MROP: Modulated Rank-One Projections for compressive radio interferometric imaging 2025 MROP: Modulated Rank-One Projections for compressive radio interferometric imaging – References
Olivier Leblanc, Chung San Chu, Laurent Jacques, Yves Wiaux · 2025
Read MROP: Modulated Rank-One Projections for compressive radio interferometric imaging 2025 MROP: Modulated Rank-One Projections for compressive radio interferometric imaging – References by Olivier Leblanc, Chung San Chu, Laurent Jacques, Yves Wiaux (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-18446.txt.
Introduction
Olivier Leblanc and co-authors squeeze radio sky data at the dish, ere it floods the bank, in a study from 2025. New radio arrays will see the sky with wild depth and grit, but their pair counts grow as the square of dish count times span count. Raw haul in that bulk can not ride.
Their squeeze, MROP, swaps each batch's full pair grid for a set of lean rank-one casts, then folds spans by chance tunes. Two reads frame it: chance beam aim at the dish, or squeeze past the pair step. Chance casts set noise flat cross all reads, sans the weight scheme. Costs in store and math are writ out for real builds.
Catalog record
| CREATOR | Olivier Leblanc, Chung San Chu, Laurent Jacques, Yves Wiaux |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 113 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-18446
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
MROP: Modulated Rank-One Projections for compressive radio interferometric imaging 2025 MROP: Modulated Rank-One Projections for compressive radio interferometric imaging – References Olivier Leblanc, Chung San Chu, Laurent Jacques, Yves Wiaux Abstract The emerging generation of radio-interferometric (RI) arrays are set to form images of the sky with a new regime of sensitivity and resolution. This implies a significant increase in visibility data volumes, scaling as 𝒪 ( Q 2 B ) \mathcal{O}(Q^{2}B) for Q Q antennas and B B short-time integration intervals (or batches), calling for efficient data dimensionality reduction techniques. This paper proposes a new approach to data compression during acquisition, coined modulated rank-one projection (MROP). MROP compresses the Q × Q Q\times Q batchwise covariance matrix into a smaller number P P of random rank-one projections and compresses across time by trading B B for a smaller number M M of random modulations of the ROP measurement vectors. Firstly, we introduce a dual perspective on the MROP acquisition, which can either be understood as random beamforming, or as a post-correlation compression. Secondly, we analyse the noise statistics of MROPs and demonstrate that the random projections induce a uniform noise level across measurements independently of the visibility-weighting scheme used. Thirdly, we propose a detailed analysis of the memory and computational cost requirements across the data acquisition and image reconstruction stages, with comparison to state-of-the-art dimensionality reduction approaches. Finally, the MROP model is validated in simulation for monochromatic intensity imaging, with comparison to the classical and baseline-dependent averaging (BDA) models, and using the uSARA optimisation algorithm for image formation. An extensive experimental setup is considered, with ground-truth images containing diffuse and faint emission and spanning a wide variety of dynamic ranges, and for a range of u v uv -coverages corresponding to VLA and MeerKAT observation. 1 Introduction Data acquisition by interferometry in radio astronomy relies on an array of strategically placed antennas measuring, through the signal covariance matrix, a noisy and nonuniform partial coverage of the spatial Fourier domain of the image of interest, i.e. , a small portion of the celestial sphere accessible to the instrument. Radio interferometry (RI) is fundamental in studying a wide range of astronomical phenomena, including star formation, galaxy evolution, and the cosmic microwave background. The classical model stores in memory all the visibilities—the Fourier samples obtained from the interferometric acquisition. While current arrays like Very Large Array (VLA) ( Thompson et al. 1980 ) or MeerKAT ( Asad et al. 2021 ) respectively have 27 and 64 antennas, future arrays such as the Square Kilometer Array (SKA) are coming up with an increasing number of antennas. SKA itself is to form petabyte-scale ( 10^{15} bytes) images from exabyte-scale ( 10^{18} bytes) data volumes ( Scaife 2020 ) . Compression is thus emerging as a critical need to reduce the data size and make RI imaging scalable. Recently, Leblanc et al. 2024 proposed a compressive sensing technique applied directly at the level of the antenna measurements. It showed that beamforming —a common technique of dephasing antenna signals—usually used to focus some region of the sky, is equivalent to sensing a rank-one projection (ROP) of the signal covariance matrix. Random beamforming was combined with random modulations across time to build the modulated rank-one projection (MROP) model that replaces the enormous number of visibilities with a tunable number of MROPs. In Leblanc et al. 2024 , the emphasis was made on the mathematical foundations of the random beamforming. Image recovery guarantees were provided for image reconstruction under a simplistic model assuming sparsity in the image domain. The associated reduced sample complexity, theoretically derived in a simpler case without modulations and numerically obtained in phase transition diagrams, was shown to scale as \mathcal{O}(K) where K is the image sparsity. This paper significantly extends the study of the MROP model while still focussing on monochromatic intensity imaging. Firstly, we discuss two acquisition schemes. The antenna-based random beamforming acquisition scales linearly with the number of antennas, while the visibility-based acquisition is closer to the classical correlation computation and facilitates visibility weighting. Secondly, the noise statistics of MROPs are derived. Thirdly, the memory requirements and the computational costs of the acquisition and imaging steps are derived and compared with other models in the literature. Finally, the reconstruction of images with high dynamic range in a realistic numerical setting demonstrates the potential of the MROP model. 1.1 Related Work Several compression techniques for RI that are connected to our method exist in the literature. Among them, compressing the observation vector that collects all the visibilities by multiplication with a Gaussian random matrix (with much fewer rows than columns) was discussed in Kartik et al. 2017 and shown to increase the computational cost of the forward model significantly. Kartik et al. 2017 proposed an efficient Fourier reduction model approximating the optimal (in the least-squares sense) dimensionality reduction that projects the data with the adjoint of \bm{U} which contains the left singular vectors of the SVD of the classical imaging operator \bm{\Phi}:=\bm{U}\bm{\Sigma}\bm{V}^{*} . On the other hand, baseline-dependent averaging (BDA) ( Wijnholds et al. 2018 ; Atemkeng et al. 2018 ) offers an effective reduction in data volume by averaging consecutive visibilities corresponding to short baselines, i.e. , those with nearly constant frequency locations. Finally, standard visibility gridding, which consists of storing the dirty image , is an efficient data dimensionality reduction technique when the original visibility data size is larger than the image size, as discussed in Kartik et al. 2017 . The dirty map can even be computed in an antenna-based way which reduces its acquisition cost ( Thyagarajan et al. 2017 ; Krishnan et al. 2023 ) . Random beamforming for RI has first been suggested in Öçal et al. 2015 . Then, the connection of random beamforming to ROPs of the signal covariance matrix has been highlighted in Leblanc et al. 2024 , leveraging the theory of recovery guarantees via ROP to derive the sample complexity in a simplified case. This paper studies image reconstruction using the uSARA algorithm ( Repetti & Wiaux 2020 ; Terris et al. 2022 ) , which is a sophisticated version of the proximal gradient method ( Beck 2017 ) . 1.2 Contributions In this work, we significantly increase the practicality of the MROP approach proposed by Leblanc et al. 2024 to RI by providing four novel contributions, all studied in the context of monochromatic intensity imaging. Acquisition of MROPs. Two perspectives on acquiring MROPs are discussed in Sections 3.2 and 3.3 on the grounds of their acquisition costs, their compatibilities with the existing correlators, and their compatibilities with visibility weighting. The first perspective, involving random beamforming , scales linearly with the number of antennas but is shown to raise challenges for visibility weighting because the visibilities are never accessed. A possible solution that decomposes the matrix of weights as a sum of rank-one matrices is proposed in Section 3.2.2 . The second perspective, coined visibility-based acquisition of MROP, applies post-correlation compression so accesses the visibilities, hence enabling their weighting, and is also compatible with the existing correlators. Noise statistics on MROPs. In Section 3.2.3 , we conduct a statistical analysis showing that MROP computation preserves the additive i.i.d. Gaussian nature