Guoliang Chen

Smartphone-based Wi-Fi round-trip time (RTT) indoor localization is significantly degraded by line-of-sight (LoS) and non-line-of-sight (NLoS) conditions. To address this, we propose an enhanced RTT localization framework that leverages wavelet transform for joint LoS/NLoS identification and ranging denoising. Specifically, continuous wavelet transform (CWT) first extracts normalized coefficients from raw RTT data, which are then fed into a Gaussian process classifier (GPC) to identify LoS/NLoS states. Based on these classification results, NLoS-induced ranging errors are subsequently mitigated via discrete wavelet transform-based denoising with an adaptive threshold strategy. Experiments demonstrate that in the office scenario, the proposed CWT-GPC identifier reduces the fail rate by 10.10, 21.50, and 15.50 percentage points compared with SVM, RF, and GPC, respectively; in the corridor scenario, the corresponding reductions are 8.26, 27.84, and 26.95 percentage points. In the office scenario, the denoising step reduces the mean localization error by 0.584 m, 0.406 m, and 0.232 m compared with LC, NLC, and GPRC, respectively. In the corridor scenario, the corresponding reductions are 1.252 m, 0.743 m, and 0.561 m, respectively. These results validate significant gains in both identification robustness and positioning precision.

With the evolution of Fifth-Generation (5G) wireless systems, high-resolution channel observations enabled by large bandwidths and massive Multiple-Input Multiple-Output (MIMO) have opened new opportunities for indoor positioning. However, while existing fingerprinting-based learning methods excel in complex multipath environments, they often rely on fixed anchor configurations. This makes them vulnerable to dynamic anchor visibility caused by signal blockages, user mobility, or hardware failures in practical deployments. To address this, we propose 5G Graph-based Localization (5GLoc), a learning-driven framework designed for robust indoor positioning under dynamic anchor availability. By modeling localization instances as graphs, 5GLoc utilizes a Dynamic Graph Attention Network (DGAT) to flexibly accommodate varying numbers and combinations of anchors, ensuring robust inference under heterogeneous observations. We further introduce a Combinatorial Dynamic Node Masking (CDNM) strategy for data augmentation, as well as a composite loss function with learnable weights for balancing accuracy and stability. Finally, a map-constrained Particle Filter (PF) is integrated to refine trajectories for spatial feasibility and trajectory consistency. Experimental results based on the IPIN 2025 dataset demonstrate that 5GLoc achieves superior performance in complex warehouse environments, with a 75th percentile error of 0.857 m. Compared to several representative positioning methods, 5GLoc reduces the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) by 42.2%–80.2% and 50.6%–80.8%, respectively, maintaining high reliability even under extreme conditions with reduced anchor availability. This framework provides an effective solution for 5G localization in dynamic environments, meeting both precision and real-time requirements.

In this paper, we have constructed a high-precision indoor positioning system by leveraging WiFi round-trip time (RTT), magnetic field and pedestrian dead reckoning (PDR). For the WiFi RTT module, a collaborative WiFi RTT positioning approach is proposed by combining RTT ranging and fingerprinting through a Cramer-Rao lower bound (CRLB)-based outlier detection method. Based on this, an enhanced version of the geomagnetic-aided WiFi RTT localization method is implemented. For better smartphone-based deployment, we improve PDR in terms of step detection, step length estimation and heading optimization. Moreover, an optimization strategy incorporating map-aided resampling and particle swarm reconstruction is designed to enhance the performance of particle filter. Finally, the WiFi RTT, magnetic positioning and PDR are integrated based on the improved particle filter (IPF). Our system employs two mechanisms to achieve fusion localization, which means that the system executes WiFi/magnetic/PDR fusion when the WiFi module is available and switches to magnetic/PDR fusion when the WiFi signal is absent. Extensive experiments conducted in a corridor environment demonstrate that the IPF-based WiFi RTT/Magnetic/PDR model achieves localization accuracy of 0.64 m and 0.91 m when the WiFi module is available and unavailable, respectively, showing meter-level continuous positioning capability even in the presence of additional WiFi ranging errors and in areas without WiFi signal coverage.

The WiFi fine time measurement (FTM)-based ranging positioning often encounters challenges instability and low accuracy in real-life scenarios due to inaccurate round-trip time (RTT)-based ranging. To address these issues, we construct a WiFi RTT/RSS/PDR/Map fusion system with an improved mapaided particle filter (PF). Our approach focuses on FTM-based localization from a fingerprinting perspective rather than WiFi ranging positioning. The WiFi FTM fingerprinting utilizes an artificial database, reducing the need for offline labor-intensive measurement work. The semi-parametric error model is employed to compensate for the WiFi RTT and RSS data and improve fingerprinting accuracy. A collaborative optimization strategy incorporating map and fingerprinting information is designed to enhance the PF’s performance in terms of particle impoverishment, propagation, and weight update. Extensive experiments demonstrate that the integrated WiFi RTT/RSS/PDR/Map system is error-tolerant under various additional error variance conditions (e.g., heading δ _Ψ ² ≤4/ deg ² , or step length δ _L ² ≤0.5/ m ² , or RTT ranging δ _d ² ≤2/ m ² and RSS δ _R ² ≤2/ dBm ² ), with an accuracy fluctuation within 0.1 m. The system obtains meter-level accuracy within 0.19 s when utilizing 1000 particles for every localization request. Compared to the state-of-the-art fusion approaches, the WiFi RTT/RSS/PDR/Map fusion system can deliver better accuracy and stability.

A perfluoroglutaric acid stapled ligand for synthesis of highly emissive CsPbBr₃ perovskite nanocrystals in polar organic solvent.

Currently, various indoor positioning technologies are widely studied, and visible light positioning (VLP) is a promising technology due to its high accuracy, low cost, and high output rate. However, the most common method based on the received signal strength (RSS) requires calibrating the model in advance, which has a weak generalization ability. This article focuses on the VLP method based on the time difference of arrival (TDOA), which does not require heavy preparatory work. Firstly, we analyze the influence of different errors on TDOA-based VLP, such as the time synchronization error, receiver noise, etc. Secondly, a convolution neural network (CNN) based network is designed for phase difference estimation, which significantly improves the accuracy of phase difference estimation compared to the traditional in-phase&quadrature signal-based method. Lastly, a particle filter based on the motion state is proposed to improve positioning accuracy and robustness. Simulated experiments evaluate the proposed methods, and the final results show a significant improvement in accuracy when compared with traditional methods. The improvements in ranging and localization accuracy can both reach over 50%.

Recently, indoor localization has played a crucial role in the construction of smart homes, smart buildings, and smart cities. Numerous localization technologies have been investigated, and visible light positioning (VLP) as one of the representative technologies has drawn much attention due to its high accuracy, low cost, and friendly infrastructure. Aiming to solve the prominent issues faced by the traditional VLP ranging-model-based localization method, a phase-difference-based ranging method assisted by a dual-LED anchor was first proposed, which can eliminate the negative influences of hardware heterogeneity, environmental differences, and the receiver’s inclination angle effectively. The key point of the proposed solution is that we innovatively designed an anchor integrated by two-closely-spaced LEDs replacing the two separated LEDs for distance calculation, which contributed to avoiding the workload of the ranging model’s calibration and the dependence on frequency exchange. Furthermore, a weight matrix was introduced in the final localization to restrain the significant error and increase the confidence coefficient of reliable observations. Finally, a brief analysis of the proposed method’s theoretical performance was presented and evaluated by extensive experiments. The proposed method was illustrated to be a very competitive system, which obtained a centimeter-level localization accuracy, and showed more advantages than the traditional ranging method, especially when the PD’s inclination angle is not zero.

Robust and accurate step counting plays an essential role in some fields, such as indoor positing, behavior recognition, and health management. Currently, there are numerous applications or methods in step counting. However, most solutions did not emphasize the elimination of false steps in the non-walking state, which still encounters the over-counting or under-counting problem. In this study, a robust and accurate step counting solution based on movement mode recognition is proposed. First, according to the characteristics of pedestrians walking, the SVM and FSM-DT classifiers were constructed to recognize the user’s motion state and phone usage mode, respectively. The purpose of classification is to solve the step-counting error of non-walking state and initialize a suitable parameter for different cases. Then, we utilized a crest-valley detection algorithm with multi-feature constrained to detect the step, and the initial threshold values for each mode can be adjusted adaptively during walking. The results indicate that the accuracy of the proposed method can reach 99.11% and 97.43% in normal and free walking. Compared with standard peak detection and an excellent method, the proposed method can improve by 44.14% and 18.71% for false walking, respectively, which is a significant improvement in the robustness and accuracy. Furthermore, we also achieve an average accuracy higher than 98% in some typically carrying modes and ways, and higher accuracy compared with four popular step-counting mobile applications in different walking states.