Sequential Covariance Fitting for InSAR Phase Linking
Dana EL HAJJAR, Guillaume GINOLHAC, Yajing YAN, and Mohammed Nabil EL KORSO · 2025
Read Sequential Covariance Fitting for InSAR Phase Linking by Dana EL HAJJAR, Guillaume GINOLHAC, Yajing YAN, and Mohammed Nabil EL KORSO (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-2502-09248.txt.
Introduction
Four authors led by Dana El Hajjar speed up a radar image step that lines up wave phases across a stack of shots. Satellites such as Sentinel-1 now send down huge stacks of radar images of the same ground. Phase linking cleans the stack so ground shift can be read with care. Classic tools for the step cost too much at this scale, and a leaner frame still leans on the full data table at once.
Their fix folds each new block of images into the fit in turn, not the whole stack at once. The math works on a torus shape, since phase values wrap around a full turn. A stepwise shrink method keeps each update sound and fast. New shots can thus join the fit the day they land. Ground watch teams gain a tool that keeps pace with the fleet.
Catalog record
| CREATOR | Dana EL HAJJAR, Guillaume GINOLHAC, Yajing YAN, and Mohammed Nabil EL KORSO |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 73 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-09248
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Sequential Covariance Fitting for InSAR Phase Linking Dana EL HAJJAR, Guillaume GINOLHAC, Yajing YAN, and Mohammed Nabil EL KORSO Abstract Traditional Phase-Linking algorithms are known for their high cost, especially with the huge volume of Synthetic Aperture Radar images generated by Sentinel- 1 1 Synthetic Aperture Radar missions. Recently, a COvariance Fitting Interferometric Phase Linking approach has been proposed, which can be seen as a generic framework for existing Phase-Linking methods. Although this method is less computationally expensive than traditional PL approaches, COvariance Fitting Interferometric Phase Linking exploits the entire covariance matrix, which poses a challenge with the increasing time series of Synthetic Aperture Radar images. However, COvariance Fitting Interferometric Phase Linking, like traditional Phase-Linking approaches, cannot accommodate the efficient inclusion of newly acquired Synthetic Aperture Radar images. This paper overcomes this drawback by introducing a sequential integration of a block of newly acquired Synthetic Aperture Radar images. Specifically, we propose a method for effectively addressing optimization problems associated with phase-only complex vectors on the torus based on the Majorization-Minimization framework. Notations Notations Description \circ element-wise (Hadamard) product .^{H} transposed and complex conjugated operator (Hermitian) .^{T} transposed operator \widetilde{.} tilde accentuation representing the entire dataset \bar{.} bar accentuation representing the new items .^{*} complex conjugated operator \textit{Re}(.) complex number real component \text{tr}(.) trace matrix operator \text{log}(.) logarithm operator \mathbb{E}(.) first-order moment operator ||.||^{2}_{F} frobenius norm |.| modulus operator I Introduction InSAR techniques have seen a great development and improvement with the immense number of available Synthetic Aperture Radar (Synthetic Aperture Radar) images [ 1 ] , thanks to ongoing remote sensing missions such as Sentinel -1 , TerraSynthetic Aperture Radar-X, etc. Interferometric Synthetic Aperture Radar (InSAR) is widely used to monitor land movements for various applications such as volcano monitoring, natural disaster damages, urban development. The first InSAR techniques involved using two SAR images acquired at two different dates ( 2 -pass InSAR) to estimate ground movements. However, when the dates are far apart, the coherence between the two images Acronyms CCG Complex Circular Gaussian COFI-PL COvariance Fitting Interferometric Phase Linking COMET COvariance Matching Estimation Technique DS Distributed Scatterer DSI Distributed Scatterer Interferometry EVD EigenValue Decomposition ILS Integer Least Square InSAR Interferometric Synthetic Aperture Radar KL Kullback-Leibler MLE Maximum Likelihood Estimator MLE-PL Phase Linking based on Maximum Likelihood Estimation MM Majorization-Minimization MSE Mean Squared Error MT-InSAR Multi-Temporal Interferometric Synthetic Aperture Radar PCA Principal Component Analysis PL Phase-Linking PO Phase-Only Sample Covariance Matrix PS Permanent Scatterer PSI Permanent Scatterer Interometry RGD Remannian Gradient Descent SAR Synthetic Aperture Radar SBAS Small BAseline Subset SCM Sample Covariance Matrix S-COFI-PL Sequential COvariance Fitting Interferometric Phase Linking S-MLE-PL Sequential Phase Linking based on Maximum Likelihood Estimation weakens, resulting in a significant information loss. As a result, leveraging the vast availability of Synthetic Aperture Radar images from various missions, time-series-based approaches have been proposed to exploit the temporal correlation among these images. This concept can be summarized as Multi-Temporal Interferometric Synthetic Aperture Radar (MT-InSAR). An essential family in Multi-Temporal Interferometric Synthetic Aperture Radar analysis is called Distributed Scatterer Interferometry (Distributed Scatterer Interferometry) which exploits groups of homogeneous scatterers called Distributed Scatterer ( DS ). A famous approach in Distributed Scatterer Interferometry, called Phase-Linking ( PL ) [ 2 , 3 , 4 , 5 , 6 ] , consists to retrieve Synthetic Aperture Radar image phases from the full Synthetic Aperture Radar images time series. For an overview of existing Phase-Linking algorithms, refer to [ 7 , 8 ] . The general principle consists of two steps: first, estimating the Sample Covariance Matrix (Sample Covariance Matrix) and using its modulus as a plug-in of the unknown coherence matrix, and second, estimating the phases vector of Synthetic Aperture Radar images. However, this plug-in is non optimal due to the fact that it does not represent the Maximum Likelihood Estimator (Maximum Likelihood Estimator) of the coherence matrix, since the modulus operator is not holomorphic. The second approach, introduced in [ 5 , 6 , 9 ] , presents the true Maximum Likelihood Estimator of the coherence matrix jointly with the phases vector and will be referenced as Phase Linking based on Maximum Likelihood Estimation, for Phase Linking based on Maximum Likelihood Estimation, in the remaining of the paper. For both approaches, the process is based on the optimization of the negative log-likelihood function of the data following a zero mean Complex Circular Gaussian ( CCG ) distribution, which may not always fit the data. Studies such as [ 10 , 9 ] have shown that Synthetic Aperture Radar images are better modeled by a heavy-tailed distribution than a Gaussian distribution. Some works developed versions of the Phase-Linking based on a Non-Gaussian distribution [ 9 ] which demonstrated an improvement in the results. These algorithms are computationally intensive as they rely on the full covariance matrix estimation and consist of an iterative algorithm with two steps [ 5 , 9 ] : i) estimating the phases vector, with an iterative Majorization-Minimization (Majorization-Minimization) algorithm, and ii) estimating the coherence matrix. These two steps are also part of an iterative algorithm called Block Coordinate Descent (Block Coordinate Descent). Additionally, in the non-Gaussian case, the estimation of the covariance matrix is computationally intensive, as it involves Tyler-type estimators [ 9 ] , for which the estimation process itself is recursive. From above, it is clear that the estimation of the covariance matrix plays an important role. An alternative of the Maximum Likelihood Estimator approach was proposed in [ 11 ] , called COvariance Matching Estimation Technique (COvariance Matching Estimation Technique), often requiring less computational effort while maintaining similar asymptotic properties. They consist of a set of statistical methods used for parameter estimation by matching the empirical and theoretical covariance matrices. They gained attention in the signal processing field due to their efficiency in solving various estimation problems. Covariance fitting approach was used in InSAR in [ 12 ] , namely COvariance Fitting Interferometric Phase Linking ( COFI-PL ). The main idea is to refine the structure of a covariance matrix estimator by minimizing a projection criterion. COvariance Fitting Interferometric Phase Linking can be seen as a generalization of most of the existing Phase-Linking problems [ 12 ] while maintaining a lower cost than traditional Phase-Linking methods. The problem centers on two key components: the chosen method for covariance matrix estimation and the selected matrix distance. These factors can significantly influence the overall performance. Various covariance matrix plug-in can be used such as the Sample Covariance Matrix, the Phase-Only Sample Covariance Matrix (Phase-Only Sample Covariance Matrix), Tyler estimator, etc. Additionally, regularization techniques can also be adopted. Although theoretically less effective, COFI-PL has demonstrated strong and promising results in practice with real-world data [ 12 ] . A key advantage of this approach is its robustness, which stems from to the choice of the covariance matrix plug-in. However, when a new image or several new images are added, traditional methods become computationally intensive because they depend on the estimation of the full covariance matrix. This matrix increases with each new Synthetic Aperture Radar image, necessitating a complete re-estimation every time a new image is introduced. As a result, traditional methods struggle to manage this growth, imposing significant constraints on real-time processing of SAR image