Towards Real-Time ECG and EMG Modeling on $\mu$NPUs

TL;DR

PhysioLite is a lightweight, NPU-compatible model architecture for ECG and EMG analysis that achieves near-state-of-the-art performance on constrained hardware, enabling real-time signal processing on microcontrollers.

Contribution

The paper introduces PhysioLite, a novel, hardware-aware model architecture and training framework optimized for low-power microcontroller units for physiological signal analysis.

Findings

PhysioLite achieves comparable accuracy to large Transformer models on ECG and EMG benchmarks.

The model size is less than 10% of state-of-the-art Transformer models, around 370KB with 8-bit quantization.

Latency and resource profiling demonstrate PhysioLite's suitability for real-time processing on microcontrollers.

Abstract

The miniaturisation of neural processing units (NPUs) and other low-power accelerators has enabled their integration into microcontroller-scale wearable hardware, supporting near-real-time, offline, and privacy-preserving inference. Yet physiological signal analysis has remained infeasible on such hardware; recent Transformer-based models show state-of-the-art performance but are prohibitively large for resource- and power-constrained hardware and incompatible with $\mu$NPUs due to their dynamic attention operations. We introduce PhysioLite, a lightweight, NPU-compatible model architecture and training framework for ECG/EMG signal analysis. Using learnable wavelet filter banks, CPU-offloaded positional encoding, and hardware-aware layer design, PhysioLite reaches performance comparable to state-of-the-art Transformer-based foundation models on ECG and EMG benchmarks, while being <10% of the size ($\sim$370KB with 8-bit quantization). We also profile its component-wise latency and resource consumption on both the MAX78000 and HX6538 WE2 $\mu$NPUs, demonstrating its viability for signal analysis on constrained, battery-powered hardware. We release our model(s) and training framework at: https://github.com/j0shmillar/physiolite.