Search
Journal Paper
Revisiting the conditional single signal model with phase uncertainty: application to time-delay estimation
Elsevier Signal Processing, On line, November 2026.
Parameter estimation from phase-based observations arises in a wide range of applications, including navigation, radar, and communication systems. Within the conditional signal model, the received signal phase is typically represented as a single unknown term by merging hardware-induced and propagation-induced contributions into a complex amplitude, under the assumption that they are not independently identifiable. In this work, we revisit this modeling assumption and show, through an identifiability analysis, that hardware-induced and propagation-induced phase components are independently identifiable under suitable conditions. This result challenges the classical formulation and enables a more general signal model in which the phase structure is explicitly parameterized. Building on this formulation, we develop a general estimation framework for single-source parameter estimation, valid for an arbitrary parameter vector, and explicitly accounting for auxiliary phase observations affected by uncertainty. Within this framework, we derive Cramér–Rao bounds and maximum likelihood estimators (MLEs) for two scenarios: ideal hardware-phase compensation and a realistic case with noisy auxiliary phase measurements. We also implement a naive estimator that directly compensates the observed phase and applies MLE under ideal compensation. The proposed estimators are analyzed theoretically and validated through Monte Carlo simulations. The results show that phase uncertainty induces a performance limitation, leading to a saturation effect at high signal-to-noise ratios, whereas ideal phase compensation yields significant accuracy improvements over the conventional model. Finally, time-delay and Doppler estimation are considered to illustrate the proposed framework. The theoretical findings are validated using GPS L1 C/A and Kasami sequences, widely used in GNSS and communication systems.
Signal and image processing / Localization and navigation and Space communication systems
Talk
A practical introduction to conformal prediction for signal processing
Seminar of TeSA, Toulouse, October 20, 2026.
Uncertainty quantification asks a machine learning (ML) model not only for a prediction, but for a confidence band or set to go with it. This is difficult for data-driven models such as neural networks, which rest on no distributional assumptions about the data. Conformal prediction has recently attracted broad interest across the statistics, computer science and signal processing communities because it fills this gap: it wraps around any model, treating it as a black box, and returns prediction sets with finite-sample, distribution-free coverage guarantees. This talk introduces the variants most widely used in regression and classification: full and split conformal prediction, conformalized quantile regression, and online conformal prediction. The emphasis is practical, usage of these tools are illustrated with a hands-on Python notebook. The talk closes with example applications in signal processing problems.
Signal and image processing / Other
Journal Paper
Enhancing the Reliability of Swarming Ad-hoc Networks
Springer Nature, vol. 32, pp 2463–2477, July 2026.
We investigate the evolution of the reliability of a mobile communication network through its capacity to avoid and withstand faults (called robustness) and to maintain proper functioning if faults occur nevertheless (called resilience). Our case study focuses on a swarm of nanosatellites orbiting the Moon and operating as a distributed space interferometer. The objective of this study is to evaluate the impact of graph division techniques on the robustness and resilience of the system, simultaneously. A high reliability level is key to guaranteeing proper quality of service by recovering from impairments while preserving the primary function or mission of the mobile network. By analyzing the effects of exploration and random selection algorithms on the network reliability, our results show that fair graph division significantly improves robustness metrics such as routing cost and network efficiency, while introducing a measurable tradeoff with specific resilience metrics such as path redundancy and disparity. In addition, our analysis highlights the superior performance of sequential exploration algorithms, such as MIRW, in optimizing robustness while preserving a decent level of resilience.
Networking / Space communication systems
Bayesian polarimetric detector for improved forest loss monitoring using dual-polarization Sentinel-1 data
ISPRS Journal of Photogrammetry and Remote Sensing, vol. 237, pp. 94-112, July, 2026.
This study investigates the synergistic use of dual-polarization Sentinel-1 time series for forest loss monitoring across mixed land-cover types. While VH polarization generally provides higher contrast between intact vegetation and deforested areas, VV polarization is assumed to be more effective over areas characterized by vegetation remnants left after clearing. These insights motivate the development of fpol-BOCD, an unsupervised Bayesian polarimetric change detection method with Near Real-Time (NRT) capabilities. The algorithm is designed to leverage the complementary scattering properties of both polarizations, improving detection sensitivity across diverse land-cover conditions. Notably, fpol-BOCD jointly processes VH and VV Sentinel-1 data with comparable computational complexity relative to single-polarization approaches. Evaluation is performed using two datasets of forest loss events, extracted from the MapBiomas Alerta reference dataset for 2020 in the Cerrado and Amazon biomes in Brazil. One dataset focuses on small-scale clearings (0.1-2 hectares), while the other includes a randomly selected subset of larger deforested patches. The method improves detection accuracy by approximately 10% compared to single-polarization BOCD methods in areas likely containing residual vegetation after clearing, e.g., small disturbances in the Cerrado and large ones in the Amazon, while producing very few false alarms relative to the simple merging of single-polarization alerts. Additionally, fpol-BOCD outperforms the operational optical-based GLAD-L alerts in the Cerrado, achieving a 47% increase in true detections In the Amazon, a comparison with the operational SAR-based RADD alerts shows a 34% increase in true positives for small clearings. The detection speed of fpol-BOCD matches single-polarization BOCD methods. Temporal analysis shows a peak in forest loss during the dry season, with unexpected early wet-season increases likely linked to a lengthening of the dry period. Overall, results highlight the value of polarimetric data for efficient, more accurate forest loss monitoring in complex tropical environments.
Signal and image processing / Earth observation
Conference Paper
Experimental Validation of Passive Intermodulation Products Simulation by a Power Law Behavioral Model
In Proc. IEEE MTT-S International Conference on Numerical Electromagnetic and Multiphysics Modeling and Optimization (NEMO), Valencia, Spain, July 1-3, 2026.
As analytic nonlinearities cannot simulate correctly passive intermodulation (PIM) behavior, a nonanalytic power law behavioral model has been proposed to explain the non-classical variation of product power versus carrier power. This model has then been applied to many different measurement cases. This has successfully explained some published measures, and it has been used to propose new measurements and predict their results. It seems that the behavioral model may be linked to an underlying physical model of the nonlinearity responsible for the PIM.
Digital communications / Space communication systems
Talk
Localization in a Swarm of Satellites
COMET PDS Seminar on Radiodetermination Roadmap, CNES Toulouse, France, May 21, 2026.
Signal and image processing / Localization and navigation
Fast Covariance Learning algorithms for Sparse Bayesian Learning
In Proc. International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Barcelone, Spain, May 4-8, 2026.
In the multiple measurement vector (MMV) model, sparse signal recovery (SSR) is formulated as identifying the common support set of sparse signal vectors using a collection of measurement vectors obtained via a shared, known overcomplete dictionary. The sparse signal vectors or their powers may be estimated along with the support set using Sparse Bayesian Learning (SBL) framework. This talk focuses on the Type-II (marginal) likelihood in SBL, which reduces to a covariance‑learning (CL) problem: estimating the sparse signal powers and their support using the measurements’ second‑order statistics. Building on successive convex approximation (SCA) and majorization–minimization (MM), we propose two fast covariance-learning algorithms for solving the type-II likelihood, the CL-SCA and CL-MM algorithms. These methods iteratively estimate the sparse signal powers and typically require an order of magnitude fewer iterations than the standard SBL expectation maximization (EM) algorithm. Numerical experiments show consistent gains in SSR over state-of-the-art baselines. We also demonstrate an application to joint activity detection and channel estimation (JADCE) for massive random access. Finally, we discuss robustness extensions that move beyond the Gaussian assumptions of classical SBL.
Signal and image processing / Other
Public-Private Collaboration Research
INSPIRE Summer School on Remote Sensing & Artiicial Intelligence, Sfax, Tunisia, May 11-15, 2026.
Signal and image processing, Digital communications and Networking / Aeronautical communication systems, Earth observation, Localization and navigation, Space communication systems and Other
Conference Paper
A Novel Intrinsic Cramer-Rao Bound for Exact Gaussian Distribution on Lie Groups
In Proc. International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Barcelone, Spain, May 4-8, 2026.
In this communication, we propose a novel Cramer-Rao bound for matrix Lie groups, based on the exact modeling of Gaussian distributions on Lie groups (LG-ExCRB), different from those in the literature based on a simplifying approximation. To achieve this, we design a generic expression and develop new analytical formulas of the Fisher Information matrix. Then, closed-form expressions are given for two Lie groups of interest in engineering applications, SO(3) and SE(2). The proposed LG-ExCRB is validated numerically by comparison with the LG-CRB derived from the approximate modeling.
Fisher Scoring Algorithm for TIime-Delay and Doppler Estimation
In Proc. International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Barcelone, Spain, May 4-8, 2026.
Time-delay and Doppler estimation are among the most critical operations for synchronizing wireless communication systems, as well as for applications such as radar and global navigation satellite system (GNSS). Typically, a maximum likelihood estimator (MLE) is employed to initialize the time-delay and Doppler parameters. How ever, due to the high computational complexity of the MLE, sub-optimal algorithms are often used for subsequent parameter tracking. In this paper, we propose a novel low-complexity Fisher-scoring estimator, which is a variant of the Newton-Raphson method. Under the band-limited assumption, we derive closed-form expressions for the Fisher Information Matrix (FIM) and the gradient, both of which depend solely on the received signal samples. The performance of the proposed estimator is evaluated against the Cram´er-Rao Bound (CRB), demonstrating asymptotic convergence and achieving the same performance as the MLE after a certain number of iterations.
ADDRESS
7 boulevard de la Gare
31500 Toulouse
France