Spam email detection remains an ongoing challenge due to the increasing sophistication and evolving tactics employed by spammers. Traditional rule–based and machine learning (ML) detection methods have demonstrated limitations in adaptability and generalization. This paper proposes SpamLLM, a multimodal spam detection framework that leverages a frozen large language model (LLM) backbone integrated with semantic embeddings and structured metadata extracted from email headers and body statistics. Rigorous evaluations are conducted across four widely used spam detection datasets, and SpamLLM is compared against classical ML, deep learning, and pretrained Transformer‐based baseline models. The experimental results demonstrate that SpamLLM outperforms existing methods and achieves state‐of‐the‐art performance in accuracy, precision, recall, and F1 score. Notably, SpamLLM excels in handling diverse spam content and provides robust detection across various datasets. These findings underscore the potential of multimodal fusion approaches for advancing spam classification systems and suggest promising directions for future research in email security.

To meet the high-precision positioning demands of Internet of Things (IoT) applications, carrier phase positioning in cellular networks is promising. However, dynamic clock offsets among base stations (BSs) and user equipments (UEs) challenge the accuracy of range-based positioning systems. To achieve centimeter-level positioning in time-varying asynchronous cellular networks, we propose a collaborative framework leveraging the coupled ranging information between UEs and BSs obtained from double-differenced (DD) Time of Arrival (ToA) and Carrier Phase of Arrival (CPoA) measurements. Coarse UE position estimates are first obtained iteratively from DD ToA, then refined via DD CPoA-based integer ambiguity resolution for joint high-accuracy estimation across time. By leveraging coupled measurements and time-invariant integer ambiguities, the method can estimate the UE positions without a dedicated reference node to eliminate clock offsets. Simulations show centimeter-level accuracy in dynamic asynchronous scenarios.

In scenarios where high-precision localization is required, even tiny calibration errors of the base station (BS) can directly and significantly impact the positioning accuracy of the device. Inspired by the excellent precision of carrier phase positioning, this paper proposes a high-precision positioning algorithm based on carrier phase and unscented Kalman filter (UKF) to address BS calibration errors. The state vector of the positioning system is initialized with the initial coordinates of the terminal, which is estimated via Chan's algorithm. The coordinates of the BSs and the terminal are then iteratively refined by combining the double-differential carrier phase and time difference of arrival (TDoA) measurements of multiple moments. Numerical simulations demonstrate that the proposed algorithm achieves centimeter-level positioning accuracy even in the presence of significant BS calibration errors, which effectively overcomes the impact of BS calibration errors on positioning performance.

This paper presents a novel user-centric sensing framework, where a user equipment (UE) acts as the receiver of a multi-static radar sensing system and utilizes the communication signals emitted by base stations (BSs) and scattered by the targets for joint UE and target tracking. Specifically, we propose to locate the UE using the least squares (LS) estimator with a one-dimensional (1D) search and then develop a two-dimensional (2D) target identification approach using the estimated target location and motion of each path based on mean-shift clustering. After the motion parameters of the UE and targets are estimated, a joint UE and target tracking algorithm is designed based on the analysis of the localization and motion estimation errors. Extensive simulations corroborate the ability of our approach to estimate target parameters and cluster and identify targets. Specifically, the average estimation error of the target number is only 0.18. The speed and heading accuracy of the UE and targets is [0.121 m/s, 3.879°] and [0.199 m/s, 4.669°], respectively. In joint UE and target tracking, our scheme outperforms the benchmarks of the extended Kalman filter (EKF) and belief propagation (BP) by at least 33.16% and 10.29%, respectively, even though the EKF and BP require a-priori knowledge of the motion parameters and target identification.

Radio simultaneous localization and mapping (SLAM) is challenging due to multipath propagation. While line-of-sight (LoS) and first-order non-LoS (NLoS) paths, referred to as NLoS-1 paths, play a critical role in SLAM, no existing techniques can effectively separate them from high-order NLoS paths, i.e., NLoS-npaths (n≥ 2). This paper presents a new framework to accurately identify the LoS/NLoS-1 paths and conduct SLAM. The key idea is to define the virtual user equipment (UE) of a NLoS-npath as then-th order reflection of the UE. We discover that the centers of the circles encompassing the UE, a virtual UE associated with a LoS/NLoS-1 path, and each of some other virtual UEs are aligned in a line, if and only if those virtual UEs are all associated with NLoS-1 paths. Accordingly, we propose to identify the LoS/NLoS-1 paths using Hough transform-based line detection, and estimate the UE’s location and the environments with the identified LoS/NLoS-1 paths using maximum likelihood estimation and mean-shift clustering. We analytically confirm that the localization error asymptotically approaches the Cramér-Rao Lower Bound. Simulations show that our approach outperforms the state of the art in localization accuracy by up to 91.93%, even when the latter assumed all NLoS-1 paths are perfectly identifieda-priori.

In complex scenarios where line-of-sight (LOS) and non-line-of-sight (NLOS) paths both exist, a LOS/NLOS identifi-cation method is necessary. In this paper, we propose a millimeter wave (mmWave) LOS/NLOS identification scheme utilizing mean-shift (MS) clustering algorithm and a 3D angle-of-arrival (AOA) localization algorithm using both LOS and one-bound reflection NLOS paths. In order to separate LOS and one-bound NLOS paths from multiple-bound NLOS paths, we first make all possible reflection condition assumptions for all paths to give all possible user equipment (UE) locations. Each path's assumption corresponds to one possible UE location. Then by applying mean-shift clustering to the calculated locations, we find the cluster with the most points as the set of correct hypothetical points. For the points in this cluster, the corresponding LOS/NLOS assumptions are considered to be correct, which means the LOS/NLOS conditions are successfully identified. Given known reflection conditions and original AOA measurements, the position estimate is then solved by the proposed AOA localization algorithm. Simulation results demonstrate that our scheme is capable of achieving high identification accuracy and localization precision.