Smartphone-based ultra-wideband (UWB) localization is sensitive to anchor orientation, smartphone posture, and motion state, which cause condition-dependent ranging errors. We propose Inconsistency-Aware Robust Trilateration (IART), a lightweight framework that predicts per-anchor inconsistency and uses it to adapt measurement uncertainty within an Iteratively Reweighted Least Squares Gauss–Newton (IRLS-GN) solver. Combined with residual-based robust reweighting, IART suppresses unreliable range measurements without temporal filtering or IMU fusion. Experiments were conducted in a furnished indoor office using three anchors and two UWB-enabled iPhones. The evaluation covered 18 combinations of three anchor-orientation configurations (each fixed during an experimental run), three smartphone postures, and two motion states. Data were collected along two trajectories within the same environment and fixed anchor layout: Trajectory 1 for model development and same-trajectory evaluation, and Trajectory 2 for evaluating robustness to trajectory change. On Trajectory 2, IART reduced the mean RMSE from 0.350 m to 0.329 m (a 6.0% improvement) and achieved the lowest mean RMSE among all evaluated methods. It performed worse than Raw in only 2 of the 18 conditions, compared with 7 conditions for explicit signed-bias correction. Although signed-bias correction achieved higher average condition-wise gains on both trajectories, its gain decreased sharply on Trajectory 2 and it showed more frequent condition-wise degradation. These results support the same-deployment cross-trajectory robustness of learned inconsistency-aware weighting in three-anchor smartphone UWB localization. The current implementation assumes externally available smartphone-posture and motion-state labels.
Indoor pedestrian localization often combines pedestrian dead reckoning (PDR) with ultra-wideband (UWB) ranging, but in fixed-anchor indoor deployments the resulting ranges are vulnerable to blockage and multipath, which produce impulsive, biased, and heavy-tailed errors that are not well handled by a Gaussian extended Kalman filter (EKF) update. This paper addresses robust handling of corrupted UWB ranges in step-driven PDR–UWB fusion for a handheld parametric-approach formulation. The proposed method uses a range-update formulation based on the maximum correntropy extended Kalman filter (MCEKF), in which kernel-based correntropy weighting selectively attenuates large residuals while jointly reweighting the prior and measurement terms through a fixed-point procedure. A practical step-interval association rule is also introduced so that asynchronous UWB ranges can be incorporated consistently into the same step-driven fusion pipeline. Experiments were conducted under intermittent obstruction, persistent non-line-of-sight blockage, and corner-induced multipath conditions, as well as on a long mixed-condition trajectory. In these evaluations, the proposed method achieved the lowest mean trial-wise root-mean-square positioning error across the evaluated corrupted-ranging conditions. The empirical error distributions also remained favorable relative to the baseline filters across distinct degradation patterns.
This study evaluates the performance of a collaborative, multicriteria integrity control method in the presence of Global Navigation Satellite System jamming and spoofing. By simultaneously leveraging several sources of information, it enables the determination of whether the position of the receiver has been compromised by an attack. In particular, it combines position, velocity, time (PVT) residuals with the carrier-to-noise density ratio ($ ext{C/N-0) to ensure a robust assessment of integrity across all available data. Its main strength lies in collaboration between receivers through comparison and information sharing. This approach is purely software-based, requiring no additional hardware, and is scalable to receiver populations of varying sizes, from small to large. This study uses real jamming data and simulated spoofing data. In both cases, the data used for PVT calculation, originating from different satellites, are real. The presented results based on data combining simulation and real data demonstrate that the method is promising.
This article explores the impact of electromagnetic interference from consumer electronics and common household appliance on the performance of Global Navigation Satellite System (GNSS) receivers in smartphones. As GNSS signals are weak and susceptible to external noise and interference, consumer appliances can introduce spurious emissions within GNSS frequency bands, leading to signal degradation. To investigate this, we conducted a series of controlled experiments simulating typical smartphone usage in proximity to different sources of EMI among household appliances (i.e., a microwave oven) and consumer electronics (i.e., Wi-Fi and Bluetooth devices), ultimately focusing on a dense Bluetooth environment. Our analysis reveals that the microwave oven can induce severe GNSS signal degradation, with substantial automatic gain control and C/N-0 drops (up to 2.3 dB-Hz) and significant satellite availability loss (up to 18 %), particularly in close-range configurations, leading to positioning root-mean-square error (RMSE) increases exceeding 7 m. In addition, Wi-Fi interference resulted in moderate reductions in signal quality, with positioning RMSE increases of less than 2m in most scenarios. In contrast, Bluetooth exhibited negligible effects under tested conditions. These findings suggest potential implications for indoor navigation applications and underscore the importance of further improvements in interference mitigation for consumer GNSS devices.
Indoor localization is a key technology for the safe operation of automatic guided vehicles (AGVs) in smart factories. The time-of-arrival (TOA) observations provide physically interpretable distance measurements, and TOA-based localization can be realized using existing wireless infrastructure but remains challenging due to the effects of multipath propagation and frequent obstruction of the line-of-sight (LOS). In a typical factory environment, the AGV and access point locations, LOS/non-LOS conditions, and obstacle layout are tightly coupled because they jointly determine the underlying radio propagation characteristics, which fundamentally affect the TOA-based localization performance. We propose an environment-guided localization framework that integrates the multiple dependencies of the propagation environment into a factor graph to reduce the uncertainty of TOA-based localization under multipath conditions. Factor graphs can be used to represent heterogeneous dependencies where each physical relationship is expressed as an individual factor. The factor graph for distributed localization comprises a TOA factor and positional constraint factor derived from a known obstacle map. We extend this factor graph to cooperative localization by adding an order-consistency (OC) factor that considers the environmental similarity among AGVs. Simulation results showed that the OC factor sharpens the marginal distributions of positions and provided insights into conditions under which the OC factor improves or degrades the accuracy of cooperative localization.
Global Navigation Satellite Systems (GNSSs) face increasing vulnerability to radio frequency interference, with chirp signals representing a particularly challenging threat due to their time-varying frequency characteristics that disrupt receivers across wide bandwidths. This study introduces a novel interference mitigation technique that partitions the interferer into segments, leveraging the fractional Fourier transform (FrFT) to enhance suppression. The proposed algorithm employs a piecewise least squares estimation method in combination with fractional-domain processing, specifically aimed at manage nonlinear wideband chirp interference impacting snapshot GNSS receivers. A generalized cross validation metric is used to balance algorithmic complexity with fitting accuracy in each segment. Mitigation occurs in the fractional domain, allowing interference energy to be concentrated at fractional coordinates matching the chirp characteristics, thereby reducing the number of affected GNSS samples compare to traditional pulse blanking. The method demonstrates particular robustness against linear frequency modulated (LFM) chirps, and for quadratic frequency modulated signals, it further divides the interference into approximately LFM segments for targeted FrFT-based suppression. Experimental results show that this approach preserves signal quality more effectively than conventional mitigation strategies.
Many industrial and scientific applications rely on high accuracy positioning. With the advancements of modern Global Navigation Satellite Systems constellations and the release of new services, access to a real-time high-accuracy positioning anywhere in the world is a reality. To enable real-time precise point positioning, Galileo High Accuracy Service (HAS) offers corrections without needing any additional infrastructures on the user side. In this article, we compare the user positioning performance obtained with HAS corrections received via Internet Data Dissemination and Signal-In-Space using the software JRC User Navigation Engine, which implement the Galileo HAS reference user algorithm developed at the Joint Research Centre. The comparison is carried out using stations of the International GNSS Service located throughout the world, using data acquired since the start of the initial service phase (January 2023). The goal of the article is to evaluate the achievable performance using the different HAS corrections, by reviewing the performance evolution of HAS since the beginning of the initial service phase, and determining the impact on the user performances.
Several precise positioning systems rely on carrier phase measurements, which enable centimeter-level accuracy. However, these measurements are significantly affected by additive noise, degrading the performance of phase-based positioning estimators. To better capture the statistical behavior of carrier phase observations, we propose to model phase measurements with a von Mises distribution. Then, to perform precise navigation, it is fundamental to estimate its parameters. To ensure estimation consistency and the respect of the underlying geometric constraints, we propose to perform their estimation within a Lie group (LG) framework. Furthermore, we adopt a Bayesian approach, where prior information about the parameters is assumed within the space SO(2) & times; R+. The proposed methodology provides a full Bayesian formulation that incorporates prior knowledge, solved through a Newton algorithm on LGs. This approach demonstrates advantages in terms of robustness and precision, especially when dealing with a small number of observations, compared to traditional Euclidean-based and frequentist methods. Furthermore, we validate the approach through simulations with a real navigation dataset.
Radio-based localization systems determine target positions by analyzing radio communication between sensor nodes. Despite numerous technological approaches, no generic information-theoretic modeling exists to compare such systems, even though a unified view of position information flow forms the conceptual link between all RF-based localization systems. This often yields suboptimal designs, where processing mismatches and hardware limits cause information loss and reduced localization performance. We propose Localization as Information Flow (LocInFlow), the first model that maps RF-based localization systems onto the functional blocks of the Shannon-Weaver communication model, enabling technology-independent analysis of position information flow. LocInFlow introduces three metrics: the a-posteriori uncertainty matrix for geometric coding, the Modulation Mutual Information for modulation, and the system-level flow efficiency. The metrics do not target perfect reconstruction; they expose where information is lost and enable blockwise comparison under explicit measurement assumptions, using standard deployment artifacts and routine calibration/evaluation data, without claiming universal predeployment error prediction. We illustrate LocInFlow by mapping a simplified Bluetooth low energy (BLE) angle-of-arrival system onto it, exposing dominant information-loss sources across functional blocks. Although introduced on a moderate BLE grid, LocInFlow scales to large infrastructures and 3-D scenarios, as its functional-block metrics are technology-independent and do not depend on scenario size or dimensionality.
This article presents an automated wireless indoor localization system that eliminates the need for site surveying and reduces the dependency on complete anchors' location knowledge. The proposed approach leverages cooperative received signal strength (RSS) measurements and a hybrid logarithmic path-loss model to automatically localize network anchors and build radio maps that can be used to localize user nodes. Starting from a limited initial number of known anchor nodes, position estimates are propagated to the rest of the network using an iterative strategy guided by a learning-based dilution of precision (DOP) metric. To address the challenge of unknown anchor locations at initialization, a novel Gaussian process regression (GPR)-based method is introduced to estimate the DOP values without requiring ground-truth coordinates. This enables a self-organizing deployment process that adapts to varying indoor deployments. The system is validated through Cram & eacute;r-Rao lower bound analysis, simulations, and real-world experiments in ZigBee-based networks. Results demonstrate a root-mean-square localization error of 3 m in a typical indoor 15 m & times; 30 m office environment covered by seven partially known anchors without prior site survey using only RSS measurements, providing a balance between cost, deployment complexity, and localization accuracy.
Distributed learning and Edge Artificial Intelligence (AI) necessitate efficient data processing, low-latency communication, decentralized model training, and stringent data privacy to facilitate real-time intelligence on edge devices while reducing dependency on centralized infrastructure and ensuring high model performance. In the context of Global Navigation Satellite System (GNSS) applications, the primary objective is to accurately monitor and classify interferences that degrade system performance in distributed environments, thereby enhancing situational awareness. To achieve this, machine learning (ML) models can be deployed on low-resource devices, ensuring minimal communication latency and preserving data privacy. The key challenge is to compress ML models while maintaining high classification accuracy. In this article, we propose variational autoencoders (VAEs) for disentanglement to extract essential latent features that enable accurate classification of interferences. We demonstrate that the disentanglement approach can be leveraged for both data compression and data augmentation by interpolating the lower-dimensional latent representations of signal power. To validate our approach, we evaluate three VAE variants—vanilla, factorized, and conditional generative—and benchmark 19 state-of-the-art VAE and generative models on five distinct datasets, including three collected in controlled indoor environments and two real-world highway datasets. In addition, we conduct extensive hyperparameter searches to optimize performance. Our proposed VAE achieves a data compression rate ranging from 512 to 8192 and achieves an accuracy up to 99.92%. Quantizing our model from float32 to int8 results in a fourfold reduction in model weight size.
Inertial sensors are widely used for pedestrian activity recognition. Recent advances in deep learning techniques have significantly improved the inertial classification task’s performance and robustness. However, a standardized benchmark for evaluating and comparing these methods remains lacking. Such a benchmark is critical for ensuring fair and consistent evaluation and future development. In this study, we aim to fill this gap by defining and analyzing 11 data-driven techniques designed to enhance neural inertial classification networks. Our investigation focuses on three key components: network architecture, data augmentation, and data preprocessing. In addition, we conduct comparative analyses to identify the optimal window size for each dataset. This is a parameter that substantially affects model performance but is often overlooked. The experiments were conducted across seven datasets collected from 229 participants and with a total of 4482 min. Among the evaluated techniques, data augmentation through rotation and multihead network architectures yielded the most consistent performance improvements. Our experimental results show that rotation-based augmentation and multihead architectures consistently yield the highest gains, improving accuracy by up to 9.72% depending on the dataset and window length. We additionally quantify the effect of temporal window size, demonstrating that longer segments (2 s) provide the largest average improvement, whereas shorter windows better suit real-time deployment. Finally, we propose a benchmarking strategy to support the future development and evaluation of deep learning models for inertial activity recognition.
Indoor positioning based on 5G data has achieved high accuracy through the adoption of recent machine learning (ML) techniques. However, the performance of learning-based methods degrades significantly when environmental conditions change, thereby hindering their applicability to new scenarios. Acquiring new training data for each environmental change and fine-tuning ML models is both time-consuming and resource-intensive. This article introduces a domain incremental learning (DIL) approach for dynamic 5G indoor localization, called 5G-DIL, enabling rapid adaptation to environmental changes. We present a novel similarity-aware sampling technique based on the Chebyshev distance, designed to efficiently select specific exemplars from the previous environment while training only on the modified regions of the new environment. This avoids the need to train on the entire region, significantly reducing the time and resources required for adaptation without compromising localization accuracy. This approach requires as few as 50 exemplars from adaptation domains, significantly reducing training time while maintaining high positioning accuracy in previous environments. Comparative evaluations against state-of-the-art DIL techniques on a challenging real-world indoor dataset demonstrate the effectiveness of the proposed sample selection method. Our approach is adaptable to real-world nonline-of-sight propagation scenarios and achieves an mean absolute error positioning error of 0.261 m, even under dynamic environmental conditions.
This article proposes a novel 6G downlink waveform designed for passive coherent location using orthogonal time-frequency space (OTFS) modulation. Targeting private campus networks, the approach enables passive sensing without active signal emission, addressing key challenges of energy consumption, interference, and cost in industrial automation. The OTFS framework operates in the delay-Doppler domain, allowing seamless integration of radar functionality into communication signals while maintaining synchronization-free operation through local signal processing.A comprehensive simulation-based analysis of OTFS grid configurations reveals fundamental tradeoffs between sensing resolution and computational efficiency. Controlled ray-traced simulations support the theoretical framework, indicating high-resolution target detection capabilities that meet third Generation Partnership Project requirements for autonomous mobile robot navigation. The proposed architecture offers power advantages through elimination of transmit amplification, the primary power consumer in active radar systems, and provides inherent privacy advantages through passive operation and distributed processing.Processing chain analysis reveals strong compatibility with multistatic extensions, requiring only evolutionary modifications rather than fundamental redesign.Simulation results suggest the feasibility of dual-use signaling in future 6G networks, with applications extending beyond industrial automation to smart cities, traffic monitoring, and public safety systems.
Multimodal transport refers to multiple transportation means (e.g., car and plane) that can be used to transport people or goods. Classifying the mode of transportation can have multiple usages toward sustainable transport solutions, such as optimizing routes, reducing transit times, having efficient logistics operations, reducing transportation costs by strategically combining different modes, or understanding how people move within cities for migration studies. Multimodal transport classification has traditionally relied on data collected from various movement sensors (e.g., accelerometers, pedometers, and gyroscopes); yet, with the opening of the access to raw global navigation satellite system (GNSS) data on mobile devices, new avenues of multimodal analysis have been created, when GNSS signals alone (without additional sensors) could be used to classify the mode of transport. This article introduces a novel Receiver Independent Exchange (RINEX)-based framework for multimodal transport classification that operates exclusively on instantaneous raw GNSS observables, without relying on position estimates or auxiliary motion sensors. Unlike traditional approaches that require at least four satellites for positioning, the proposed method achieves classification using data from as little as one strongest satellite in view. By leveraging machine learning algorithms, transportation modes are inferred directly from single and double differences of pseudorange, Doppler, and carrier-to-noise ratio features extracted from raw RINEX data. The framework was validated using an extensive dataset collected from 18 volunteers across five European countries, using 409 tracks and ten transportation modes. The results show that accurate and stable classification is possible even with limited satellite visibility, demonstrating the feasibility of low-power, privacy-preserving, and geometry-aware mobility analytics based solely on raw GNSS measurements.
Radio-based localization systems conventionally require stationary reference points (e.g., anchors) with precisely surveyed positions, making deployment time-consuming and costly. This article presents an empirical evaluation of collaborative self-calibration for ultra-wideband (UWB) networks, extending a Bayesian approach based on grid-based uncertainty propagation. The enhanced algorithm reduces measurement availability requirements while maintaining positioning accuracy through probabilistic state estimation. We validate the approach using real-world data from controlled indoor experiments with 12 nodes in a static environment. Experimental evaluation yields 0.28 m mean ranging error under line-of-sight conditions and 1.11 m overall ranging error across mixed propagation scenarios. Results confirm the algorithm's resilience to measurement noise and partial connectivity scenarios typical in industrial deployments. We evaluate algorithm robustness under nonline-of-sight-contaminated initialization, showing graceful accuracy degradation (median error 0.62-0.99 m) compared to closed-form methods that exhibit substantial performance collapse (median error up to 2.43 m). The findings contribute to automated UWB network initialization for indoor positioning applications, reducing infrastructure dependence compared to manual anchor calibration procedures.
While Wi-Fi positioning is still more common indoors, using magnetic field features has become widely known and utilized as an alternative or supporting source of information. Magnetometer bias presents significant challenge in magnetic field navigation and SLAM. Traditionally, magnetometers have been calibrated using standard sphere or ellipsoid fitting methods and by requiring manual user procedures, such as rotating a smartphone in a figure-eight shape. This is not always feasible, particularly when the magnetometer is attached to heavy or fast-moving platforms, or when user behavior cannot be reliably controlled. Recent research has proposed using map data for calibration during positioning. This paper takes a step further and verifies that a pre-collected map is not needed; instead, calibration can be done as part of a SLAM process. The presented solution uses a factorized particle filter that factors out calibration in addition to the magnetic field map. The method is validated using smartphone data from a shopping mall and mobile robotics data from an office environment. Results support the claim that magnetometer calibration can be achieved during SLAM with comparable accuracy to manual calibration. Furthermore, the method seems to slightly improve manual calibration when used on top of it, suggesting potential for integrating various calibration approaches.
