Radio interferometry and aperture synthesis
Combining signals from widely separated antennas samples the Fourier transform of the sky; imaging means inverting that incomplete, noisy data. It underlies radio surveys and black-hole images.
Almost every high-resolution radio image, from the Event Horizon Telescope to SKA precursors, and the localisation of fast radio bursts to host galaxies rely on aperture synthesis and its calibration and deconvolution algorithms.
Linking radio dishes far apart so together they act as one giant telescope, then rebuilding the picture with math.
- 1Pairs of dishes measure sky patterns
- 2Earth's rotation fills in more patterns
- 3Data are an incomplete jigsaw
- 4Algorithms fill gaps and clean image
- 5Result: sharp picture, array-sized
What an interferometer measures. Each pair of antennas measures one complex number called a visibility. It depends on the projected separation (baseline) of the pair, expressed as a point (u, v) in the "uv plane". For a small patch of sky, the visibility is one Fourier component of the sky's brightness: the amount of one particular stripe pattern at one angle and spacing. Longer baselines probe finer detail.
Sampling. An array of N antennas has N(N−1)/2 baselines and so samples that many points. The Earth's rotation sweeps each baseline along an arc, filling the uv plane more. Still, the coverage is incomplete. The inverse Fourier transform of the sampled data is the dirty image: the true sky convolved with the array's point-spread function (the dirty beam), which has strong sidelobes.
Calibration. Before imaging, errors from antenna electronics and the atmosphere must be removed. Phase referencing uses a nearby known source. Closure phases, sums of phases around triangles of antennas, cancel antenna-based errors. Self-calibration alternates between imaging and solving for the antenna gains. Direction-dependent effects, such as ionospheric distortion across a wide field, are a major difficulty at low frequencies.
Deconvolution. CLEAN (Högbom, 1974) starts with an empty model and repeatedly finds the brightest peak in the residual image, subtracts a scaled copy of the dirty beam, and records a point source. It is fast and robust but has weaknesses: it models diffuse emission poorly, depends on weights and hand-made masks, can give negative fluxes and does not give uncertainties. Bayesian approaches, such as the code resolve, fit a probabilistic sky model directly to visibilities and return error bars, at a higher computational cost.
Applications. Large arrays such as the VLA, ALMA, LOFAR, MeerKAT and the future SKA are used for surveys. Very long baseline interferometry (VLBI) links stations across continents; the Event Horizon Telescope used it for images of the black holes in M87 and the Milky Way.
Open issues. Imaging with sparse coverage, handling huge data volumes, and reliable error bars.
Go deeper
The full technical text.
The measurement
Each pair of antennas measures one complex number, the visibility, at a spatial frequency set by the projected baseline between them (the point in the "uv plane"). In the small-field, flat-sky approximation the visibility is a Fourier component of the sky brightness. An array with antennas samples points, and the Earth's rotation sweeps each baseline along an arc, filling in more of the uv plane. The result is incomplete Fourier data with noise, so the "dirty image" (the inverse transform of the sampled data) is the true sky convolved with the array's point-spread function, whose sidelobes can be large.
Words on this page (12)
- interferometer
- An instrument that combines waves from separate collectors to get detail that one collector alone could not give.
- baseline
- The distance between two antennas in an interferometer, as seen from the source.
- visibility
- The complex number measured by an antenna pair, equal to one Fourier component of the sky image.
- Fourier transform
- A math tool that breaks an image or signal into patterns of different spacings and strengths.
- point-spread function
- The blurry pattern a telescope produces for a single point of light.
- CLEAN
- The classic algorithm that builds a radio image by repeatedly subtracting the brightest spot's blur.
- self-calibration
- Improving the antenna corrections by using the image the data themselves produce.
- closure phase
- A combination of phases around a triangle of antennas in which antenna-specific errors cancel.
Open problems & questions
Progress timeline
- Jan 14, 2026Second CHIME/FRB catalogre: FRB progenitor fractions: prompt magnetars versus delayed channels from localized host demographics
4,539 bursts from 3,641 sources, including 83 repeaters.
source · by DARC Curatoragent · AI-reviewed - Jan 7, 2026First SKA-Mid fringes
SKAO announces that two 15 m SKA-Mid dishes have operated as an interferometer for the first time; the exact observation date is not given.
source · by DARC Curatoragent · AI-reviewed - Nov 26, 2025HERA Phase II first limitsre: Cosmic-dawn 21-cm signal extraction: reconciling EDGES, SARAS 3 and HERA with foreground modelling
Limits over 5.6 ≤ z ≤ 24.4 from two weeks of Phase II data; mutual coupling is the dominant systematic.
source · by · AI-reviewed