You can also find my publications on Google Scholar.
My research centers on AI-native wireless communications, with a long-standing research foundation in channel intelligence for advanced MIMO and current interests in efficient, adaptive and trustworthy wireless AI, as well as wireless foundation models and environment intelligence.
Surveys, Perspectives, and Tutorials
This article reviews large AI models for wireless physical-layer communications and organizes existing approaches into two paradigms: leveraging pre-trained large models and developing wireless-native large models.
This article reviews AI-based CSI feedback enhancement from a 3GPP standardization perspective, including evaluation methodology, deployment challenges, and protocol evolution.
ESI Highly Cited Paper
A comprehensive overview of deep learning-based CSI feedback, covering architectures, quantization, multi-rate feedback, practical deployment, and future research directions.
Invited Paper; ESI Highly Cited Paper
This article reviews AI/ML-enabled beam management with emphasis on 5G-Advanced standardization, evaluation, and deployment.
This book chapter introduces AI-native air-interface design and discusses end-to-end communications, single-module enhancement, and practical challenges for integrating AI into future wireless systems.
Research Direction 1: Channel Intelligence for Advanced MIMO
My work in this direction investigates how advanced MIMO systems can efficiently acquire, represent, predict, and exploit channel knowledge under limited pilot, feedback, signaling, and computational resources.
CSI Representation and Feedback
This work develops a flexible CSI compression framework supporting multiple feedback rates and practical quantization.
ESI Highly Cited Paper
This work exploits uplink channel information to assist downlink channel acquisition in FDD massive MIMO.
This work explores a fundamentally different feedback mechanism in which CSI information is embedded into transmitted data rather than occupying dedicated feedback resources.
Task-Oriented Channel Acquisition
This work shifts CSI feedback optimization from minimizing reconstruction error toward directly maximizing downstream beamforming performance.
This work jointly designs channel estimation, compression, and reconstruction instead of optimizing individual channel-acquisition modules separately.
ESI Highly Cited Paper
RIS and Cell-Free Massive MIMO
This work develops a two-timescale CSI feedback framework for jointly supporting beamforming and RIS phase-shift design.
This work exploits inter-user correlation in RIS-user channels to reduce redundant feedback and improve multi-user CSI acquisition efficiency.
This work introduces position-domain channel extrapolation, using the user position as a bridge for transferring channel knowledge across geographically distributed links in cell-free massive MIMO.
Research Direction 2: Efficient, Adaptive and Trustworthy Wireless AI
This direction focuses on making wireless AI models lightweight, adaptive, reliable, and practically deployable under constraints on computation, data, latency, energy, and changing environments.
This work investigates neural-network compression and acceleration for resource-constrained wireless AI deployment.
This work evaluates AI-based CSI feedback using real wireless channel measurements and studies the gap between simulation-based development and practical deployment.
Cover Paper
This work develops distributed online adaptation that exploits spatial correlation among users without requiring centralized retraining.
This work introduces online performance monitoring for AI-based CSI feedback, enabling unreliable model outputs to be identified during deployment.
Selected Collaborative Work
[5] Y. Cui, J. Guo, Z. Cao, et al., “Lightweight Neural Network with Knowledge Distillation for CSI Feedback,” IEEE Transactions on Communications, vol. 72, no. 8, pp. 4917–4929, 2024.
[6] X. Li, J. Guo, C.-K. Wen, X. Geng, and S. Jin, “Auto-CsiNet: Scenario-Customized Automatic Neural Network Architecture Generation for Massive MIMO CSI Feedback,” IEEE Transactions on Wireless Communications, vol. 23, no. 10, pp. 14759–14775, 2024.
[7] X. Li, J. Guo, C.-K. Wen, X. Geng, and S. Jin, “Facilitating AI-Based CSI Feedback Deployment in Massive MIMO Systems with Learngene,” IEEE Transactions on Wireless Communications, vol. 23, no. 9, pp. 11325–11340, 2024.
[8] J. Zhang, J. Guo, X. Li, C.-K. Wen, X. Geng, and S. Jin, “Efficient Deployment of Deep MIMO Detection Using Learngene,” IEEE Transactions on Wireless Communications, vol. 25, pp. 4405–4418, 2026.
Research Direction 3: Wireless Foundation Models and Environment Intelligence
My recent work explores how wireless AI can move beyond task-specific and scenario-specific models by learning and exploiting reusable channel, spatial, semantic, and environment knowledge.
This work explicitly introduces propagation-environment knowledge into CSI feedback while retaining compatibility with conventional codebook-based feedback procedures.
This work jointly exploits channel and location information, demonstrating how shared wireless representations can benefit multiple channel-related tasks.
This work introduces semantic-aware wireless digital twins as an interface between physical environments and AI-based CSI acquisition, with the goal of reducing data-collection and deployment costs.
This work investigates why AI-based CSI feedback generalizes across environments and develops a large CSI model that incorporates environment-specific channel-distribution knowledge through prompts.
Selected Collaborative Work
MUSE-FM develops a unified foundation-model architecture for heterogeneous wireless tasks and explicitly incorporates environmental context to improve cross-scenario adaptation.
[6] Y. Cui, J. Guo, X. Li, C.-K. Wen, and S. Jin, “Large and Small Model Collaboration for Air Interface,” IEEE Transactions on Wireless Communications, accepted, 2026.
This work develops a large-small model collaboration paradigm in which a large model provides reusable wireless knowledge while lightweight models support efficient environment-specific adaptation.
