Compressed Sensing for Wearable ECG and EMG

ISEF Category: Embedded Systems

Subcategory: Signal Processing  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Your smartwatch cannot afford to record every heartbeat forever. Battery drains fast, and medical wearables need long runtime. Compressed sensing tries to fix that by recording fewer samples and rebuilding the signal later. You can test how much detail survives that shortcut.

What Is It?

Compressed sensing is a way to capture a signal with fewer measurements than normal, then reconstruct the original shape with math. Think of it like hearing only some notes from a song and using structure in the music to fill in the rest. For ECG and EMG, the goal is to keep the important wave patterns even when the sensor samples far below the usual rate.

ECG measures electrical activity from the heart. EMG measures electrical activity from muscles. Both signals have patterns that are not random, so reconstruction algorithms can often recover them better than you might expect. Your project can test how well that works on a microcontroller, and how much battery life the lower sampling rate could save.

Why This Is a Good Topic

This topic works well because you can change one clear variable, the sampling rate, and measure real outputs like reconstruction error, waveform shape, and estimated power savings. It connects to wearable health tech, remote monitoring, and medical devices that need long battery life. You can learn signal processing, evaluation metrics, and embedded implementation choices without needing a wet lab. A strong project can compare different reconstruction methods or signal types and show where the method breaks down.

Research Questions

• How does sampling at 1/10× Nyquist affect ECG reconstruction error compared with full-rate sampling?
• What is the effect of reconstruction algorithm choice on EMG signal fidelity at low sampling rates?
• To what extent does noise level change compressed-sensing performance for wearable biosignals?
• Which sparsifying transform gives the best reconstruction for ECG signals on an MCU?
• How does record length affect reconstruction quality and estimated battery savings?
• Does compressed sensing preserve clinically relevant ECG features better than simple downsampling?
• To what extent can fixed-point MCU code match floating-point reconstruction quality?

Basic Materials

• Microcontroller board with ADC support, such as an Arduino or ESP32.
• ECG or EMG sensor kit for educational use.
• Electrodes and lead wires.
• Computer for coding and data analysis.
• USB cable and power source for the microcontroller.
• Breadboard and jumper wires.
• Notebook or spreadsheet for tracking test runs.
• Public ECG or EMG dataset from PhysioNet for comparison.

Advanced Materials

• Microcontroller with stronger processing support, such as an STM32, ESP32-S3, or similar MCU.
• Reference ECG or EMG acquisition hardware.
• Biomedical signal acquisition electrodes and shielded cables.
• Oscilloscope or logic analyzer for timing and power checks.
• Current measurement tool, such as a USB power meter or bench supply with readout.
• Development board or breakout for low-power measurements.
• Access to MATLAB, Python, or embedded C toolchain.
• Datasets from PhysioNet for offline benchmarking.

Software & Tools

• Python: Runs reconstruction experiments, error metrics, and plots.
• NumPy: Handles arrays and matrix operations for signal processing.
• SciPy: Provides filtering, transforms, and optimization helpers.
• MATLAB: Tests reconstruction methods and compares algorithms quickly.
• ImageJ: Not needed for this topic, so skip it and focus on signal analysis tools.

Experiment Steps

1. Define the exact signal family you will test, such as ECG, EMG, or both.
2. Choose one reconstruction method first, then decide what makes it fair to compare against a full-rate baseline.
3. Build a metric set that includes waveform error, feature retention, and estimated power savings.
4. Plan a data pipeline that separates training, tuning, and final testing so you do not overfit your results.
5. Design controls that test noise, motion artifacts, and different sparsity assumptions.
6. Decide whether you will evaluate the method on raw data, MCU code, or both, then document that tradeoff clearly.

Common Pitfalls

• Comparing a reconstructed signal to a filtered version of the same signal, which hides real reconstruction error.
• Using public datasets with different sampling rates or sensor setups, which makes results hard to compare.
• Testing only one ECG or EMG record, which gives a false sense of performance.
• Ignoring feature-level metrics, which can miss lost QRS peaks, R peaks, or burst timing.
• Treating software speed as battery life, which skips the real cost of ADC duty cycle, memory, and MCU compute.

What Makes This Competitive

A strong project would go past a basic reconstruction demo and ask where the method actually helps a wearable. You could compare more than one biosignal, more than one algorithm, or more than one MCU implementation style. Better yet, tie signal quality to a real device tradeoff, like power budget, memory use, and clinically relevant feature loss. Careful validation on independent data sets would make the project much stronger.

Project Variations

• Test compressed sensing on ECG signals from exercise data instead of resting data, then compare motion artifact sensitivity.
• Swap EMG for respiration or pulse oximetry waveforms to see whether the same sparse reconstruction idea still works.
• Compare floating-point reconstruction with fixed-point MCU code to measure the accuracy and speed tradeoff.

Learn More

• PhysioNet: Search for ECG and EMG databases, sample records, and published benchmark papers.
• NIH PubMed: Search for review articles on compressed sensing in biomedical signal processing.
• MIT OpenCourseWare: Search for signals and systems course materials that cover sampling, transforms, and reconstruction.
• NASA Open Science Data Repository: Browse examples of signal analysis workflows and data handling practices.
• IEEE Xplore: Search for recent peer-reviewed papers on wearable biosignal compression and low-power sensing.