Jae-Young Pyun

Received signal strength indicator (RSSI)-based positioning with Bluetooth Low Energy (BLE) is widely used for indoor localization, and weighted centroid localization (WCL) is among the most common methods because of its simplicity and low computational cost. However, BLE RSSI is subject to temporal fluctuation and multipath interference, which induce an unstable RSSI-to-distance relationship and degrade the accuracy of conventional WCL. Its distance-based weighting cannot distinguish reliable line-of-sight (LOS) anchors from misleadingly strong non-line-of-sight (NLOS) anchors. This limitation is most severe in long, anchor-dense corridors and high-traffic lobbies, where multipath propagation is high. This paper presents a confidence-adaptive Kalman filtering framework for BLE RSSI-based localization in multipath-affected environments. A scan-level confidence score combines relative RSSI ranking, probabilistic LOS/PLOS/NLOS classification, and the proportion of high-confidence LOS predictions. The resulting score adapts the measurement covariance assigned to WCL position observations without modifying individual anchor weights. Relative to conventional WCL, the proposed method reduced the mean absolute error by 14.6% in corridors, 9.3% in a classroom, 3.3% in an office, and 7.9% along a high-traffic lobby trajectory. Relative to a fixed-covariance Kalman filter, the additional reduction was small in the low-traffic zones (0.4–2.0%) and largest in the high-traffic lobby (7.7%).

Long range wide area network (LoRaWAN) has become a leading low-power wide area network (LPWAN) technology for Industrial Internet of Things (IIoT) applications, offering long-range communication with low energy consumption. A fundamental challenge lies in selecting the appropriate spreading factor (SF) for each device since this directly influences coverage, packet success ratio (PSR), and airtime. The default adaptive data rate (ADR) mechanism is static and fails to adapt under dynamic network conditions such as mobility. This article proposes a lightweight linear upper confidence bandit (LinUCB–SF)-based reinforcement learning (RL) approach for adaptive SF selection. Each end device (ED) uses locally observable features to autonomously select its SF, balancing exploration and exploitation. The method is implemented and validated in ns-3 simulations across a range of deployment densities. Results show that our proposed LinUCB–SF algorithm reduces energy consumption by 19.3% in mobile scenarios and 37.8% in static scenarios, while improving PSR by 9.1% and 9.0%, respectively, compared to the EXP3 baseline.

Because a typical adaptive data rate (ADR) of the long-range wide-area network (LoRaWAN) is designed for static services such as smart grid, monitoring, and metering, its performance is limited in mobile environments for location-based services (LBS). Indeed, ADR cannot respond to frequent channel environment changes, thus allocates incorrect data rate (DR) settings to devices, resulting in generation of massive packet loss. To improve mobile support in LoRaWAN location-based services, this paper proposes a robust ADR (RADR) based on an artificial intelligence (AI) approach. The proposed RADR enhances the DR allocation by applying an AI model trained on various channel environments of LBS. Its AI model infers efficient DR, ensuring reliable packet delivery even in mobile scenarios. To verify and demonstrate our AI-driven RADR performance, we built a testbed composed of RADR-embedded LoRaWAN network server/end devices and evaluated their performance in field tests assuming LBS. The results showed a significant improvement in the packet success rate, with the proposed RADR scheme achieving an average increase of 30% compared to the typical ADR.

This study introduces a device mobility identification (DMI) method required for Long-Range Wide Area Networks location-based services, including both mobile and stationary applications. The proposed DMI, trained under various channel conditions, can determine the mobility status of the device in real-time without motion sensors or positioning systems. The training data given to the DMI consisted of packet loss, signal-to-noise ratio, and received signal strength indicator histories recorded at the network server. The performance evaluation provided insights into the generalization capability of the proposed deep learning model for the DMI across both known and unknown environments. Furthermore, the proposed method improves mobility identification performance, achieving an accuracy of 87.30%, as compared with 74.17% of the existing support vector machine approach in an online test assuming a real-time pedestrian monitoring service.

This letter introduces a denoising-based two-way ranging (D-TWR) utilizing artificial intelligence (AI) for an ultra-wideband (UWB) indoor positioning system. The proposed D-TWR reduces the distance measurement error with a denoising neural network pre-trained on UWB channels in various indoor environments. The denoising model adopts an extended input layer comprising the TWR-measured distance and channel impulse response observations and infers the ground truth distance for UWB ranging. The experimental results provide insights into the performance of the proposed denoising AI model in various indoor environments. Furthermore, it was confirmed that our method can reduce the prediction error of the ranging distance by approximately 34.44% compared to the existing AI regression method, resulting in improved UWB positioning accuracy.

Location estimation in indoor environments using radiofrequency (RF) has garnered considerable attention in recent years owing to the widespread adoption of mobile devices. RF-based fingerprinting—a direct approach that allows location estimation based on observed signals—relies on manual surveys during the offline phase to create a radio map with coordinates and RF measurements at multiple locations. The accuracy of RF fingerprint-based localization is related to the number of reference points. However, conventional site survey procedures tend to incur substantial expenses. To alleviate the workload of site surveys and address the challenge of incomplete fingerprint databases, we propose a data-augmentation method to complement existing fingerprint data. Our approach leverages a conditional generative adversarial network with long short-term memory (CGAN-LSTM) prediction model to effectively learn the intricate patterns inherent in the initial training data and generate high-quality synthetic data that align with the underlying data distribution. In an experimental evaluation conducted on a real testbed, our data augmentation framework increased the average localization accuracy by 15.74% compared with fingerprinting without data augmentation. Furthermore, experiments conducted in two typical indoor environments using sparse data highlighted the significant performance enhancement of the proposed approach in reducing localization error and was comparable to state-of-the-art data-augmentation methods.

The popularity of radio frequency (RF)-based fingerprinting for indoor localization has grown owing to its relatively low cost of equipment deployment and satisfactory accuracy. However, generating a complete radio map by associating unlabeled RF signals with the corresponding location information remains challenging, especially in crowdsourcing-based fingerprinting. In this article, we propose a semi-crowdsourced radio map construction method based on Bluetooth low-energy (BLE) landmarks that harnesses reference points (RPs) in the radio map for coarse localization and facilitates the labeling of location information to WiFi signals. Principally, we acquire RF-received signal strength (RSS) measurements annotating them with location coordinates recorded while a user is walking to provide an efficient method of data collection. Furthermore, we introduce a generative adversarial network (GAN)-based method to increase the amount of training data collected at each RP and reduce human effort by augmenting the fingerprint database. Our proposed method demonstrates promising results, including improved localization accuracy and localization performance comparable to that of traditional site surveys while reducing measurement time and human effort.

The Internet of Things is rapidly growing with the demand for low-power, long-range wireless communication technologies. Long Range Wide Area Network (LoRaWAN) is one such technology that has gained significant attention in recent years due to its ability to provide long-range communication with low power consumption. One of the main issues in LoRaWAN is the efficient utilization of radio resources (e.g., spreading factor and transmission power) by the end devices. To solve the resource allocation issue, machine learning (ML) methods have been used to improve the LoRaWAN network performance. The primary aim of this survey paper is to study and examine the issue of resource management in LoRaWAN that has been resolved through state-of-the-art ML methods. Further, this survey presents the publicly available LoRaWAN frameworks that could be utilized for dataset collection, discusses the required features for efficient resource management with suggested ML methods, and highlights the existing publicly available datasets. The survey also explores and evaluates the Network Simulator-3-based ML frameworks that can be leveraged for efficient resource management. Finally, future recommendations regarding the applicability of the ML applications for resource management in LoRaWAN are illustrated, providing a comprehensive guide for researchers and practitioners interested in applying ML to improve the performance of the LoRaWAN network.

Terahertz (THz) is a promising technology for future wireless communication networks, particularly for 6G and beyond. The ultra-wide THz band, ranging from 0.1 to 10 THz, can potentially address the limited capacity and scarcity of spectrum in current wireless systems such as 4G-LTE and 5G. Furthermore, it is expected to support advanced wireless applications requiring high data transmission and quality services, i.e., terabit-per-second backhaul systems, ultra-high-definition streaming, virtual/augmented reality, and high-bandwidth wireless communications. In recent years, artificial intelligence (AI) has been used mainly for resource management, spectrum allocation, modulation and bandwidth classification, interference mitigation, beamforming, and medium access control layer protocols to improve THz performance. This survey paper examines the use of AI in state-of-the-art THz communications, discussing the challenges, potentials, and shortcomings. Additionally, this survey discusses the available platforms, including commercial, testbeds, and publicly available simulators for THz communications. Finally, this survey provides future strategies for improving the existing THz simulators and using AI methods, including deep learning, federated learning, and reinforcement learning, to improve THz communications.