrograms, and filter response plots proved critical for debugging detection failures and tuning thresholds iteratively.

What this taught me

01

Simulating before building saves months

A week of Python modeling revealed that naive single-channel thresholding would produce unacceptable false positive rates. Multi-channel coherence checks added complexity but eliminated the problem. Hardware teams found this out the hard way. We caught it in simulation.

02

Biomedical signals are hostile

Textbooks show clean sinusoids. Real EMG is fractional-volt noise contaminated with motion artifacts, ECG cross-talk, and 60 Hz hum. Every design choice (filter order, feature set, threshold) had to survive adversarial noise tests before acceptance.

03

Regulatory thinking must start early

EMC, biocompatibility, and design controls are not afterthoughts. Building regulatory awareness into the signal processing design (choosing components with certified datasheets, documenting filter design rationale) streamlines later validation.

Interested in this work?

I bring signal processing rigor, medical device regulatory depth, and a bias toward validation-before-build to every project. Looking for roles in wearable biomedtech, diagnostics, or neural interfaces.