How does the Hogbom CLEAN algorithm work step by step?

Step 1: Start with the dirty image (inverse Fourier transform of the measured visibilities). Step 2: Find the pixel with the highest absolute value. Step 3: Subtract a scaled copy of the dirty beam (PSF) centered at that pixel, where the scale factor (loop gain) is typically 0.1 to 0.3 of the peak value. Step 4: Record the position and amplitude of the subtracted component in a list of clean components. Step 5: Repeat steps 2-4 until the maximum residual falls below a threshold (typically the noise level) or a maximum number of iterations is reached. Step 6: Convolve the clean component list with the clean beam (a Gaussian fit to the main lobe of the dirty beam) and add the final residual image to produce the clean image.

What variants of CLEAN exist for RF imaging?

Clark CLEAN separates the process into minor cycles (operating on a smaller patch in image space for speed) and major cycles (subtracting in visibility space for accuracy). Cotton-Schwab CLEAN improves accuracy by degridding clean components back to visibility space. Multi-scale CLEAN represents sources as extended Gaussians of different sizes instead of point sources, better handling resolved emission. Multi-frequency synthesis CLEAN combines data across bandwidth to improve UV coverage. W-projection CLEAN handles wide-field non-coplanar baseline effects. For radar applications, iterative adaptive CLEAN variants use spatially varying PSFs to handle range-dependent beam patterns in SAR and ISAR imagery.

Signal Processing & Imaging

CLEAN Algorithm

Q: Where is CLEAN used outside radio astronomy?

CLEAN is widely applied in ISAR (inverse synthetic aperture radar) imaging for ship and aircraft recognition, where sidelobe artifacts from limited aperture integration angles obscure target features. It is used in SAR image enhancement, phased array direction-of-arrival estimation (spatial spectrum cleaning), acoustic imaging (microphone array beamforming), and medical ultrasound imaging. In antenna measurement, CLEAN deconvolves the probe pattern from spherical near-field measurements to improve the accuracy of far-field pattern reconstruction. The algorithm's simplicity and robustness make it applicable whenever the imaging process can be modeled as convolution with a known PSF.

/kleen al-guh-rith-uhm/

An iterative deconvolution method developed by Jan Högbom (1974) for removing point spread function (PSF) sidelobe artifacts from radio interferometric images. The algorithm finds the peak in the "dirty image," subtracts a fraction (gain 0.1 to 0.3) of the PSF centered at that peak, records the component in a model, and repeats until the residual reaches the noise floor. The final clean image is the model convolved with an idealized Gaussian beam plus the residual. Also applied in SAR/ISAR radar imaging, phased array beamforming, and acoustic imaging.

Category: Signal Processing & Imaging

Developed: 1974 (Högbom)

Loop Gain: 0.1 to 0.3

Understanding the CLEAN Algorithm

In radio interferometry, an array of antennas samples the spatial frequency (UV) plane at discrete points determined by the baseline geometry. The inverse Fourier transform of these incomplete UV samples produces the "dirty image," which is the true sky brightness convolved with the "dirty beam" (the PSF of the incomplete sampling). The dirty beam has a narrow main lobe (resolution element) but strong sidelobes extending across the entire image, creating artifacts that can obscure faint sources near bright ones.

CLEAN solves this deconvolution problem by modeling the sky as a collection of point sources. At each iteration, the brightest remaining pixel is identified, a scaled version of the dirty beam is subtracted from its location, and the source position and flux are recorded. This process is remarkably robust despite its simplicity: it converges for most radio images because astronomical sources are sparse (few bright sources in a large field). The loop gain (fraction subtracted per iteration, typically 0.1 to 0.3) controls stability versus convergence speed. Lower gains are more stable but require more iterations. After convergence, the clean components are convolved with the "clean beam" (a Gaussian fitted to the dirty beam's main lobe), producing an image with correct resolution but without sidelobe artifacts. The algorithm extends naturally to radar imaging where limited aperture creates similar sidelobe problems in ISAR target images and SAR stripmap/spotlight modes.

CLEAN Algorithm Parameters

Dirty Image:
I_dirty(x,y) = I_true(x,y) * B_dirty(x,y)

Iteration Step:
I_residual⁽ⁿ⁺¹⁾ = I_residual⁽ⁿ⁾ - γ · I_max⁽ⁿ⁾ · B_dirty(x - x_max, y - y_max)

Final Clean Image:
I_clean = ∑ C_i · B_clean(x - x_i, y - y_i) + I_residual^(final)

Where * = convolution, γ = loop gain (0.1 to 0.3), C_i = clean component flux, B_clean = Gaussian fit to dirty beam main lobe. Typical: 1,000 to 100,000 iterations.

CLEAN Variants

Variant	Key Improvement	Application
Högbom (1974)	Original point-source model	Compact radio sources
Clark (1980)	Minor/major cycle acceleration	Large images
Cotton-Schwab	Visibility-space subtraction	High dynamic range
Multi-scale	Extended source Gaussians	Resolved emission
MFS CLEAN	Multi-frequency synthesis	Wideband imaging

Common Questions

Frequently Asked Questions

How does Högbom CLEAN work step by step?

Find peak in dirty image, subtract γ×peak×PSF at that location, record component, repeat until residual reaches noise. Final image: convolve clean components with Gaussian beam + residual. Loop gain of 0.1 to 0.3 balances stability and speed. Typically 1,000 to 100,000 iterations.

What CLEAN variants exist?

Clark (minor/major cycles for speed), Cotton-Schwab (visibility-space accuracy), multi-scale (extended sources as Gaussians), MFS (multi-frequency UV improvement), W-projection (wide-field correction). Radar uses iterative adaptive variants with spatially varying PSFs.

Where is CLEAN used outside radio astronomy?

ISAR imaging (ship/aircraft recognition), SAR enhancement, phased array DOA estimation, acoustic microphone array imaging, medical ultrasound, and antenna near-field measurement deconvolution. Applicable wherever imaging involves convolution with a known PSF.

Related Terms

← Clausius-Mossotti Cleanroom →