Project Harmonic Engine: Reviving the Mechanical Fourier Synthesizer for Gravitational Wave Data Analysis

The mechanical Fourier synthesizer, perfected by Lord Kelvin and others in the 1800s, could sum dozens of sine waves using rotating gears, cams, and linkages. These machines predicted tides with remarkable accuracy for decades before digital computers rendered them obsolete. But for ultra-low-power, real-time spectral analysis — such as searching for gravitational wave signals in noisy LIGO data — the analog approach offers surprising advantages. Digital FFTs require millions of operations per second and consume significant power; a mechanical harmonic engine could perform the same summation physically, with power consumption dominated by friction and motor drive, potentially under 10 watts. This project aims to build a modern, high-precision mechanical Fourier synthesizer using CNC-machined components, rare-earth magnets, and optical encoders. The machine would sum up to 1000 sinusoids concurrently, with frequencies from 10 Hz to 10 kHz, controlled by stepper motors that adjust gear ratios and amplitudes. The summed motion is read by a linear variable differential transformer (LVDT) and compared to a reference signal. If successful, this could lead to a new class of ultra-low-power analog coprocessors for signal processing in space missions or remote sensors.

Gravitational wave detection demands massive real-time computation — matched filtering requires correlating noisy data against thousands of templates, each requiring Fourier transforms. Modern supercomputers handle this with gigawatts, but for space-based detectors (like LISA) or low-power Earth stations, energy efficiency is critical. Digital FFTs are inherently power-hungry due to transistor switching and memory access. In contrast, a mechanical harmonic synthesizer performs the same linear combination of sine waves through physical motion, with energy losses only from friction and control electronics. Could a resurrected 19th-century technology outperform silicon for this specific task?

• Build a prototype mechanical Fourier synthesizer capable of summing 1000 sine waves across 10 Hz–10 kHz.
• Achieve real-time correlation with simulated gravitational wave signals (e.g., from binary black hole mergers).
• Demonstrate total power consumption under 10 watts for the mechanical core.
• Compare magnitude accuracy with digital FFT: target <0.1% error at matched frequencies.
• Document scaling laws for number of channels vs. physical size and precision.

• Not intended to replace digital computers for general-purpose signal processing.
• Not aiming for phase accuracy better than 0.1 degree (digital still superior).
• Will not handle non-periodic or transient signals beyond sinusoidal decomposition.
• No attempt to miniaturize beyond tabletop size for proof-of-concept.

• CNC mill and lathe for precision metal parts (aluminum, brass).
• Rare-earth magnets for low-friction bearings and magnetic coupling.
• Optical shaft encoders (1000-line) for angular position feedback.
• Stepper motors with microstepping for amplitude and frequency adjustment.
• Linear variable differential transformer (LVDT) for displacement sensing.
• FPGA for real-time control and data acquisition.
• MATLAB/Python for simulation and comparison with digital FFT.

The machine consists of a central drive shaft rotated at constant speed by a servo motor. Along the shaft are 1000 modular gear trains, each mounted on a separate spindle. Each gear train has a set of changeable gears that determine the frequency ratio to the main shaft (e.g., ratio 1:100 gives 100 cycles per revolution). A cam with adjustable radius (via a lead screw and stepper) sets the amplitude. The follower of each cam moves a push rod that actuates a summing lever. All 1000 lever outputs are summed via a mechanical linkage that adds displacements linearly. The total displacement is sensed by an LVDT whose core is attached to the sum point. A control computer reads the LVDT output and can correlate it with a target waveform (e.g., gravitational wave template) by adjusting amplitudes and frequencies in real time. The machine is enclosed in a vibration-damped chassis with temperature control to minimize drift.

• Mechanical wear and backlash degrade accuracy over time; may require periodic recalibration.
• Precision machining tolerances (microns) may be challenging for 1000 channels.
• Resonances in the mechanical structure at high frequencies (>500 Hz) could introduce errors.
• Thermal expansion of components alters gear ratios and cam profiles.
• Achieving phase coherence across 1000 channels is difficult; digital feedback loops may be needed.

• Can we achieve <0.1% amplitude accuracy using off-the-shelf cams and followers?
• How to handle frequencies above 1 kHz without excessive shaft speed? (e.g., use helical gears.)
• Is there an alternative to cams (e.g., magnetic or electrostatic) that reduces friction?
• What is the practical limit on number of channels before mechanical complexity becomes prohibitive?
• Could this be integrated into a hybrid analog-digital system for low-power edge computing?

The team evolved from a garage workshop to professional research, and funding for a large-scale mechanical prototype never materialized. Life moved on, and the idea remains a fascinating 'what-if' — a plan that is complete on paper but waiting for the right workshop and sponsor.

This project draws on the rich history of mechanical computing, from the Antikythera mechanism to the NACA Differential Analyzer. Modern CNC and sensor technology make it feasible to revive these designs with orders-of-magnitude better precision. If built, the Harmonic Engine could be a centerpiece for STEM outreach, demonstrating that old ideas can find new life.