Signal Processing & Statistical Learning
Extracting structure from noisy, high-dimensional measurements through probabilistic inference, representation learning, and clustering.
About Me
I am a full-time, tenure-track Assistant Professor in the College of Computer Science and Software Engineering at Shenzhen University, with teaching and research responsibilities. As a Ph.D. and master's supervisor, I am authorized to independently supervise Ph.D. students in Computer Science.
I received my Ph.D. in Computer and Information Engineering from The Chinese University of Hong Kong, Shenzhen, in August 2025. My research spans signal processing, optimization, and statistical learning, with a focus on wireless communications and localization. I develop methods that connect data-driven inference with physical models for channel understanding, radio map construction, and positioning. I am a member of IEEE and the China Computer Federation (CCF).
Education
The Chinese University of Hong Kong, Shenzhen
& The University of Hong Kong
Ph.D., Computer and Information Engineering
Beihang University
& University of Malaya
Master's degree
Ocean University of China
& Dalian University of Technology
Bachelor's degree
Honors & Awards
- Outstanding Master's Graduate, Beihang University
- First-Class Scholarship, Beihang University
- Outstanding Undergraduate Graduate, Shandong Province
- Outstanding Undergraduate Graduate, Ocean University of China
- Outstanding Undergraduate Thesis, Ocean University of China
- National Scholarship
- Silver Award, National Undergraduate Electronic Design Contest
- First-Class Scholarship, Ocean University of China
Research Interests
Learning and optimizing from signals to understand wireless environments and locate devices.
Optimization for Wireless Systems
Developing efficient optimization methods for wireless communication, channel estimation, resource allocation, and data-driven network design.
Localization & Radio Mapping
Inferring positions and trajectories from wireless signals, and building radio maps with limited or no location labels.
Publications
38 publications · 22 journal articles · 16 conference papers
About venue metrics and illustrations
Badges use the CCF 2026 international directory (or the separate CCF 2025 Chinese-journal tiers), CAS 2025 broad-category divisions, and the 2024 Journal Impact Factor. These are venue-level indicators; each badge links to its source. Unlisted or unverified indicators are omitted. All publication thumbnails are AI-generated concept illustrations.
Summary sources and illustration notes38 publications
202622 papers
Concept illustration HAO-IDS: Hierarchical Multi-Agent Orchestration for Adaptive Network Intrusion Detection
Introduces a hierarchical team of language-model agents for adaptive network intrusion detection, allocating analysis effort according to traffic difficulty. Environment awareness and continual memory adaptation help the system respond to changing network conditions.
Concept illustration Radiomap-Assisted Vehicle Trajectory Retrieving in Massive MIMO Networks
Investigates the recovery of vehicle trajectories using radio maps in massive MIMO networks.
Concept illustration GAP-LWLM: Geometry-Aware Prompt Tuning for Cross-Scene Localization with a Wireless Foundation Model
Studies geometry-aware prompt tuning to adapt a wireless foundation model for localization across different scenes.
Concept illustration Continuous Location Sensing via Adaptive Radio Map Matching in Massive MIMO Networks
Studies continuous position sensing in massive MIMO networks through adaptive radio map matching.
Concept illustration Survey-Free Radio Map Construction via HMM-Based Coarse-to-Fine Inference
Builds radio maps from unlabeled received-signal-strength sequences using a hidden Markov model and coarse-to-fine location inference. Known indoor layouts and movement continuity provide structure for recovering trajectories without a manual fingerprint survey.
Concept illustration QoE-Aware Resource Allocation for AIGC Services in Low-Altitude Economy Networks
Studies resource allocation for AI-generated-content services in low-altitude networks, with a focus on user quality of experience.
Concept illustration Probabilistic Semantic Communication Enabled MEC Systems for Low-Altitude Economy
Investigates probabilistic semantic communication for mobile edge computing systems serving low-altitude networks.
Concept illustration Mobile Edge Computing with Retrieval-Augmented Large Language Models: A Tutorial
Reviews how retrieval-augmented large language models can support decision-making in mobile edge computing. The tutorial organizes design trade-offs and open challenges, and illustrates grounded decision-making through a multiuser computation-offloading case study.
Concept illustration AudioLLM-Enabled Semantic Communication for Multi-Task Speech Transmission
Explores audio large language models as a basis for semantic communication across multiple speech transmission tasks.
Concept illustration Topology-Aware Embedding Network for Label-Free Radio Map Construction
Develops a topology-aware embedding approach for constructing radio maps without labeled location measurements.
Concept illustration Wireless Signals as Tokens: Repurposing Large Language Models for Indoor Localization
Represents wireless measurements as tokens so that a large language model can be adapted for indoor localization. The RATLoc framework combines lightweight adaptation with coordinate prediction to accommodate variable wireless inputs.
Concept illustration Constructing Radio Map via Spatio-Temporal Learning of Crowdsourced Trajectory Data
Uses spatial and temporal structure in crowdsourced trajectory data to construct radio maps.
Concept illustration Integrated Sensing and Communication with a Shared Channel Map
Uses a shared channel map as prior knowledge for both wireless sensing and communication. The framework combines map-assisted target tracking with channel estimation to connect the two tasks through a common environmental representation.
Concept illustration AEGIS: Anatomy-Embedded Group-Invariant Segmentation for Fair Medical Foundation Models
Combines attribute-conditioned feature alignment and history-aware loss weighting to reduce subgroup disparities when adapting medical segmentation foundation models. The design aims to preserve anatomical information while reducing sensitivity to acquisition-related variation.
Concept illustration Energy-Efficient Multi-UAV-Enabled MEC Systems over Space-Air-Ground Integrated Networks
Studies energy-efficient mobile edge computing across a network of satellites, UAVs, and ground users. Joint optimization of user association, UAV trajectories, computation offloading, processing resources, and transmission power coordinates communication and computation.
Concept illustration Movable-Antenna Enabled Robust Vehicular Consumer Networks Under Imperfect CSI
Jointly optimizes movable-antenna positions and transmit beamforming for vehicular networks with uncertain channel information. The robust formulation targets worst-case communication performance while accounting for power and antenna-motion constraints.
Concept illustration An Efficient Federated Large Language Model Recommendation Framework for Enhanced Personalization
Studies an efficient federated large language model framework for personalized recommendation.
Published in Chinese; title translated into English.
Concept illustration Blind Radio Map Construction via Topology Guided Manifold Learning
Constructs radio maps from unlabeled channel measurements by learning a low-dimensional representation that preserves their spatial topology. Alignment with geographic structure connects the learned representation to physical locations.
Concept illustration Wireless Human Activity Recognition: A Survey of Recent Advances and a Novel Zero-Shot LLM-Based Inference Framework
Surveys recent wireless human activity recognition methods and their practical challenges. It also introduces a zero-shot large-language-model framework that interprets wireless measurements using physical activity knowledge and temporal reasoning.
Concept illustration Temporal visual semantics-induced human motion understanding with large language models
Uses large language models to extract temporal visual semantics and guide subspace clustering of human motion. A feedback loop couples segmentation with representation learning to keep motion groups temporally consistent.
Concept illustration Clustering structure identification with ordering graph
Reorders graph similarity matrices to expose block-diagonal cluster structure, then identifies clusters with a greedy partition algorithm. The method combines density-aware graph traversal with theoretical guarantees under separated-cluster conditions.
Concept illustration Physical Embedding for Radio Map Construction
Learns a two-dimensional physical embedding of unlabeled CSI by combining angle-of-arrival, time-of-arrival, and power–angle–delay similarity. The embedding preserves spatial topology to support radio-map construction without location labels.
20259 papers
Concept illustration Blind Construction of Angular Power Maps in Massive MIMO Networks
Infers measurement locations and angular power maps from unlabeled massive-MIMO CSI using a hidden Markov model of propagation and mobility. The work also analyzes localization limits and uses reconstructed maps for channel prediction.
Concept illustration Trajectory Map-Matching in Urban Road Networks Based on RSS Measurements
Reconstructs vehicle trajectories from RSS measurements while respecting road-network connectivity and motion constraints. An HMM-based embedding combines signal propagation with temporal dependencies, aided by speed-constrained initialization.
Concept illustration Cervical-RG: Automated Cervical Cancer Report Generation from 3D Multi-sequence MRI via CoT-guided Hierarchical Experts
Generates cervical-cancer reports from multi-sequence 3D MRI with a hierarchy of diagnostic and staging experts. Chain-of-thought reasoning and cross-stage consistency checks connect image findings with FIGO staging.
Concept illustration Unsupervised Radio Map Construction in Mixed LoS/NLoS Indoor Environments
Jointly models channel power, delay, angle, and user motion to recover indoor trajectories without location labels. The HMM framework distinguishes LOS and NLOS propagation when building a radio map.
Concept illustration HMM-Based CSI Embedding for Trajectory Recovery via Feature Engineering on MIMO-OFDM Channels in LOS/NLOS Regions
Uses engineered MIMO-OFDM channel features and hidden Markov modeling to recover trajectories in both line-of-sight and non-line-of-sight regions.
Concept illustration Bayesian-based Transmission and Channel Knowledge Map Reconstruction in LEO Satellite-terrestrial Communication
Studies Bayesian transmission and channel knowledge map reconstruction for low-Earth-orbit satellite–terrestrial communication.
Concept illustration Constructing Angular Power Maps in Massive MIMO Networks Using Measurements without Location Labels
Builds angular power maps from massive-MIMO measurements without location labels. A hidden Markov model links temporal CSI variation to the mobile trajectory, providing inferred locations for map construction and channel prediction.
Concept illustration Block-Diagonal Structure Learning for Subspace Clustering
Exposes subspace clusters by permuting the similarity matrix into block-diagonal form and identifying its diagonal blocks. The approach combines density-based traversal with a greedy segmentation rule.
Concept illustration Calibration-Free Indoor Positioning via Regional Channel Tracing
Builds calibration-free indoor radio maps from RSS sequences by tracing movement through known regions. Signal-subspace clustering and regional location tagging recover measurement positions without inertial sensors or manually supplied labels.
20245 papers
Concept illustration Segmentation and completion of human motion sequence via temporal learning of subspace variety model
Jointly segments and completes incomplete human-motion sequences with a temporally constrained subspace-variety model. Alternating subspace learning and motion assignment exploit temporal consistency, while missing entries are reconstructed from the learned motion subspaces.
Concept illustration Block-Diagonal Guided DBSCAN Clustering
Guides DBSCAN-style traversal with the block-diagonal structure of a learned similarity graph. Ordering followed by split-and-refine partitioning identifies clusters while addressing sensitivity to density parameters and high-dimensional data.
Concept illustration HMM-based CSI Embedding for Trajectory Recovery from RSS Measurements of Non-Cooperative Devices
Recovers trajectories of non-cooperative devices through hidden Markov modeling of received-signal-strength measurements.
Concept illustration Unsupervised action segmentation via fast learning of semantically consistent actoms
Segments video without a training stage by first splitting it into semantically consistent action units, or actoms. A spatio-temporal merging step reunites units representing the same action, balancing segmentation quality and computational cost.
Concept illustration Constructing indoor region-based radio map without location labels
Constructs region-based indoor radio maps from RSS measurements collected without location labels. Sequential subspace clustering groups measurements, and graph-based matching associates the resulting clusters with physical regions.
20222 papers
Concept illustration Integrated segmentation and subspace clustering for RSS-based localization under blind calibration
Reduces indoor-localization calibration effort by jointly segmenting and clustering blindly collected RSS sequences. The method exploits collection order and region-specific signal subspaces to recover useful location regions without recording the device route.
Concept illustration Spectrum efficiency prediction for real-world 5G networks based on drive testing data
Predicts spectrum efficiency in real 5G networks from drive-test RSRP measurements using model-assisted learning. Joint interference estimation and classification-assisted prediction address information that signal-strength measurements alone do not capture.
No publications match your search. Try another keyword or change the filters.
Prospective Students
Graduate Research Opportunities
I welcome motivated students interested in signal processing, optimization, statistical learning, wireless communications, and localization. Master's opportunities are available in Software Engineering, Artificial Intelligence, and Computer Technology.
Prospective Ph.D. students in Computer Science are welcome to contact me to discuss research interests and supervision.
zhengx@szu.edu.cnWhat I Look For
- A solid foundation in computer science, calculus, linear algebra, numerical analysis, probability, and statistics.
- Proficiency in at least two of MATLAB, Python, and C/C++.
- Strong English reading and writing skills, curiosity, persistence, and a responsible approach to research.
Undergraduate Research
I welcome first- through third-year undergraduates who are interested in signal processing, statistical learning, and optimization, particularly wireless signal analysis and localization. Strong mathematics and programming skills, along with sustained time for research, are essential. Relevant preparation includes linear algebra, probability and statistics, matrix theory, and MATLAB or Python.
Academic Service & Teaching
I serve as a reviewer for international journals and conferences.
Journal Reviewing
- IEEE Journal on Selected Areas in Communications (JSAC)
- IEEE Transactions on Signal Processing (TSP)
- IEEE Transactions on Wireless Communications (TWC)
- IEEE Transactions on Mobile Computing (TMC)
- IEEE Transactions on Communications (TCOM)
Conference Reviewing
- International Conference on Learning Representations (ICLR)
- International Conference on Machine Learning (ICML)
- AAAI Conference on Artificial Intelligence (AAAI)
- IEEE International Conference on Communications (ICC)
- IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
Teaching
University Computer Fundamentals · Shenzhen University
Footprints
Contact
College of Computer Science and Software Engineering
Shenzhen University · Shenzhen, Guangdong, China