A Unified Maximum-Likelihood Framework for 3D InISAR Phase Unwrapping with Outlier Rejection
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Electrical Engineering. The day in electrical engineering: signals, systems and the hardware that carries them.
Rosa: I'm Rosa.
Dev: And I'm Dev, here with Taro.
Rosa: We're looking at the paper: "A Unified Maximum-Likelihood Framework for 3D InISAR Phase Unwrapping with Outlier Rejection".
Dev: Here's the gist of the paper. A novel mathematical framework for phase unwrapping in three-dimensional interferometric ISAR (3D InISAR) imaging is presented, which provides a unified way to handle generic sensor geometries and produces a natural a posteriori quality metric for outlier rejection The paper introduces a novel mathematical framework for phase unwrapping in three-dimensional interferometric ISAR (3D InISAR) imaging This approach works on a scatterer-by-scatterer basis and does not rely on any spatial continuity assumptions, making it suitable for sparse point clouds The formulation is derived from the Mixed-Integer Least Squares (MILS) theory, an optimal maximum-likelihood framework for joint estimation of integer and real unknowns in the presence of Gaussian noise This provides a unified way to handle generic sensor geometries, multi-baseline, multi-frequency, or hybrid setups MILS Theoretical Formulation The observation model (4) is written by renaming variables as y = Aa + Bb + e where y ≡ ∆ϕ ∈ R m is the observation vector The optimal way to estimate a and b is through maximum-likelihood estimation (MLE): L(a, b y):= ∥y − Aa − Bb∥ squared Q-1 yy for the joint estimation of integer and real unknowns The operational algorithm consists of two steps: (i) estimate the integer vector aˆ using (9); (ii) estimate the real vector bˆ using (8) with the resulting aˆ Search Space and Constraints To obtain a practical implementation, the search space of the integer variables must be bounded by prior knowledge of maximum target size A constraint on the real unknowns can be imposed by requiring b to lie in a bounded set B ⊆ R p, specifically using simple box constraints ξ1 ≤ Lmax squared and ξ3 ≤ Lmax squared The set A is then defined as the set of all a ∈ Z n satisfying the constraint (11) and producing a bˆ(a) from (8) within B A Posteriori Quality Assessment The quality of the integer solution can be evaluated using the posterior probability of aˆ given the data, defined as P(a y) ∝ R b∈B P(y a, b) db This leads to a natural a posteriori quality measure defined as AP:= P(aˆ y) The Ambiguity Posterior (AP) value lies in
0, 1: and expresses how likely the estimated integer solution is correct Outlier Rejection Criterion The AP value can be used to define a statistical test to accept or reject a scatterer by setting Accept if AP ≥ APthr The Acceptance Rate (AccR) is defined as P(AP ≥ APthr), and the Conditional Failure Rate (CoFaR) is P(aˆ ≠ a AP ≥ APthr) Two strategies can be used to set the threshold: 1) Fixed-threshold, using a constant value like APthr = 90%, or 2) Fixed-CoFaR, where the corresponding APthr for each scatterer is computed based on its SNR Simulation Results and Validation Monte Carlo simulations show that when APthr increases, more scatterers are rejected, and the AccR gradually decreases until it reaches values close to zero The results indicate that for the considered configuration to work reliably, the SNR must be above roughly 25 dB The final reconstruction after the accept/reject step shows a significant improvement, with the RMSE computed only on correctly unwrapped scatterers being 0.10 m This demonstrates that the proposed approach can achieve reliable 3D reconstruction even in challenging ambiguity conditions Future Work Directions Several future developments are possible, including designing smarter and more efficient strategies to define the search space of the integer unknowns, studying the AccR, CoFaR, and RMSE metrics for different 3D InISAR systems with various geometries, and assessing 3D InISAR reconstruction quality using more advanced and realistic simulations The paper also suggests a possible post-processing strategy to further reduce the impact of residual outliers based on local point density References
1: M.
Rosa: Martorella, D.
Dev: Stagliano, F.
Rosa: Salvetti, and N.
Paper summary: Rosa: We just talked about how this paper sets up a unified framework for phase unwrapping in three dee InISAR imaging <ref:2605.28431#pg1,framework for phase unwrapping in>. The main thesis is that by using Mixed-Integer Least Squares theory, you can jointly estimate the integer and real variables in the presence of noise.
Dev: The authors are arguing that this lets us handle generic sensor geometries, which means it’s not locked into one specific setup like a simple two-baseline scenario.
Rosa: And they claim that by doing this, we can get an a posteriori quality metric for each phase solution. This is key because it lets you decide whether to accept or reject a scatterer measurement statistically.
Dev: So, the point is that this method offers an optimal way to estimate those unknowns, and it naturally gives us a measure of how likely the integer solution actually is correct.
Rosa: It matters because in real-world scenarios, you’re going to have noise and bad data points in your interferometric measurements. This framework gives you a statistical tool to filter out those outliers before they mess up your whole three dee reconstruction <ref:2605.28431#pg1>.
Dev: It moves the problem from just finding *a* solution to finding the most probable solution, and then using that probability to make informed decisions about what data is trustworthy.
Rosa: So, it’s about moving towards a system that’s not just mathematically sound for ideal cases but is also practical for messy, real-world three dee InISAR data <ref:2605.28431#pg1>.
Dev: And this sets the stage for better reliability in target characterization using these types of imaging systems.
Rosa: So, next up, we're going to look at what they actually found when they ran their simulations.
Conclusion: Dev: So, wrapping up this discussion on "A Unified Maximum-Likelihood Framework for three dee InISAR Phase Unwrapping with Outlier Rejection," the core contribution is that it provides a way to handle phase unwrapping without needing spatial continuity assumptions <ref:2605.28431#pg1,A Unified Maximum-Likelihood Framework for 3D InISAR Phase Unwrapping with Outlier>.
Rosa: The authors show that their method can be used with various sensor setups, like multi-baseline or hybrid ones. It’s not just for one specific geometry anymore.
Dev: And the real practical part is the addition of that a posteriori quality metric, which lets you use a statistical test to reject outliers based on how likely the solution is.
Rosa: This means for someone listening who only cares about getting a usable three dee map, this approach suggests that you can get a much cleaner reconstruction even when your raw data has some noise or errors <ref:2605.28431#pg1>.
Dev: It implies that for these complex three dee InISAR systems, we don't have to just hope the initial unwrapped phase is right; we have a principled way to check its reliability <ref:2605.28431#pg1>.
Rosa: The authors suggest that if you tune the threshold for accepting a scatterer based on this quality metric, you can control your final reconstruction accuracy.
Dev: So, it’s about giving engineers and researchers a tool to manage uncertainty directly in the unwrapping process for these types of imaging systems.
Matteo Pardi, Francesco Mancuso, Elisa Giusti, Marco Martorella
Department of Information Engineering, University of Pisa · Radar and Surveillance Systems National Laboratory, CNIT, Pisa
eess.SP
Submitted: 2026-05-27
Updated: 2026-05-27
Comments: 6 pages, 4 figures. Accepted for publication in the Proceedings of the 2026 IEEE Radar Conference (RadarConf26)
Journal ref: 2026 IEEE Radar Conference (RadarConf'26)
DOI: 10.1109/RadarConf2663773.2026.11674931
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 77/100
The gist: A novel mathematical framework for phase unwrapping in three-dimensional interferometric ISAR (3D InISAR) imaging is presented…
Key concepts
- Mixed-Integer Least Squares (MILS)
- This is an optimal framework for finding both integer and real values simultaneously when dealing with noisy data. It's used here to estimate the unknown integer phase shifts and continuous real parameters in the radar measurement model, ensuring a mathematically sound joint estimation.
- Ambiguity Posterior (AP)
- The AP value measures how likely the estimated integer solution is correct, ranging from 0 to 1. It is calculated by integrating the probability of observing the data given a specific integer solution over all possible real parameter values. A high AP suggests a reliable estimate.
- Outlier Rejection Criterion
- This criterion uses the AP value to decide whether an estimated scatterer should be kept or discarded. If the AP is below a set threshold (APthr), the scatterer is rejected. This statistical test helps filter out erroneous measurements, improving reconstruction accuracy.
- Scatterer-by-Scatterer Basis
- The framework analyzes the imaging process based on individual scatterers rather than relying on assumptions about spatial continuity between them. This makes it highly effective for sparse point clouds and allows it to be applied to generic sensor geometries.
Terminology
Summary
A novel mathematical framework for phase unwrapping in three-dimensional interferometric ISAR (3D InISAR) imaging is presented, which provides a unified way to handle generic sensor geometries and produces a natural a posteriori quality metric for outlier rejection
Introduction to the Problem
The paper introduces a novel mathematical framework for phase unwrapping in three-dimensional interferometric ISAR (3D InISAR) imaging This approach works on a scatterer-by-scatterer basis and does not rely on any spatial continuity assumptions, making it suitable for sparse point clouds The formulation is derived from the Mixed-Integer Least Squares (MILS) theory, an optimal maximum-likelihood framework for joint estimation of integer and real unknowns in the presence of Gaussian noise This provides a unified way to handle generic sensor geometries, multi-baseline, multi-frequency, or hybrid setups
MILS Theoretical Formulation
The observation model (4) is written by renaming variables as y = Aa + Bb + e where y ≡ ∆ϕ ∈ R m is the observation vector The optimal way to estimate a and b is through maximum-likelihood estimation (MLE): L(a, b y):= ∥y − Aa − Bb∥ squared Q-1 yy for the joint estimation of integer and real unknowns The operational algorithm consists of two steps: (i) estimate the integer vector aˆ using (9); (ii) estimate the real vector bˆ using (8) with the resulting aˆ
Search Space and Constraints
To obtain a practical implementation, the search space of the integer variables must be bounded by prior knowledge of maximum target size A constraint on the real unknowns can be imposed by requiring b to lie in a bounded set B ⊆ R p, specifically using simple box constraints ξ1 ≤ Lmax squared and ξ3 ≤ Lmax squared The set A is then defined as the set of all a ∈ Z n satisfying the constraint (11) and producing a bˆ(a) from (8) within B
A Posteriori Quality Assessment
The quality of the integer solution can be evaluated using the posterior probability of aˆ given the data, defined as P(a y) ∝ R b∈B P(y a, b) db This leads to a natural a posteriori quality measure defined as AP:= P(aˆ y) The Ambiguity Posterior (AP) value lies in [0, 1] and expresses how likely the estimated integer solution is correct
Outlier Rejection Criterion
The AP value can be used to define a statistical test to accept or reject a scatterer by setting Accept if AP ≥ APthr The Acceptance Rate (AccR) is defined as P(AP ≥ APthr), and the Conditional Failure Rate (CoFaR) is P(aˆ ≠ a AP ≥ APthr) Two strategies can be used to set the threshold: 1) Fixed-threshold, using a constant value like APthr = 90%, or 2) Fixed-CoFaR, where the corresponding APthr for each scatterer is computed based on its SNR
Simulation Results and Validation
Monte Carlo simulations show that when APthr increases, more scatterers are rejected, and the AccR gradually decreases until it reaches values close to zero The results indicate that for the considered configuration to work reliably, the SNR must be above roughly 25 dB The final reconstruction after the accept/reject step shows a significant improvement, with the RMSE computed only on correctly unwrapped scatterers being 0.10 m This demonstrates that the proposed approach can achieve reliable 3D reconstruction even in challenging ambiguity conditions
Future Work Directions
Several future developments are possible, including designing smarter and more efficient strategies to define the search space of the integer unknowns, studying the AccR, CoFaR, and RMSE metrics for different 3D InISAR systems with various geometries, and assessing 3D InISAR reconstruction quality using more advanced and realistic simulations The paper also suggests a possible post-processing strategy to further reduce the impact of residual outliers based on local point density
References
[1] M. Martorella, D. p. 0 Stagliano, F. p. 5 Salvetti, and N. p. 5 Battisti, “3D Interferometric ISAR Imaging of Non-Cooperative Targets,” IEEE Transactions on Aerospace and Electronic Systems, vol. p. 5 50, no. 4, 2014
[2] X. Xu and R. M. Narayanan, “Three-dimensional Interferometric ISAR Imaging for Target Scattering Diagnosis and Modeling,” IEEE Transactions on Image Processing, vol. p. 5 10, no. 7, 2001
[3] B. Tian, Z. Lu, Y. Liu, and X. p. 5 Li, “Review on Interferometric ISAR 3D Imaging: Concept, Technology and Experiment,” Signal Processing, vol. p. 5 153, 2018
[4] F. Salvetti, M. p. 5 Martorella, E. p. 0 Giusti, and D. p. 0 Stagliano, “Multiview Threedimensional Interferometric Inverse Synthetic Aperture Radar,” IEEE Transactions on Aerospace and Electronic Systems, vol. p. 5 55, no. 2, 2018
[5] W. Jiang, Y. Wang, Y. Li, Y. Lin, and W. p. 5 Shen, “Radar Target Characterization and Deep Learning in Radar Automatic Target Recognition: A Review,” Remote Sensing, vol. p. 5 15, no. 15, 2023
[6] G. Meucci, E. Giusti, A. Kumar, F. Mancuso, S. p. 0 Ghio, and M. p. 5 p. 5 Martorella, “Transformer-based Automatic Target Recognition for 3DInISAR,” IEEE Transactions on Radar Systems, 2025
[7] C. Y. Pui, B. Ng, L. Rosenberg, and T. p. 5-T. Cao, “Target classification for 3D-ISAR using CNNs,” IEEE Transactions on Aerospace and Electronic Systems, vol. p. 5 60, no. 1, 2023
[8] E. Giusti, S. Ghio, and M. Martorella, “Drone-based 3D Interferometric ISAR Imaging,” in 2021 IEEE Radar Conference (RadarConf21). p. 5 IEEE, 2021
[9] B. W.-H. Ng and H.-T. Tran, “Long-baseline 3D Interferometric ISAR,” in 2017 IEEE Radar Conference (RadarConf). p. 5 IEEE, 2017
[10] F. Mancuso, E. p. 0 Giusti, B. Ng, and M. Martorella, “Dual-frequency Phase Unwrapping for 3D InISAR Imaging of Non-Cooperative Targets,” IEEE Transactions on Radar Systems, vol. p. 5 2, 2024
[11] H. Yu, Y. Lan, Z. Yuan, J. Xu, and H. Lee, “Phase Unwrapping in InSAR: A Review,” IEEE Geoscience and Remote Sensing Magazine, vol. p. 5 7, no. 1, 2019
[12] Z. Lin, Y. Duan, Y. Deng, W. Tian, and Z. p. 5 Zhao, “An Improved Multibaseline Phase Unwrapping Method for GB-InSAR,” Remote Sensing, vol. p. 5 14, no. 11, 2022
[13] M. Costantini, F. p. 5 Minati, F. Vecchioli, and M. p. 5 Zavagli, “An Efficient Approach to Spatio-Temporal 3D Phase Unwrapping,” in IGARSS 2024. p. 5 IEEE, 2024
[14] W. Yang, Y. He, Q. Zhu, L. Zhang, and L. p. 5 Jin, “Unwrap-Net: A Deep Neural Network-based InSAR Phase Unwrapping Method Assisted by Airborne LiDAR Data,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. p. 5 218, 2024
[15] P. J. G. Teunissen, “Carrier Phase Integer Ambiguity Resolution,” in Springer Handbook of Global Navigation Satellite Systems, P. p. 5 J. G. Teunissen and O. p. 5 Montenbruck, Eds. p. 5, Springer, 2017, ch. p. 5 23
[16] P. Teunissen, “On InSAR Ambiguity Resolution for Deformation Monitoring,” Artificial Satellites, vol. p. 5 41, no. 1, 2006
[17] B. M. Kampes and R. p. 5 F. Hanssen, “Ambiguity Resolution for Permanent Scatterer Interferometry,” IEEE Transactions on Geoscience and Remote Sensing, vol. p. 5 42, no. 11, 2004
[18] S. S. Esfahany, “Exploitation of Distributed Scatterers in Synthetic Aperture Radar Interferometry,” PhD thesis, Delft University of Technology, Delft, The Netherlands, 2017
[19] A. Ferretti, A. p. 5 Monti-Guarnieri, C. p. 5 Prati, F. Rocca, and D. p. 5 Massonet, InSAR Principles: Guidelines for SAR Interferometry Processing and Interpretation, K. p. 5 Fletcher, Ed. p. 5, ESA Publications, 2007
[20] S. Verhagen, “The GNSS Integer Ambiguities: Estimation and Validation,” PhD thesis, Netherlands Geodetic Commission (NCG), Delft, The Netherlands, 2005
--- Page 1 ---
A Unified Maximum-Likelihood Framework for 3D InISAR Phase Unwrapping with Outlier Rejection
Matteo Pardi †∗, Francesco Mancuso ∗, Elisa Giusti ∗, Marco Martorella ‡∗
Pruning redundant reasoning tokens
The paper introduces a novel mathematical framework for phase unwrapping in three-dimensional interferometric ISAR (3D InISAR) imaging This approach works on a scatterer-by-scatterer basis and does not rely on any spatial continuity assumptions, making it suitable for sparse point clouds The formulation is derived from the Mixed-Integer Least Squares (MILS) theory, an optimal maximum-likelihood framework for joint estimation of integer and real unknowns in the presence of Gaussian noise This provides a unified way to handle generic sensor geometries, multi-baseline, multi-frequency, or hybrid setups
How the training signal is built
The observation model (4) is written by renaming variables as y = Aa + Bb + e where y ≡ ∆ϕ ∈ R m is the observation vector The optimal way to estimate a and b is through maximum-likelihood estimation (MLE): L(a, b y):= ∥y − Aa − Bb∥ squared Q-1 yy for the joint estimation of integer and real unknowns The operational algorithm consists of two steps: (i) estimate the integer vector aˆ using (9); (ii) estimate the real vector bˆ using (8) with the resulting aˆ
Where the gains come from
The quality of the integer solution can be evaluated using the posterior probability of aˆ given the data, defined as P(a y) ∝ R b∈B P(y a, b) db This leads to a natural a posteriori quality measure defined as AP:= P(aˆ y) The Ambiguity Posterior (AP) value lies in [0, 1] and expresses how likely the estimated integer solution is correct
Monte Carlo simulations show that when APthr increases, more scatterers are rejected, and the AccR gradually decreases until it reaches values close to zero The results indicate that for the considered configuration to work reliably, the SNR must be above roughly 25 dB The final reconstruction after the accept/reject step shows a significant improvement, with the RMSE computed only on correctly unwrapped scatterers being 0.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Mixed-Integer Least Squares (MILS) Phase Unwrapping Framework: This framework provides a
statistically optimal way to jointly estimate integer ambiguities and scatterer positions according to the maximum-likelihood criterion,
allowing for joint estimation of real and integer unknowns in generic sensor geometries. -
A Posteriori Quality Metric for Outlier Rejection: The Ambiguity Posterior (AP) is introduced as an
a posteriori metric for outlier rejection,
expressed as P(a y), which allows the system to define a statistical test to "Accept if AP >= APthr, where APthr is user-defined." -
Dynamic Threshold Setting for Outlier Rejection: The system can employ a
Fixed-CoFaR
strategy where a desired Conditional Failure Rate (e.g., 5%) is chosen, and the corresponding threshold APthr for each scatterer is computed based on its SNR, keeping the CoFaR constant across all scatterers. -
Reliable 3D Reconstruction in Ambiguous Conditions: The improved system can achieve
reliable 3D reconstruction even in challenging ambiguity conditions
by using the MILS algorithm combined with an accept/reject stage, significantly reducing geometric distortions compared to unwrapping without this step. -
System Design Optimization: By analyzing the performance metrics AccR and CoFaR as functions of SNR and APthr, designers can
optimize system parameters to reach the desired performance,
such as ensuring reliable operation only whenthe SNR is above roughly 25 dB
for a specific configuration.
Related papers
- Open-H-RF: A Large-Scale Ultrasound Raw Channel Dataset Initiative
- Scalable Attention for 5G NR Channel Estimation
- Low-Altitude Wireless Networks: A Comprehensive Survey
- Sensing-Assisted OFDM with Index Modulation: A Waveform for Optical Wireless Integrated Communication and Localisation
- Iteration-Free Candidate Pruning and an Index-Domain VLSI Architecture for Sort-Free MIMO Box Decoding up to 1024-QAM
- A Foundation Model for Large-Scale Wireless Network Planning, Operation and Optimization