New Technique Cleans Up Synthetic Aperture Radar Images

RFI Removal From SAR Imagery via Sparse Parametric Estimation of LFM Interferences | IEEE Journals & Magazine | IEEE Xplore

Progressive Removal of Multiple Interference Components from SAR Imagery

This figure demonstrates the step-by-step process of removing radio frequency interference (RFI) from a Sentinel-1 synthetic aperture radar image of northern Sweden near the Baltic Sea coast.

Panel Breakdown:

Panel (a) - Original Corrupted Image: Shows the raw SAR image with severe RFI contamination. The bright vertical and diagonal streaks are interference artifacts that completely obscure the underlying terrain features.

Panel (b) - Estimated FM Rate Parameters: A plot displaying the estimated azimuth and range frequency modulation (FM) rates for the detected LFM interference components. The blue triangular markers indicate the locations of three distinct interference components in the parameter space.

Panel (c) - First LFM Component: Shows the most dominant interference component isolated by the algorithm. This represents the strongest interfering signal that needs to be removed.

Panel (d) - After First Removal: The cleaned image after removing the most significant LFM component. Notice that while some interference has been eliminated, substantial artifacts remain, revealing the multi-component nature of the RFI.

Panel (e) - Second LFM Component: The second most significant interference pattern isolated from the partially cleaned image.

Panel (f) - After Second Removal: Further improvement in image quality after removing the second interference component. More of the underlying terrain structure becomes visible.

Panel (g) - Third LFM Component: The third interference pattern detected and isolated, showing the weakest but still significant interfering signal.

Panel (h) - Final Cleaned Image: The final result after iteratively removing all three LFM components. The terrain features are now clearly visible, with the bright vertical strips largely eliminated. The underlying scene structure—including what appears to be coastline and land features—is now interpretable.

Key Insights:

This figure illustrates the paper's central contribution: RFI in SAR imagery often consists of multiple overlapping interference components, not just a single source. Previous methods that assumed single-component interference would have stopped after panel (d), leaving substantial artifacts. The iterative sparse parametric estimation approach successfully identifies and removes each component sequentially, dramatically improving image quality and making the data suitable for scientific analysis and operational use.

 

Detecting Multiple Interference Patterns

Advanced sparse recovery algorithm removes radio frequency interference from satellite radar imagery, improving clarity for Earth observation missions

Synthetic aperture radar (SAR) satellites capture detailed images of Earth's surface day and night, through clouds and darkness. But these systems face a persistent problem: radio frequency interference (RFI) from ground-based radars and other SAR satellites creates bright streaks and artifacts that obscure the underlying terrain. Now, researchers have developed a more sophisticated approach to identify and remove these interference patterns from radar images.

A team led by Dehui Yang at Xi'an Jiaotong-Liverpool University and colleagues at Nanjing University of Science and Technology has introduced a method that recognizes RFI as a mixture of multiple linear frequency-modulated (LFM) signal components, rather than treating it as a single interference source. Their technique, published in October 2025 in IEEE Geoscience and Remote Sensing Letters, uses sparse mathematical representations to efficiently estimate the characteristics of each interference component before systematically removing them.

"One of the challenges in spaceborne synthetic aperture radar is modeling and mitigating radio frequency interference artifacts in SAR imagery," the researchers wrote. Previous approaches assumed interference consisted of a single LFM signal, limiting their effectiveness when multiple radio sources create overlapping artifacts in the same image.

The Growing Interference Problem

Radio frequency interference has become increasingly problematic for SAR systems as more satellites share crowded spectrum bands. The European Space Agency's Sentinel-1 satellites, which provide free radar imagery for environmental monitoring and disaster response, frequently encounter interference from ground-based radars and other space-based systems. In one documented case, China's GaoFen-3 satellite and Europe's Sentinel-1A experienced mutual interference when their radar signals crossed paths.

The interference appears as distinctive bright strips in SAR images—linear frequency-modulated signals that sweep across a range of frequencies, creating characteristic patterns. When one SAR satellite's transmitted signal is received by another satellite as unwanted interference, it manifests as two-dimensional LFM artifacts that can completely obscure ground features.

A Sparse Solution

The new approach models RFI as multiple overlapping LFM components, each characterized by specific frequency modulation rates in both the azimuth (along-track) and range (across-track) directions. The key innovation lies in using sparse parametric estimation—a technique that assumes most of a pre-defined "dictionary" of possible interference patterns will have zero contribution, with only a few patterns actually present in any given corrupted image.

"In practical scenarios, each RFI artifact in SAR images typically contains only a few LFM components," the researchers explained. By solving an optimization problem that minimizes the number of active interference patterns while matching the observed data, the algorithm efficiently identifies which specific LFM components are present.

The method splits the estimation process into two computationally efficient steps: first estimating the azimuth frequency modulation rate (which remains constant across all interference components), then separately estimating the range frequency modulation rates (which vary for each component). Once these parameters are identified, the algorithm uses spectral analysis to focus each interference component, applies notch filtering to remove it, and repeats the process iteratively for multiple components.

Superior Performance

Testing on Sentinel-1 single-look complex images from various locations—including Sweden's Baltic Sea coast, Iran's central desert, and Yemen's Red Sea coast—demonstrated clear improvements over existing techniques. The researchers compared their method against three representative approaches: principal component analysis (PCA), robust PCA (RPCA), and an earlier two-dimensional spectral analysis method that assumed single-component interference.

Using the average gradient metric to quantitatively assess image quality (with higher values indicating richer structural detail), the new method achieved scores of 24.82 and 6.23 on two test images, compared to 21.16 and 6.21 for the previous best approach. Visual inspection revealed that competing methods either removed too much legitimate image content along with the interference, over-smoothed the results, or left significant residual artifacts.

In simulation experiments where synthetic interference was added to clean reference images at various signal-to-interference ratios, the new technique consistently achieved the lowest recovery errors across all interference levels. The advantage was particularly pronounced when multiple interference sources were present—the scenario the method was specifically designed to address.

Implications for Earth Observation

The ability to more effectively remove radio frequency interference has practical implications for numerous SAR applications. These radar systems support disaster response by imaging flood-affected areas through cloud cover, monitor deforestation in remote regions, track ice sheet changes, detect ground subsidence, and provide intelligence for both civilian and defense purposes.

As more nations launch SAR satellites and the orbital environment becomes increasingly congested, interference mitigation will become even more critical. The European Space Agency operates two Sentinel-1 satellites as part of its Copernicus program, with Sentinel-1C launched in 2024 to continue the mission. These satellites alone generate thousands of images daily, many of which contain some degree of radio frequency interference.

The new method's computational efficiency makes it practical for operational use. The researchers successfully processed a full Sentinel-1 interferometric wide-swath image measuring 1,488 by 20,546 pixels using blockwise processing, demonstrating scalability to production environments.

Future Developments

The research team indicated that future work will explore algorithms that can better integrate with their LFM model and sparse parametric estimation scheme. Other researchers continue developing complementary approaches, including preprocessing methods applied to raw radar data before image formation and machine learning techniques that could potentially identify interference patterns automatically.

The growing sophistication of interference mitigation techniques reflects both the importance of SAR imagery for Earth observation and the increasing challenges posed by spectrum congestion as satellite constellations proliferate. Each advance in cleaning corrupted radar images extends the scientific utility of these valuable datasets and improves the reliability of applications that depend on them.

---

Sources

1. Yang, D., Xi, F., Cao, Q., & Yang, H. (2025). RFI Removal From SAR Imagery via Sparse Parametric Estimation of LFM Interferences. IEEE Geoscience and Remote Sensing Letters, 22, 4013705. DOI: 10.1109/LGRS.2025.3615975

2. Tao, M., Su, J., Huang, Y., & Wang, L. (2019). Mitigation of radio frequency interference in synthetic aperture radar data: Current status and future trends. Remote Sensing, 11(20), 2438. https://doi.org/10.3390/rs11202438

3. Li, N., Lv, Z., & Guo, Z. (2022). Observation and mitigation of mutual RFI between SAR satellites: A case study between Chinese GaoFen-3 and European Sentinel-1A. IEEE Transactions on Geoscience and Remote Sensing, 60, 5112819. DOI: 10.1109/TGRS.2022.3216819

4. Yang, H., He, Y., Du, T., Zhang, T., Yin, J., & Yang, J. (2022). Two-dimensional spectral analysis filter for removal of LFM radar interference in spaceborne SAR imagery. IEEE Transactions on Geoscience and Remote Sensing, 60, 5219016. DOI: 10.1109/TGRS.2022.3146575

5. Yang, H., Tao, M., Chen, S., Xi, F., & Liu, Z. (2020). On the mutual interference between spaceborne SARs: Modeling. IEEE Transactions on Geoscience and Remote Sensing, 59(10), 8470-8485. DOI: 10.1109/TGRS.2020.3011076

6. Yang, H., Li, K., Li, J., Du, Y., & Yang, J. (2022). BSF: Block subspace filter for removing narrowband and wideband radio interference artifacts in single-look complex SAR images. IEEE Transactions on Geoscience and Remote Sensing, 60, 5211916. DOI: 10.1109/TGRS.2021.3138804

7. Lu, X., Yang, J., Yu, W., Su, W., Gu, H., & Yeo, T. S. (2021). Enhanced LRR-based RFI suppression for SAR imaging using the common sparsity of range profiles for accurate signal recovery. IEEE Transactions on Geoscience and Remote Sensing, 59(2), 1302-1318. DOI: 10.1109/TGRS.2020.3003492