Low-Cost Single-Lead ECG Front-End Trainer with Noise and Filtering Study

ECG-Trainer is a build-and-learn instrument for the biomedical engineering lab: a low-cost single-lead electrocardiograph front end, plus the measurement study that makes it a project rather than a kit assembly. An AD8232 single-lead analogue front end on a breakout board picks up a lead-I ECG through three disposable gel electrodes worn by a team volunteer. An Arduino Uno samples the conditioned waveform at 500 Hz, writes raw samples to a microSD card and streams them over serial, so every recording is available afterwards for offline analysis.

The study is the heart of the bundle. Three artefacts dominate classroom ECG traces: baseline wander (respiration and electrode motion), 50 Hz powerline pickup (Indian mains couples capacitively into lead wires) and muscle (EMG) noise. The team provokes each artefact deliberately, quantifies signal-to-noise ratio in dB before and after treatment, and compares the AD8232's onboard analogue band-pass filtering against digital cures implemented in Python/SciPy and Scilab — a 0.5–40 Hz Butterworth band-pass and a 50 Hz IIR notch.

The rig is explicitly a teaching prototype, not a medical device: it claims no IEC 60601 compliance, never touches a patient connection beyond the team's own volunteers, and the report documents what a real product would need (patient isolation, leakage-current limits, defibrillation protection) as an awareness study. C firmware runs the Arduino (the VTU 2022 scheme teaches C and instrumentation labs), Python and Scilab handle analysis, and the oscilloscope and function generator already in the lab serve as references.

Biomedical engineering students learn ECG instrumentation from textbook figures because real trainers and patient simulators cost tens of thousands of rupees. As a result, they never experience the three artefacts that dominate real traces — baseline wander from respiration and motion, 50 Hz powerline pickup, and muscle noise — and never measure how well analogue front-end filtering and digital filtering actually work against them. Ready-made sensor kits are affordable, but assembling one and shelving it proves nothing measurable: there is no gain calibration, no quantified SNR before and after filtering, and no protocol that turns the hardware into defensible experimental evidence.

There is a need for a low-cost, well-characterised ECG front-end trainer with a written calibration and provocation protocol, an offline analysis pipeline that reports SNR and R-peak detection figures objectively, and an explicit prototype-not-product framing that teaches the safety obligations (patient isolation, leakage current, IEC 60601-1) a real instrument would carry — turning a ₹2,500 kit into a viva-defensible measurement study.

Abstract

Biopotential measurement is a core subject in biomedical engineering, yet most departments teach it from textbook figures because a commercial ECG trainer or simulator costs far more than a lab budget allows. This project builds a single-lead ECG front-end trainer for about ₹2,700: an AD8232 analogue front end with three gel electrodes feeds an Arduino Uno that logs 500 Hz raw samples to microSD and serial. The team then runs a controlled noise-and-filtering study on real traces — quantifying baseline wander, 50 Hz mains interference and muscle artefact by SNR in dB — and compares the AD8232's onboard analogue band-pass against digital Butterworth and notch filters in Python/SciPy. Ten volunteer recordings, a gain calibration and a written provocation protocol form the evidence base.

Introduction

The electrocardiogram is a millivolt-level signal (0.5–4 mV peak) buried in interference, which is why ECG instrumentation is the classic vehicle for teaching instrumentation amplifiers, common-mode rejection and filtering together. Indian classrooms add a practical twist: 50 Hz mains, fans and unshielded bench wiring make the powerline artefact visible on almost every unguarded trace, which makes the filtering study motivating rather than artificial.

Existing System

• Textbook waveforms: safe, but students never see real artefacts or electrode-skin effects.
• Commercial ECG patient simulators (₹40,000 and above): clean synthetic output, no real skin-electrode interface, out of budget.
• Ready-made AD8232 kits assembled once and shelved: hardware exists but is rarely turned into a quantified experiment — no SNR numbers, no filter comparison, no protocol.
• Limitations: no calibration evidence and no discussion of why such a rig must not be called a medical device.

Proposed System

• A documented, reproducible front-end trainer: AD8232 breakout (instrumentation amplifier, right-leg drive, onboard 0.5–40 Hz band-pass), three-electrode lead-I placement, Arduino Uno sampling at 500 Hz with timestamped raw logging to SD.
• A calibration step with the lab function generator: overall gain and frequency response measured with a 1 mV sine sweep, tabulated against the datasheet.
• A provocation protocol: motion, 50 Hz pickup, muscle clench and electrode dry-out — each provoked by a written script.
• A digital comparison: Butterworth band-pass, 50 Hz IIR notch and R-peak detection before and after filtering; SNR and detection sensitivity computed in Python and cross-checked in Scilab.
• An honesty layer: consent-and-safety protocol, battery or USB-isolated power during recordings, and an IEC 60601-1 awareness section stating why the prototype is not a clinical device.

Feasibility Study

• Technical: every block is testable with lab instruments; the AD8232 datasheet gives the transfer functions, SciPy/Scilab provide the filter designs, and the highest circuit voltage is 5 V DC.
• Economic: the bill of materials is about ₹2,700 one-time, with consumable electrodes at ₹350 per pack of ten — cheaper than a week of simulator rental.
• Operational: recordings take five minutes per volunteer under the written protocol and analysis runs on any laptop. The main risk — overclaiming clinical capability — is designed out: outputs are labelled teaching traces throughout.

In scope

• One single-lead (lead-I) channel, 0.5–40 Hz working band, 500 Hz sampling, recordings of 1–5 minutes per volunteer session.
• Volunteers drawn from the project team and consenting classmates, with a written consent form, single-use gel electrodes and a stop-if-irritation rule.
• Bench calibration against the lab function generator and oscilloscope; artefact provocation on the bench under the written protocol.
• Offline analysis in Python/SciPy (cross-checked in Scilab): PSD/SNR metrics, Butterworth band-pass, 50 Hz notch, simplified R-peak detection.

Out of scope (v1)

• Clinical use of any kind: no diagnosis, no monitoring claim, no comparison against a certified ECG machine's interpretation.
• Multi-lead (3/12-lead) recording, stress-test protocols, and patients other than the volunteers described above.
• Real-time filtering on the microcontroller — filtering is done offline so the study can compare parameter sets.
• IEC 60601 type testing (isolation, leakage current, defibrillation withstand) — these are lab-level activities; the project documents them as an awareness study, not as compliance.

The project follows the V-model, because a measurement instrument demands a matching verification stage for every design decision, and the volunteer-recording phase must be preceded by completed verification of the hardware.

Testing approach: unit checks (ADC sample rate verified against a known 50 Hz source; SD write integrity via CRC), instrument comparisons (function generator gain sweep vs datasheet), protocol tests (each provocation scripted with expected effect), and black-box test cases recorded as ID / input / expected / actual / result. Every volunteer session is logged with time, electrode batch and any incident, making the trace library auditable — the discipline examiners look for in a measurement project.

The trainer is a linear biopotential chain followed by an offline analysis pipeline. Sensing: three disposable gel electrodes (right arm, left arm, right forearm for the drive/reference) connect through the AD8232 breakout's 3.5 mm jack. The AD8232 provides the instrumentation amplifier, a right-leg-drive output that cancels common-mode 50 Hz pickup, and an onboard two-pole band-pass (0.5 Hz high-pass, 40 Hz low-pass) with overall gain around 100–1100 depending on configuration — the calibration step measures the actual value. Acquisition: the Arduino Uno's 10-bit ADC samples A0 at exactly 500 Hz from a timer interrupt, appends samples to 1-second blocks and writes them to microSD with a timestamp header, while streaming the same blocks over USB serial for live display. Analysis: recordings are imported into Python, where PSDs, SNR figures, the digital filter set and R-peak detection run; Scilab scripts reproduce the same tables as a cross-check.

flowchart TD
  V["Volunteer (lead-I, 3 gel electrodes)"] -->|"microvolt-level ECG"| AFE["AD8232 AFE breakout: instrumentation amp, right-leg drive, 0.5-40 Hz band-pass"]
  AFE -->|"analog output, gain ~100-1100"| ADC["Arduino Uno ADC A0, timer-sampled at 500 Hz"]
  ADC --> SD["microSD: raw timestamped blocks"]
  ADC -->|"USB serial, battery-powered laptop"| LIVE["Live trace display"]
  SD --> PY["Python pipeline: PSD, SNR, R-peak detection"]
  PY --> DIG["Digital filters: 0.5-40 Hz Butterworth + 50 Hz IIR notch"]
  PY --> REP["Results: SNR tables, spectrograms, detection metrics"]
  FG["Function generator: 1 mV sine sweep"] -->|"calibration input"| AFE
  SCOPE["Bench oscilloscope"] -->|"waveform verification"| AFE

The calibration branch (function generator into the electrode jack, oscilloscope on the output) keeps the study quantitative: gain and cut-off frequencies are measured, not assumed. The analysis pipeline keeps raw data untouched and writes filtered copies alongside, so every figure in the report can be regenerated from the original trace with one script — the property that makes the comparison defensible in the viva.

• Two-channel comparison board: run lead-I and modified lead-II simultaneously to study axis effects, still on the same teaching protocol.
• On-device filtering port: move the Butterworth and notch into fixed-point C on the Uno or an ESP32 and compare real-time vs offline results.
• Open reference traces: integrate PhysioNet MIT-BIH excerpts as replay signals through a DAC so the trainer can also test its pipeline without a volunteer.
• Quantified artefact atlas: grow the trace library across batches and publish an SNR atlas by provocation type and electrode batch.
• Safety-engineering extension: add a full patient-isolation front end (isolated DC-DC + digital isolator) and measure leakage current against the IEC 60601-1 limits as a documented educational exercise under faculty supervision.