The paper asks whether a practical single-axis acoustic levitator can be built without the high-voltage, hand-matched and temperature-sensitive Langevin horns used by previous designs. The authors design, build and evaluate TinyLev, a non-resonant levitator made of 72 low-cost, low-voltage (ca. 20 V) 40 kHz ultrasonic transducers arranged in two 3D-printed curved bowls and driven by an Arduino and a motor-driver board. Simulations (piston-source field and Gor'kov force) guided the transducer arrangement and bowl radius; experiments measured the minimum excitation voltage needed to hold samples

Two bowls of tiny speakers sing at each other, far above the pitch you can hear. Where their waves meet they build an invisible ladder of quiet spots — and anything small enough to sit in one simply floats.

Acoustic levitators used to be finicky, high‑voltage lab instruments. In 2017, three engineers in Bristol rebuilt one from parts anyone can buy — 72 ultrasonic speakers, an Arduino and a 3D‑printed frame — and hoped it could become a standard science demonstration in schools. Could it? Start by playing with theirs.

Acoustic levitators used to be finicky, high‑voltage lab instruments. In 2017, three engineers in Bristol rebuilt one from parts anyone can buy — 72 ultrasonic speakers, an Arduino and a 3D‑printed frame — and hoped it could become a standard science demonstration in schools. Could it?

Two waves make a cage

Why can sound hold anything at all? Start with one bowl, then add the other.

Sound is a travelling wave of squeezed and stretched air. Like any wave it carries momentum, so it can push on whatever it hits — physicists call this the acoustic radiation force . A normal voice pushes far too weakly to notice. Focused ultrasound is a different story.

TinyLev's speakers vibrate 40,000 times a second, above human hearing. At that pitch one wave in air is only 8.65 mm long (at 25 °C), so the sound's pattern is about the size of a raindrop.

When two identical waves run into each other from opposite directions they stop looking like travelling waves at all. They lock into a standing wave : bands where the air pressure swings hugely (antinodes) alternate with thin layers where it barely changes (nodes), half a wavelength apart.

A small, dense object is pushed out of the violent bands from above, below and the sides, and squeezed into a quiet layer. Switch to Trap map to see those layers as violet wells. In TinyLev's focused geometry the wells sit about 4.7 mm apart — a little more than the 4.3 mm half‑wavelength, because the waves arrive at an angle.

That spacing also sets a size limit: an object has to fit in one trap. TinyLev holds things up to roughly half a wavelength, about 4 mm across.

Field. Each transducer is treated as a small vibrating disc (a “piston source”). Its sound spreads out with a known directivity, and the pressure at any point is the sum of all 72 contributions — exactly the approach the paper describes.

Force. For an object much smaller than the wavelength, the energy it has in the sound field is given by the Gor’kov potential:

U ∝ |p|² − (3 / 2k²) |∇p|²

The first term is large where the pressure swings hard; the second is large where the air sloshes back and forth fastest. The force is the downhill slope, −∇U, so objects collect where the pressure is quiet. The trap map shows U.

Echoes. Sound also bounces between the bowls, which the model ignores. The authors checked this: a short pulse sent from the top array still had 86% of its amplitude after bouncing between the arrays, and they judged reflections negligible because simulation and experiment agreed well.

One big horn, or 72 small voices?

Levitation had worked for decades. The hard part was making it practical.

Earlier single‑axis levitators were built around Langevin horns : stacks of piezoelectric discs clamped between a heavy backing block and a resonating metal horn. One horn can produce enormous sound pressure, but the authors list the catches:

They need high voltage — typically 100 to 1000 volts, which is potentially dangerous. They are hard to tune. Two opposing horns must resonate at the same frequency; one team reported building dozens and keeping the two that matched best. They drift. Horns heat up as they work, their resonance shifts, so they need a warm‑up and lose power after intense use. Resonant designs can also detune when the temperature changes the speed of sound, or when a large sample is put in.

TinyLev's idea is the same move lighting made when one powerful lamp gave way to arrays of LEDs in traffic lights, projectors and spotlights: replace one demanding part with many small, cheap, well‑matched ones . The 10 mm transducers sold for distance sensors already share a consistent 40 kHz resonance and run at about 20 volts.

Cheap parts aren't identical: the authors measured each transducer's timing (phase) scatter. For the budget models it was about 14°, which in their simulation costs only about 3% of the trapping force.

But a little speaker the size of a fingertip doesn't focus its sound. How do you get 72 of them to work as one?

The design bench

Aim the speakers, size the bowls, set the voltage — and see whether a 3 mm water drop survives your design.

Aim: four ways to point 36 speakers

A big horn focuses sound because of its shape. An array of small speakers has to be arranged so that all their waves arrive at the centre in step. The authors simulated four options (their Fig. 3):

Flat — speakers on a flat plate. Trivial to build, but the waves arrive out of step and largely cancel: about 2% of the best grip. Flat + timing — delay each speaker's signal electronically so the waves line up (a “phased array”). About half the best grip, but it needs complex electronics producing many separate signals. Stepped — build the same delays into the geometry by setting each ring at a different height. Same result as electronic timing, with just one signal per bowl. Stepped + tilted — also tilt every speaker to point straight at the centre. This is TinyLev, and the strongest of all.

Size: how far apart should the bowls be?

Each bowl is part of a sphere centred on the middle of the chamber, so the bowl's curve and the gap between bowls change together. Small bowls concentrate the sound: in the paper's simulations the vertical grip peaked at a radius of 4.5 cm . Bigger bowls give a longer ladder of traps but a weaker grip.

The authors chose 6 cm anyway. Steeper bowls caused more unwanted reflections, and the frame is 3D‑printed in one piece with the bowls facing up, so a deep bowl means overhangs that print badly. The 6 cm bowl also gives more usable traps. Slide the radius and count them.

How many speakers? Rings of 6, 12 and 18 come from packing circles in a hexagon. The centre spot is left empty: a hole for a camera or an injector, and an exit for drops that fall. A fourth ring of 24 would add grip, but also cost and complexity, and it would close the bowl into a more resonant, more temperamental shape.

The elevator

Every trap in the ladder can be moved at once, using nothing but timing.

The electronics are almost embarrassingly simple. An Arduino Nano produces two square waves at 40 kHz, and a cheap dual H‑bridge motor‑driver board amplifies them. Wired “push‑pull”, it swings each transducer between plus and minus the supply, doubling the peak‑to‑peak voltage (up to 70 V).

Square waves are easy to make digitally, but they aren't the smooth sine waves sound should be. That's fine: the transducers only resonate at 40 kHz, so they act like a filter and turn the square drive into a nearly pure sine — compare the faint and bright traces on the scope.

One bowl's signal stays fixed. The other's can be delayed in steps of π/12 — 15°, a twenty‑fourth of a cycle. Delaying one side slides the whole standing wave towards it, and every trapped object rides along.

Press + a few times or Run elevator , and watch the ringed bead against the ruler. In a real TinyLev this is how levitated samples are moved up and down; the authors used it to carry samples across several traps.

What can it lift?

The paper's key measurement: put a sample in, then turn the voltage down until it falls.

Here is a neat simplification. For objects smaller than half a wavelength, the grip on them and their weight both grow in proportion to their volume . Double the volume and you double both, so size cancels out. Only density decides whether something can float.

So the authors' test was: pick materials of known density, levitate a sample, and slowly lower the drive voltage until it drops. They repeated each drop test four times. Lower voltage means weaker sound, and the grip grows with the square of the voltage.

In the paper, water fell at about 11 V . Fused silica (glass) is 2.2 times denser. At what voltage do you expect it to fall?

The paper's four measured samples — isopropyl alcohol, water, sugar and fused silica — fall close to the simulated curve (blue points with error bars). That agreement is the evidence that the simple model of the sound field and force really describes the device. The live simulation on this page is calibrated to that same curve, so your falls should land on it too.

They didn't stop there. Moulding sand, Blu Tack, ceramic, a surface‑mount MOSFET chip and even sapphire spheres at 3.9 g/cm³ also flew. For those, the transducers were driven above their 20 V rating, the shapes were irregular and some samples absorbed sound, so measurements and simulation agreed less well. A scaled‑up version, BigLev, 160% larger with 16 mm transducers, held samples up to 6.5 g/cm³ .

Photos: Fig. 5 of Marzo, Barnes & Drinkwater (2017), CC BY 4.0 — cropped into separate tiles, panel labels removed.

Tough enough for a classroom

A school demo has to survive real rooms, real hands and long lessons. The authors stress‑tested it.

A 350 °C soldering iron held 5 mm from the sample for 10 minutes. The sample stayed stably trapped, despite the hot air around it.

An ultrasonic mister was pointed at the levitation zone. Droplets formed and were held, so very humid air is no problem.

Solutions were levitated continuously for more than two hours. The authors used this to measure how sugary water and jelly evaporate.

The whole levitator draws about 10 watts and works from the moment it is switched on, without the warm‑up that horn levitators need.

A levitator anyone can build

Everything you have used on this page, as a parts list. Tap a part.

Each piece of TinyLev is cheap and off the shelf, or printed. The paper comes with assembly instructions, a parts list, 3D models and the Arduino code as supplementary material.

What changed was not the physics of standing waves, which was understood long before. The authors showed that a design using 72 cheap, matched, low‑voltage parts aimed by geometry, simulated first and then measured, holds liquids, solids and even insects reliably, for hours, while a soldering iron sits next to it. They call it a democratization of acoustic levitation. The same journal has since published work that builds on it, such as a heating system for drying droplets inside the TinyLev levitator (2025).

Source & credits

Asier Marzo, Adrian Barnes and Bruce W. Drinkwater, “TinyLev: A multi‑emitter single‑axis acoustic levitator”, Review of Scientific Instruments 88 , 085105 (2017). doi.org/10.1063/1.4989995

Article content © 2017 Author(s), licensed under Creative Commons Attribution 4.0 (CC BY 4.0). The article's supplementary material (assembly video, 3D models, Arduino code and additional tests) is available with the article.

Figures. Photos and simulation maps from Figs. 1, 2, 3 and 5 of the paper, cropped and resized, with panel labels removed. Charts on this page are redrawn from the paper's Figs. 2b, 3b and 4.

Interactive levitators. A simplified re‑implementation of the paper's approach (piston‑source field and Gor’kov force on the mid‑plane, echoes ignored), calibrated so its central trap follows the paper's simulated curve in Fig. 4. Extra damping is added so beads settle, and waves are shown slowed about 40,000×. Its results are a teaching model; values marked “Measured” come from the paper's experiments.

Illustrations. The horn morph, test cards and parts diagram are schematic drawings made for this page.