Aperture Synthesis

The two-dish trick

Suppose you want the sharpness of a 300-metre telescope but can only afford two 25-metre dishes. Build the big dish and you would resolve detail of order θ ≈ 1.22 λ/D, with D = 300 m. The remarkable fact behind aperture synthesis is that the resolution of that dish comes entirely from the separation between its outermost edges — the points 300 m apart — and not from the metal in between. Two small dishes placed 300 m apart reproduce exactly that finest interference fringe. What they cannot reproduce on their own is the collecting area and the full set of intermediate spacings the filled dish samples automatically.

So the bargain is clear. A pair of antennas gives you the angular resolution of a dish as large as their separation — their baseline — at the cost of sensitivity and of having to measure many spacings one at a time. Aperture synthesis is the bookkeeping that turns a sparse collection of small dishes, each pair measuring one piece of the puzzle, into a single sharp image. The "aperture" is synthesized: you never build the giant dish, you reconstruct what it would have seen.

Each pair measures one Fourier component

Point two antennas at the same source and combine their voltages in a correlator — a device that multiplies and time-averages the two signals. The output is a single complex number called the visibility V, with an amplitude and a phase. The central result of interferometry, the van Cittert-Zernike theorem (Pieter Hendrik van Cittert 1934, Frits Zernike 1938), states that this visibility is one Fourier component of the sky brightness distribution I(l, m):

V(u, v) = ∬ I(l, m) · exp[−2πi (u·l + v·m)] dl dm

Here (l, m) are direction cosines on the sky, and (u, v) are the components of the baseline vector between the two antennas, measured in wavelengths, projected onto the plane perpendicular to the line of sight. Each baseline therefore samples one point in this uv-plane: a short baseline samples a low spatial frequency (coarse structure), a long baseline samples a high spatial frequency (fine structure). The image you want is just the inverse 2D Fourier transform of V(u, v). The whole game is to measure V at enough (u, v) points to invert it cleanly.

Earth-rotation synthesis fills the uv-plane

One pair of fixed antennas seems to give just one uv-point — hopelessly little. Martin Ryle's key insight was that the Earth is a free, slow turntable. As the planet rotates beneath a source, the line of sight to that source sweeps across the array, so the projected baseline — its appearance as seen from the source — continuously changes. Over a night, a single pair traces out an ellipse in the uv-plane instead of a single dot. Because the sky brightness of a real source is real-valued, its Fourier transform is Hermitian: V(−u, −v) = V*(u, v). So every measured point gives you its mirror partner for free, and each baseline actually paints two arcs.

Now scale up. An array of N antennas has N(N−1)/2 distinct pairs, each with its own baseline and its own ellipse. The 27-antenna Very Large Array has 27·26/2 = 351 simultaneous baselines; over a 12-hour track those 351 ellipses interleave to cover the uv-plane densely. The synthesized "aperture" is the union of all those tracks. Wherever the uv-plane is left unsampled, you have missing Fourier components — gaps that show up as artefacts (the "dirty beam" sidelobes) that deconvolution algorithms like CLEAN must then clean up.

What sets resolution, field of view, and the largest visible structure

Three different spacings control three different image properties, and conflating them is the most common source of confusion:

Angular resolution   θ_res ≈ λ / B_max        (longest baseline)
Field of view        θ_FOV ≈ λ / D             (single-dish primary beam)
Largest visible scale θ_max ≈ λ / B_min        (shortest baseline)

The longest baseline B_max sets how fine a detail you can resolve — this is the synthesized aperture. The individual dish diameter D sets the field of view, because each antenna only "sees" the patch of sky within its own primary beam of width ≈ λ/D. And the shortest baseline B_min sets the largest angular scale you can detect: structures broader than λ/B_min vary too slowly across the array to produce fringes, so they are "resolved out" and vanish — which is why interferometers are blind to smooth, extended emission unless short spacings (or single-dish data) are added. A long-baseline array is a sharp but narrow magnifying glass, not a wide-angle camera.

The numbers — real arrays, real resolutions

Plugging real wavelengths and baselines into θ ≈ λ/B (in radians, then converted to arcseconds via 206265 arcsec/rad) shows the enormous dynamic range the technique spans:

Instrument	Wavelength λ	Max baseline B_max	Resolution θ ≈ λ/B	N antennas
One-Mile Telescope (1964)	21 cm (1.4 GHz)	1.6 km	~27″	3
VLA (A-configuration)	21 cm (1.4 GHz)	36 km	~1.2″	27
VLA (A-config, 7 mm)	7 mm (43 GHz)	36 km	~0.04″	27
ALMA (longest config)	1.3 mm (230 GHz)	16 km	~0.018″	66
VLBA (continental)	3.6 cm (8.4 GHz)	8600 km	~0.001″ (1 mas)	10
Event Horizon Telescope	1.3 mm (230 GHz)	~10,700 km (≈ Earth)	~25 μas	8 telescopes

The Event Horizon Telescope line is the punchline: at 1.3 mm with a baseline equal to the Earth's diameter, the resolution is about 25 microarcseconds — sharp enough to read a credit card in Los Angeles from New York, and exactly what was needed to image the 40-microarcsecond shadow of M87*'s 6.5-billion-solar-mass black hole, published 10 April 2019.

Worked example: the resolution of the VLA at 21 cm

Take the VLA in its widest "A" configuration, observing the 21 cm hydrogen line. The longest baseline is B_max = 36 km = 3.6 × 10⁴ m, and λ = 0.21 m. The diffraction-limited resolution is

θ ≈ λ / B_max
  = 0.21 m / 3.6 × 10⁴ m
  = 5.8 × 10⁻⁶ rad
  × 206265 arcsec/rad
  ≈ 1.2 arcseconds

Now compare to a single VLA dish, D = 25 m: its standalone resolution at 21 cm is θ ≈ λ/D = 0.21/25 ≈ 8.4 × 10⁻³ rad ≈ 1700″ ≈ half a degree. The interferometer is sharper by a factor B_max/D = 36000/25 = 1440×. That same factor is exactly why you go to the trouble: 27 small dishes spread over 36 km resolve the sky 1440 times finer than any one of them, matching a single hypothetical dish 36 km across — a dish nobody could ever build.

The sensitivity, by contrast, scales only with the total collecting area actually built: 27 dishes of 25 m give an effective area of 27 × π(12.5)² ≈ 1.3 × 10⁴ m², the same whether they are packed together or spread to 36 km. Spreading them out buys resolution for free but does nothing for sensitivity — that is the fundamental trade of synthesis imaging.

History — Michelson to Ryle to the EHT

The lineage runs through optics first. In 1920 Albert Michelson and Francis Pease mounted a 6-metre beam across the 2.5-metre Hooker telescope on Mount Wilson and measured the angular diameter of Betelgeuse — 0.047″ — by watching interference fringes from light entering at the beam's two ends. That is aperture synthesis in embryo: resolution set by the 6 m beam separation, not the 2.5 m mirror.

Radio astronomy industrialised it. After wartime radar work, Martin Ryle and his Cambridge group spent the 1950s building interferometers, culminating in the One-Mile Telescope (1964) and later the 5 km Ryle Telescope. Ryle formalised Earth-rotation synthesis and the iterative reconstruction of images from incomplete uv-coverage. In 1974 he and Antony Hewish shared the Nobel Prize in Physics — the first Nobel given for astronomy — Ryle "for his observations and inventions, in particular of the aperture-synthesis technique." The method then scaled relentlessly: the VLA opened in New Mexico in 1980, the continent-spanning VLBA in 1993, ALMA in the Atacama in 2011-2013, and the Event Horizon Telescope turned the entire planet into a single instrument with its black-hole images of M87* (2019) and Sgr A* (2022). Antony Hewish's co-prize was for the discovery of pulsars; the aperture-synthesis half of the prize is Ryle's alone.

Variants and relatives

• Connected-element interferometry. Antennas linked by cable or optical fibre to a central correlator in real time — the VLA, ALMA, the SKA. Baselines up to tens of kilometres; phase is preserved over the link.
• Very Long Baseline Interferometry (VLBI). Antennas on different continents that cannot be cabled. Each records its data stream against a hydrogen-maser atomic clock; the streams are shipped to a correlator and combined offline. Baselines reach an Earth diameter; the EHT and VLBA are VLBI.
• Closure phase and closure amplitude. Atmosphere and instrument corrupt the visibility phase on each baseline, but the sum of phases around a closed triangle of three antennas — the closure phase — cancels all per-antenna errors and is a true property of the source. Closure quantities were essential to the EHT reconstructions.
• Self-calibration and CLEAN. Iterative algorithms (CLEAN, Högbom 1974; self-cal, 1980s) that deconvolve the "dirty beam" caused by incomplete uv-coverage and solve for residual antenna phase errors using the source itself as a reference.
• Intensity (Hanbury Brown-Twiss) interferometry. Correlates intensity fluctuations rather than amplitudes, sidestepping phase stability — used at Narrabri in the 1960s to measure stellar diameters, and a conceptual cousin rather than a Fourier-imaging method.
• Mosaicking and short-spacing correction. To recover large-scale structure resolved out by the array, observations are tiled across many primary-beam pointings and combined with single-dish "total power" data that supply the missing zero- and short-spacing visibilities.

Common misconceptions and subtleties

• "More dishes = more sensitivity AND resolution." Adding antennas multiplies the number of baselines (better uv-coverage and image fidelity) and adds collecting area (sensitivity), but resolution is set only by the longest baseline. You can sharpen the image purely by moving two existing dishes farther apart, gaining no sensitivity at all.
• "The synthesized telescope has the collecting area of the big dish." No — it has the resolution of a B_max-wide dish but only the collecting area of the small dishes you actually built. Resolution and sensitivity are decoupled in synthesis imaging.
• "Interferometers see everything a single dish sees, only sharper." They are blind to structure larger than λ/B_min (resolved out) and outside the λ/D primary beam. A diffuse, smooth source can be completely invisible to a long-baseline array even when a small single dish sees it easily.
• "Each baseline gives a piece of the image." Each baseline gives one Fourier component, not one patch of sky. Every visibility carries information about the entire field; the image only emerges after Fourier-transforming the whole ensemble.
• "Phase doesn't matter, only fringe amplitude." The visibility phase encodes where structure sits on the sky; throwing it away destroys positional information. Defeating atmospheric phase corruption (via closure phase and self-calibration) is the central practical challenge of high-frequency and VLBI synthesis.