Text from the interactive explanation
Triplicator phase-only hologram and its use as a snapshot optical convolver and correlator
Shang Gao, María del Mar Sánchez-López, Pascuala García-Martínez, Ignacio Moreno
Original paperTriplicator hologram · snapshot optical convolver & correlator
Could a single flash of light compare two patterns — and tell you where they match? It starts with a hologram that splits one laser beam into a picture, its upside-down twin and a bright point, all with the same energy. Drag the flower.
Light measured on the simulated cameraLeft 32% · Centre 34% · Right 32% of the light
Simulated with the paper's equations (Fourier plane of a lens, 256 × 256-pixel hologram). Lab photos below. Grab the flower — or its twin — or focus the image and use the arrow keys.
Chapter 1
Bending light with delays
A spatial light modulator (SLM) is a liquid-crystal screen. Its pixels don't dim the laser — each one delays it by an amount the computer chooses, shifting the wave's phase . A uniform delay changes nothing you can see. But a repeating pattern of delays — a diffraction grating — sends one beam out as several beams, the diffraction orders 0, ±1, ±2 …
Pick a delay profile and watch where the light goes. The efficiencies below come from a Fourier transform of one period of the profile.
—
Challenge: choose Triplicator and find the strength a where three beams are equally bright.
In 1998 Gori and colleagues derived the best possible phase-only profile for three equal beams: φ(x) = arctan[a·cos(2πx/p)] with a = 2.65. Each of the three beams gets about 30.9 %, together 92.6 % of the light; the rest leaks into weak ±2, ±3 … orders. And it needs delays of only ±0.39π. Try Draw your own: can you beat 92.6 % with three equal beams? (Their analysis says a phase-only grating can't.)
Chapter 2
From stripes to pictures
A lens is a Fourier-transform machine: one set of straight, evenly spaced stripes on the screen becomes one point at the camera, and the stripes' spacing and tilt decide where the point lands. A picture is just many points, so its hologram is many stripe patterns added together. Keep only the delays — the phase, φ = arg(IFFT(picture)) — and you have a phase-only Fourier hologram . One extra set of stripes, the carrier cx, shifts the picture sideways: a single off-axis image, which the authors call a blazed hologram.
—
Diffuser off → an outline. A smooth picture's Fourier transform is huge at a few low frequencies and tiny elsewhere. Throwing away that brightness information flattens it, which boosts the sharp changes — so edges glow (edge enhancement, the paper's Fig. 3a).
Diffuser on → filled, but grainy. Multiplying the target by a random phase spreads its light evenly over the whole hologram, so discarding brightness hurts much less and the shape fills in (Fig. 3b). The price is speckle. Hold on to that grain — it turns out to be useful.
Chapter 3
The triplicator hologram — the paper's key move
Now combine chapters 1 and 2. Take the hologram's phase φ(x), add the carrier, and pass every pixel through the triplicator curve (the paper's eq. 4):
uh(x) = arctan[ a · cos( φ(x) + cx ) ]
—
Why three? The curve is periodic, so eiuh can be expanded like a grating (eq. 5). Its three strongest terms carry e+iφ, a constant, and e−iφ. The first rebuilds the picture h on the right; the constant becomes a single point on axis; the last rebuilds h*, the complex conjugate — which a lens shows rotated by 180°, on the left (eq. 6). All three carry the same energy; the point looks far brighter only because its share is focused into one spot.
Only 0.78π of delay. Every value of uh stays within ±0.39π, so the hologram also works on screens that cannot reach a full 2π. Their SLM reaches 2π, so the authors folded a lens phase (f = 160 mm) into the same pattern and needed no external lens (Fig. 2e). Separation between orders at the camera: x′0 = λf / p.
Chapter 4
Light that compares
Correlation is a sliding comparison: slide one pattern across the other and, at each shift, add up how well they agree. Where they line up, the total is large. Convolution is the same thing with one pattern flipped first. Slide the lower barcode.
—
A smooth shape agrees with itself over a broad range of shifts: a gentle hill. A random barcode agrees with itself at exactly one shift: a needle. A different random barcode — or the flipped one — never lines up. That is why the diffuser's speckle, a nuisance for pictures, makes an optical match peak needle-sharp, and why a different noise pattern makes it vanish.
The correlator camera
Now multiply the triplicator hologram by a second phase pattern Q = eiφq — on the SLM that just means adding the delays. The paper's eq. 8 then puts, in one snapshot: the probe pattern q itself in the 0 order, the convolution q∗h in the +1 order and the correlation q⋆h in the −1 order. When the probe is the hologram's own phase, the delays cancel perfectly in the −1 order and it collapses to a point.
Correlation order, measured on the simulated camera—
Drag the image (or use the arrow keys) to shift the probe: the correlator also tells you where the match is.
Chapter 5
Nine answers at once
Cross two triplicators: one whose stripes make orders left–right (from φ1), one whose stripes make orders up–down (from φ2). Add their phases, uh1 + uh2 (eq. 10), and three times three orders appear (eq. 12): the point in the centre, replicas at left, right, top and bottom, convolutions h1∗h2 on one diagonal and correlations h1⋆h2 on the other.
Sharp points found in the simulated image—
Same pattern, same noise in both directions: the two correlations become points (Fig. 5a, b).
With a different noise realisation in x and y, the shapes are still identical — yet the correlation points disappear and only the centre point survives (Fig. 5c): a sharp match needs the patterns to agree in every respect, amplitude and random phase. Multiplying by e−iφ1 cancels the horizontal hologram exactly where it sits, so the point jumps to the right-hand cell (Fig. 5d) — a phase-only correlation peak, which the authors suggest as an efficient way of sensing and analysing a given input beam.
Chapter 6
In the lab: what was measured
Everything above is a numerical simulation of the paper's equations. The authors demonstrated it experimentally ; these are their photos.
Measured (paper) ↔ Simulated (this page)
What the photos show
Three equal orders, measured. Fig. 3c–f: the flower on the right, its inverted twin on the left with the same intensity, and the bright zero-order point. With a circle, both sides look identical.
A match spot, measured. Fig. 4a: adding the hologram's own phase gives a narrow autocorrelation peak on the left, the flower in the centre with the same intensity as the replicas in Fig. 3d, and a noisy autoconvolution cloud on the right. Fig. 4b swaps the sides.
Symmetric shapes, two spots. Fig. 4c: for a circle without diffuser, autoconvolution equals autocorrelation, so both side orders are narrow spots.
Spots vanish with different noise. Fig. 5c: the same circle with a different random phase in x and y gives no correlation points — only the centre point remains. Fig. 5d moves the point to the right-hand cell.
Chapter 7
Back to the question
Could a single flash of light compare two patterns — and tell you where they match?
In this paper's experiments, one passive phase pattern on one screen produced a copy, a convolution and a correlation simultaneously, in one camera snapshot — and the correlation became a sharp spot only when the two patterns matched exactly, shape and random phase alike. The crossed version delivered nine such answers at once.
What this could bring closer, in the authors' own framing: optical correlation can be done "in a single step and with inherent parallelism"; triplicator holograms "could find applications in the detection and quantitative evaluation of structured light beams"; and the phase-only correlation peak of Fig. 5d is "an interesting efficient method for sensing and analyzing a given input beam". They also place it in the wider context of diffractive neural networks, where optical convolution is explored as a "high-speed and energy-efficient" alternative to electronic processing.
1A repeating pattern of delays splits a beam into orders.
2The arctan[a cos] profile with a = 2.65 makes three equal beams, 92.6 % in total.
3A Fourier hologram is a stack of stripes whose points draw a picture.
4Pass the hologram through the triplicator curve: picture, inverted twin, point — equal energy.
5Add a probe's phase: copy, convolution, correlation in one snapshot.
6Cross two of them: a 3 × 3 grid of answers.
Source
Shang Gao, María del Mar Sánchez-López, Pascuala García-Martínez and Ignacio Moreno, “Triplicator phase-only hologram and its use as a snapshot optical convolver and correlator”, Journal of the European Optical Society-Rapid Publications 21 , 33 (2025). DOI: 10.1051/jeos/2025027 · Article page
Published open access under the Creative Commons Attribution License (CC BY 4.0). Figures 1–5 are reproduced from the article under that licence.
Simulations on this page solve the paper's equations numerically; laboratory images are the authors' measurements. The optimum triplicator profile is due to Gori et al., Opt. Commun. 157, 13 (1998).
Topics in this iPaper
Physics
we extend the optimum phase triplicator diffraction grating profile to a phase-only hologram
Abstract, page 1
Engineering
liquid-crystal on silicon (LCOS) SLM (Thorlabs, model Exulus-HD1)
Section 3, page 2
computer generated holography
phase-only Fourier transform (FT) computer-generated hologram (CGH)
Section 1, page 1
diffraction gratings
A triplicator is a diffraction grating that generates three equally intense 0th and ±1st diffraction orders.
Section 1, page 1
spatial light modulators
high-resolution liquid-crystal phase-only spatial light modulator (SLM)
Abstract, page 1
optical correlation
the resulting hologram yields convolution and correlation terms in the ±1st orders
Abstract, page 1
fourier optics
Optical convolution is a basic operation in Fourier optics
Section 4, page 4