ECG Arrhythmia Classification on an Ultra-Low-Power Microcontroller.

Area of Science:

• Biomedical Engineering
• Embedded Systems
• Cardiology

Background:

• Wearable biomedical systems enable continuous patient monitoring for detecting rare events like cardiac arrhythmias.
• Modern systems integrate signal processing to identify events and reduce data transmission.
• Existing solutions often lack a complete, integrated beat-to-beat classification system on a single chip.

Purpose of the Study:

• To develop and document the first complete beat-to-beat arrhythmia classification system on a custom ultra-low-power microcontroller.
• To integrate analog front-end (AFE) for electrocardiogram (ECG) acquisition and a digital back-end (DBE) for classification.
• To demonstrate the system's feasibility and performance in a real-world prototype.

Main Methods:

• Designed a single-channel ECG AFE with a low-noise instrumentation amplifier (1.4 μW, 0.9 μV RMS noise).
• Implemented an all-digital time-based Analog-to-Digital Converter (ADC) with 10-bit resolution and 250-Hz bandwidth.
• Integrated a Support Vector Machine (SVM) classification algorithm on a Cortex-M4 CPU within the DBE.

Main Results:

• The AFE instrumentation amplifier achieved 1.4 μW power consumption and 0.9 μV RMS input-referred noise.
• The digital ADC occupied only 900 μm² and provided 10-bit effective resolution.
• The SVM classifier achieved 82.6% sensitivity for supraventricular and 88.9% for ventricular arrhythmias on the MIT-BIH database.

Conclusions:

• A complete beat-to-beat arrhythmia classification system was successfully implemented on an ultra-low-power microcontroller.
• The prototyped system on the SleepRider SoC demonstrated high classification accuracy and ultra-low power consumption (13.1 μW).
• This technology advances wearable biomedical systems for continuous, efficient cardiac monitoring.