Wenjun Yu

Wenjun Yu

Auto-generated profile, based on publicly available information

Wenjun Yu

School of Communications and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing, China

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
11 February 2025 in IEEE Internet of Things Journal
Verified
Journal Article
Closed access
QPQiaolin Pu
Qiaolin Pu
WYWenjun Yu
Wenjun Yu
XLXin Lan
Xin Lan
MZMu Zhou
Mu Zhou
Xiaolong YangXiaolong Yang
Xiaolong Yang
With the continuous development of 6G communication technology and artificial intelligence (AI), reconfigurable intelligent surface (RIS) technology and generative AI (GAI) have received widespread attention. Hence, combining these two techniques to solve the problem of nonlocalizability in single access point (AP) scenarios is promising. This article proposes a novel RIS-aided localization scheme via generative artificial intelligence (GAI). Specifically, first, considering the presence of a certain amount of noise and the low discriminative features of the original collected received signal strength (RSS) brought by integrated reflective elements of RIS, we construct a generative adversarial network (GAN) model named variational autoencoder-convolutional neural network (VAE-CNN). It can effectively perform noise reduction in the data preprocessing stage to reduce unnecessary redundant information, and extract more discriminative features through convolutional networks to improve the differentiation between data. Second, the target’s location in the spatial domain is formulated as a sparse vector, and then a sparse recovery model is introduced to solve the location estimation problem in RIS-aided localization scenarios. Finally, considering the case that RIS has multiple reflective elements, which will lead to a high dimension of the measurement matrix in the sparse recovery model, we further apply the semi-tensor product (STP) sparse recovery theory on the model to reduce the storage space and high time consumption. Experimental results show that our proposed methods outperform the traditional approaches and reduce the computational complexity simultaneously.

4 March 2025 in IEEE Transactions on Instrumentation and Measurement
Verified
Journal Article
Closed access
QPQiaolin Pu
Qiaolin Pu
FJFengyi Jiang
Fengyi Jiang
MZMu Zhou
Mu Zhou
WYWenjun Yu
Wenjun Yu
MCMin Chen
Min Chen
The Direct Position Determination (DPD) algorithm, which combines the parameter estimation and position calculation procedures together, has attracted significant attention due to its good performance in low Signal-to-Noise Ratio (SNR) situations. However, the high computing consumption and being only suitable for non-coherent signals make its application limited in real situations. To address these problems, we first propose to use the Kronecker theory to decompose the original signal into signal and noise subspaces, based on which a cost function is constructed thus the localization is transformed into a non-convex optimization problem. Compared with traditional decomposition algorithms, this processing can improve the freedom degree of linear arrays and address the challenge to identify the direct transmitting path in the case of coherent signals. Then, to solve the non-convex optimization problem, an adaptive reverse particle swarm optimization (ARPSO) algorithm is proposed, in which the adaptive elite mutation strategy is introduced into the traditional particle swarm optimization (PSO) algorithms to avoid obtaining the local optimal solution caused by local convergence. Moreover, we adopt a non-inertial velocity update scheme to guide the particle flight trajectories. This scheme accelerates the process of convergence, saving the computing overhead a lot. The experimental results show that our proposed decomposition algorithm performs well for coherent and incoherent signals, and the improved optimization algorithm outperforms existing algorithms in both precision and computing consumption.

17 April 2026 in IEEE Transactions on Vehicular Technology
Verified
Journal Article
Closed access
MCMin Chen
Min Chen
QPQiaolin Pu
Qiaolin Pu
MZMu Zhou
Mu Zhou
WYWenju Yu
Wenju Yu
This work addresses the challenges of feature redundancy, fingerprint drift, and insufficient matching robustness in Channel State Information (CSI)-based fingerprint matching and indoor positioning. To address these issues, we propose a feature optimization and matching algorithm based on a graph attention mechanism. Firstly, a Graph Attention Network (GAT)–Variational Graph Auto-Encoder (VGAE) graph embedding model is developed to adaptively allocate node weights, enhancing feature discriminability and generalizability. Secondly, the domain-agnostic mutual prompting (DAMP) strategy is introduced to achieve robust cross-domain distribution matching through bidirectional mutual feature alignment, effectively suppressing CSI drift. Finally, the FM index-based alignment algorithm (FMAlign), inspired by biological sequences, is employed for fast anchor segment matching of embedded fingerprints, thereby improving the accuracy and robustness of the localization process. Experimental results show a Root Mean Square Error (RMSE) of 1.95 m and an average positioning error of 0.57 m. The method demonstrates high efficiency and stability in feature optimization, drift suppression, and sequence matching, providing a reliable solution for high-precision indoor positioning.

12 January 2026 in IEEE Transactions on Vehicular Technology
Verified
Journal Article
Closed access
QPQiaolin Pu
Qiaolin Pu
WYWenjun Yu
Wenjun Yu
MZMu Zhou
Mu Zhou
MCMin Chen
Min Chen
Channel State Information (CSI), encompassing both spatial richness and temporal dynamics, can be effectively represented as a three-dimensional tensor to retain its inherent structural characteristics. However, effectively processing such tensors while maintaining their structural integrity remains a significant challenge. Moreover, CSI data is highly susceptible to noise, which can severely degrade localization performance. To address these issue, this study proposes a novel indoor positioning method named LSFSTM based on CSI tensor. Specifically, first, a Variational Autoencoder-Convolutional Neural Network (VAE-CNN) model is developed to denoise tensor data samples, solving the problem of noise in CSI data that can affect the accuracy of positioning. Second, the Fuzzy Support Tensor Machine (FSTM) is proposed based on the traditional STM where a sigmoid fuzzy membership function is defined to handle the internal noise, to better capture the fuzzy relationships in the data, thus further improving robustness. Moreover, the binary FSTM model is then extended to a multi-class framework through the integration of Lagrange multipliers and Tensor-Tucker decomposition, resolving the challenge of tensor data vectorization, which can disrupt the structural integrity of the data. Finally, a Least Squares is integrated into the FSTM to reduce the computational complexity associated with solving the optimization problem, making the model more efficient and scalable for large datasets. Experimental results demonstrate that the proposed method significantly improves positioning accuracy.