Skip to main content

Healthcare / Prosthetics / Biosignal

EMG Sensor Front-End: Low-Noise Myoelectric Electrode Design for Bionic-Hand Control

For an Indian prosthetics company's bionic-hand programme we designed a low-noise surface-EMG electrode front-end (90–450 Hz, proportional envelope output). Status: simulation-validated, fab-ready 4-layer layout (2026); prototype not yet built.

Indian prosthetics companyDesign and simulation: June 2026Rapid Circuitry analog design team

Published Last reviewed:

Simulated
Status: layout fab-ready, not yet built
0.61 µVrms
Input noise, vendor-model simulation
90–450 Hz
EMG pass band
27 × 18 mm
4-layer board, 0 DRC errors
Illustrative photo: a small EMG front-end circuit board with snap electrodesIllustrative image

The Challenge

Myoelectric hands are driven by the tiny voltages muscles produce under the skin. The client needed an electrode front-end that turns signals from about 10 µV upward into a clean, proportional control voltage, small enough to sit in a socket and robust against mains hum and electrostatic discharge.

Signals near the noise floor

Surface EMG starts at around 10 µV, so input-referred noise and amplifier choice decide whether weak contractions are usable at all.

Impact: Usable from ~10 µV

Mains hum and common-mode noise

The body picks up 50/60 Hz interference that can be far larger than the muscle signal. The design needed high common-mode rejection and a notch option.

Impact: 50 Hz and 60 Hz variants

Socket-sized board

The electrode had to match the size of commercial myoelectrodes, which forces a dense, 0402-heavy layout around a sensitive high-impedance input.

Impact: 27 × 18 mm envelope

Proportional control output

The prosthesis controller needs a smooth envelope proportional to muscle effort, not raw EMG, delivered as a simple analog voltage.

Impact: 0–3 V envelope

Our Solution

We designed and simulated two front-end topologies in ngspice, then carried the production-style variant through a full KiCad schematic, a routed 4-layer board and a fabrication package. A structured review of placement, routing and QC before release caught a protection-circuit fault that simulation alone had missed.

System Architecture

Electrode signal to proportional control voltage on a single supply with a mid-rail reference.

Input and instrumentation amplifier

  • Dual-rail BAV99 clamps on each input for ESD
  • Series input resistors to the instrumentation amplifier
  • AD623 instrumentation amplifier with trimmable gain
  • Short, symmetric, via-free differential input pair

Filtering

  • Sallen-Key high-pass at 90 Hz
  • Sallen-Key low-pass at 450 Hz
  • Twin-T 50 Hz notch in the discrete variant (60 Hz option)
  • OPA2378 dual op-amp; MCP6004 simulated as a low-cost drop-in

Reference, supply and output

  • MAX6125 2.5 V reference as the mid-rail bias
  • LP2980 4.7 V LDO from the battery
  • Discrete variant: precision rectifier and envelope filter to 0–3 V
  • Test points and fiducials for assembly and bring-up

Key Parts and Simulated Performance

Instrumentation ampAD623
Filter op-ampOPA2378 (MCP6004 alternative)
Mid-rail referenceMAX6125, 2.5 V
Mid-band gain (sim)47.7 dB
Pass band (sim)89.8–453.9 Hz
Input noise (sim, vendor model)0.61 µVrms
CMRR (sim, vendor model)112 dB
Board4-layer, 27 × 18 mm, 38 footprints

What We Delivered

  • ngspice simulation decks for AC response, noise and transient behaviour
  • Discrete zero-drift design and a production-style AD623 design
  • KiCad schematic with 0 schematic-to-board parity errors
  • Routed 4-layer board: 0 DRC violations, 0 unconnected nets
  • Gerbers, drill files, pick-and-place and assembly drawings
  • 3D STEP model of the board
  • Detailed BOM with manufacturer part numbers
  • Design document and simulation results summary

Key Engineering Decisions

What we chose, and why.

Fix the ESD clamps

Found in layout review, not in simulation

Single Schottky clamps to ground would conduct at the 2.5 V bias

Replaced with BAV99 dual-rail clamps

Check against vendor models

Behavioural models can mislead

Re-ran with the manufacturer's AD623 macromodel

Noise 0.61 vs 0.66 µVrms; CMRR 112 vs 100 dB

Cheap filter op-amps are fine

Filters sit after 47.7 dB of gain

MCP6004 matched OPA2378 to four decimals in simulation

Noise and CMRR are set by the input stage

Verify every pinout

Checked against datasheets and KiCad libraries

Fixed three pin and package errors before release

Reference, op-amp package and LDO pin 4

Outcome and Status

The design is simulation-validated and the board is ready for fabrication. No prototype had been built when this was written, so every performance figure on this page is a simulation result, not a bench measurement. The next step is a prototype build and bench testing on electrodes.

Input-referred noise

0.61 µVrms

Simulated with the manufacturer's AD623 model, 90–450 Hz band

Discrete zero-drift variant

0.20 µVrms

Simulated; CMRR 100 dB (+33 dB with driven right leg), 37 dB notch, envelope linearity under 0.5%

Board checks

0 / 0 / 0

DRC violations, unconnected nets and schematic parity errors

Fabrication package

Ready

Gerbers, drill, pick-and-place, assembly drawings, STEP and BOM

Technologies Used

AD623OPA2378MCP6004MAX6125LP2980BAV99ngspiceKiCadFreeroutingPython

Related Case Studies

Related services, industries and guides