Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Reconstruction of Surface EMG Signal using IMU data for Upper Limb Actions

Shubhranil Basak, Mada Hemanth, Madhav Rao

TL;DR

This work addresses reconstructing normalized sEMG signals from wearable 6-axis IMU data to enable muscle-intent detection without direct EMG sensing. It introduces a sliding-window WaveNet architecture with dilated causal convolutions and a context-aggregation layer to map IMU motion to EMG envelopes, trained on synchronized sEMG-IMU data collected at 1 kHz from two subjects performing upper-limb motions. The model achieves high temporal fidelity in activation timing but underestimates peak amplitudes and exhibits limited high-frequency content replication, highlighting a trade-off between timing precision and spectral detail. The approach offers a viable path toward real-time muscle activity monitoring for prosthetics and rehabilitation, with future work aimed at improving amplitude accuracy, generalization across a broader cohort, and real-time deployment feasibility.

Abstract

Surface Electromyography (sEMG) provides vital insights into muscle function, but it can be noisy and challenging to acquire. Inertial Measurement Units (IMUs) provide a robust and wearable alternative to motion capture systems. This paper investigates the synthesis of normalized sEMG signals from 6-axis IMU data using a deep learning approach. We collected simultaneous sEMG and IMU data sampled at 1~KHz for various arm movements. A Sliding-Window-Wave-Net model, based on dilated causal convolutions, was trained to map the IMU data to the sEMG signal. The results show that the model successfully predicts the timing and general shape of muscle activations. Although peak amplitudes were often underestimated, the high temporal fidelity demonstrates the feasibility of using this method for muscle intent detection in applications such as prosthetics and rehabilitation biofeedback.

Reconstruction of Surface EMG Signal using IMU data for Upper Limb Actions

TL;DR

This work addresses reconstructing normalized sEMG signals from wearable 6-axis IMU data to enable muscle-intent detection without direct EMG sensing. It introduces a sliding-window WaveNet architecture with dilated causal convolutions and a context-aggregation layer to map IMU motion to EMG envelopes, trained on synchronized sEMG-IMU data collected at 1 kHz from two subjects performing upper-limb motions. The model achieves high temporal fidelity in activation timing but underestimates peak amplitudes and exhibits limited high-frequency content replication, highlighting a trade-off between timing precision and spectral detail. The approach offers a viable path toward real-time muscle activity monitoring for prosthetics and rehabilitation, with future work aimed at improving amplitude accuracy, generalization across a broader cohort, and real-time deployment feasibility.

Abstract

Surface Electromyography (sEMG) provides vital insights into muscle function, but it can be noisy and challenging to acquire. Inertial Measurement Units (IMUs) provide a robust and wearable alternative to motion capture systems. This paper investigates the synthesis of normalized sEMG signals from 6-axis IMU data using a deep learning approach. We collected simultaneous sEMG and IMU data sampled at 1~KHz for various arm movements. A Sliding-Window-Wave-Net model, based on dilated causal convolutions, was trained to map the IMU data to the sEMG signal. The results show that the model successfully predicts the timing and general shape of muscle activations. Although peak amplitudes were often underestimated, the high temporal fidelity demonstrates the feasibility of using this method for muscle intent detection in applications such as prosthetics and rehabilitation biofeedback.

Paper Structure

This paper contains 17 sections, 3 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Data Acquisition
  3. Data Pre-processing and Segmentation
  4. Filtering
  5. Segmentation
  6. Data merge
  7. Proposed Model Architecture
  8. Network Overview
  9. Input Layer
  10. Dilated Causal Convolution Blocks
  11. Context Aggregation Layer
  12. Output Layer
  13. Results
  14. Qualitative Analysis
  15. Quantitative Analysis
  16. ...and 2 more sections

Figures (5)

  • Figure 1: Snapshot of an individual performing Bicep Curl Exercise.
  • Figure 2: Segmented Plot for seven sets of bicep curl
  • Figure 3: Model Block Diagram
  • Figure 4: Dilation Block
  • Figure 5: Synthesis of Normalized sEMG signal (for Bicep Curl) and its corresponding envelope waveform.