Tom Dhaene

Visible light positioning (VLP) based on received signal strength (RSS) offers a low-cost solution for indoor localization, being easily implemented in a warehouse based on existing infrastructure. However, RSS-based VLP remains challenging in 3D and yields subpar performance compared to 2D due to the larger localization space, as well as the presence of dark spots where many LEDs are not bright enough. This limits the practical use cases of RSS-based VLP in industrial applications. We study the performance of RSS-based VLP on a 3D simulated environment by training various machine learning models, including Gaussian processes and Kolmogorov–Arnold networks on different representations of RSS data. Our findings show that the use of Gaussian processes for predicting distances to LEDs coupled with a logarithmic transformation and multilateration leads to both high-accuracy and high-precision predictions under thermal noise (p95 localization error of 10 cm under 50 dB SNR). With this technique, RSS-based VLP reaches levels of accuracy in our simulated 3D environment that are comparable to those reported for 2D applications, supporting the extension of RSS-based VLP to height-varying industrial use cases.

This study evaluates the application of a Bayesian variational framework for 3D activity reconstruction in 220-liter radioactive waste drums. The method combines Angular Segmented Gamma Scanning measurements and attenuation information derived from X-ray Computed Tomography. Computations are made tractable using stochastic variational inference together with a multi-resolution spatial prior for the inferred activities. A model to convert CT-derived gray levels into linear attenuation coefficients (CT-to-LAC), is developed to estimate energy-dependent attenuation maps from CT images, allowing the construction of the forward model under experimental conditions. Tests were performed using both homogeneous and heterogeneous mock-ups, with both point and linear sources made of either $^{241}$Am or $^{152}$Eu. Results show that the approach recovers spatial activity distributions broadly consistent with known source locations. Regarding the total activity of the drum, for multi-peak isotopes such as $^{152}$Eu, the true total activity is recovered within the posterior distribution and the posterior mean is more accurate than the estimates obtained from conventional averaging. In contrast, for the $^{241}$Am isotope, the reconstructions remain limited by the combined effect of single gamma energy line non-identifiability and the stronger sensitivity of the LAC calculation errors at lower energies. This work provides the first experimental validation of Bayesian 3D activity reconstruction from ASGS and CT measurements, for radiological waste characterization.

Traditional far-field characterization of wireless devices relies on dense, fixed-grid measurements over the entire measurement surface, leading to high time and cost overheads. This paper develops a data-efficient sequential sampling strategy for far-field radiation measurements using Bayesian optimization (BO) with Gaussian Process (GP) surrogate modeling on spherical domains. A central contribution is the adaptation and evaluation of GP kernels to capture the continuity and smoothness of spherical radiation patterns, enabling accurate surrogate models for BO. To demonstrate practicality, we compare kernel choices within the proposed framework and introduce a motion-aware acquisition function that reduces travel time between sampling points. The methodology is validated on both simple and more directive antenna scenarios, showing that the proposed approach reliably identifies key radiation features with significantly fewer samples than dense grid-based methods. These results highlight the potential of BO-driven sequential sampling for efficient and reliable radiation characterization on spherical domains.

This work presents a novel composite Gaussian process (GP) model for representing the complex-valued scattering parameters (S-parameters) of electrically long microwave interconnects as a function of frequency. The proposed model integrates a rational Szegö kernel, which captures the holomorphic structure and Hermitian symmetry of S-parameters, with a delay-informed periodic kernel designed to capture oscillatory behavior induced by wave propagation, thus resulting in improved predictive accuracy.