Qi Sun
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Qi Sun has a total of 48 co-authors. Here are the top co-authors by number of shared publications:
- Chih-Lin I.
China Mobile Research Institute
10 shared publications - Nan Li
China Mobile Research Institute
10 shared publications - Xiang Li
China Mobile Research Institute, Beijing, China
5 shared publications - Jinri Huang
China Mobile Research Institute, China
3 shared publications - Ziqi Chen
China Mobile Research Institute, Beijing, China
3 shared publications - Fengxian Guo
State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing, China
2 shared publications
Latest Publications
- 20 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Wireless Communications Letters
- Vol. 15 (21622337) , 3328-3332
D2AN: Delay-Doppler Attention Network for Channel Estimation in OFDM Systems
Accurate channel estimation is essential for fully exploiting the advantages of orthogonal frequency division multiplexing (OFDM) systems. Compared with conventional estimators, deep neural networks can achieve higher estimation accuracy. Nevertheless, most existing deep learning–based approaches are purely data-driven, which limits their ability to effectively capture the inherent channel correlations. As a result, their performance degrades in dynamic environments and their interpretability remains limited. To address these challenges, this paper proposes a novel architecture, referred to as the Delay–Doppler Attention Network (D2AN). In particular, we design an attention mechanism where the attention coefficients are generated from basis functions sampled in the delay and Doppler domains. By incorporating the structure of channel correlation functions into the attention layers, D2AN explicitly models the underlying spectral and temporal dependencies of wireless channels. Simulation results demonstrate that the proposed D2AN significantly outperforms state-of-the-art deep learning–based counterparts as well as conventional channel estimators. Moreover, D2AN exhibits strong robustness across diverse signal-to-noise ratios, user mobilities, and delay spreads, while maintaining low computational complexity.
- 20 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Wireless Communications Letters
- Vol. 15 (21622337) , 3328-3332
D2AN: Delay-Doppler Attention Network for Channel Estimation in OFDM Systems
Accurate channel estimation is essential for fully exploiting the advantages of orthogonal frequency division multiplexing (OFDM) systems. Compared with conventional estimators, deep neural networks can achieve higher estimation accuracy. Nevertheless, most existing deep learning–based approaches are purely data-driven, which limits their ability to effectively capture the inherent channel correlations. As a result, their performance degrades in dynamic environments and their interpretability remains limited. To address these challenges, this paper proposes a novel architecture, referred to as the Delay–Doppler Attention Network (D2AN). In particular, we design an attention mechanism where the attention coefficients are generated from basis functions sampled in the delay and Doppler domains. By incorporating the structure of channel correlation functions into the attention layers, D2AN explicitly models the underlying spectral and temporal dependencies of wireless channels. Simulation results demonstrate that the proposed D2AN significantly outperforms state-of-the-art deep learning–based counterparts as well as conventional channel estimators. Moreover, D2AN exhibits strong robustness across diverse signal-to-noise ratios, user mobilities, and delay spreads, while maintaining low computational complexity.
- Open Access
- 12 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Vehicular Technology
- Vol. 7, 1559-1575
Measured Energy Consumption for AI-Based Precoding
This paper presents the first measurement-based study on the energy consumption of artificial intelligence (AI) algorithms for radio access networks (RAN), aiming to investigate whether the integration of AI leads to higher energy consumption than traditional non-AI algorithms. We use the multiuser precoding problem as a representative example, where AI is widely recognized for its potential in supporting the growing deployment of antennas for 6G systems. Our assessment examines three AI algorithms using the convolutional neural network (CNN), edge graph neural network (EGNN), and communication model-based graph neural network (MGNN), in comparison with two non-AI algorithms, specifically the weighted minimum mean square error (WMMSE) and regularized zero-forcing (RZF) algorithms. Through extensive measurements of energy and power consumption, we examine how efficient architecture designs of deep neural networks (DNNs) affect energy consumptions, and provide a quantitative comparison between measured results and estimates from existing power estimation models. The results indicate that the MGNN, which is the most efficient architecture among the three DNNs, consumes substantially less energy than the non-AI algorithms for large-scale systems, while the CNN consumes significantly more energy for both training and inference compared to the MGNN. Moreover, utilizing GPUs does not always lead to higher power consumption compared to CPUs, the power consumption of CPUs should be taken into account when GPUs are used, and DRAM contributes little to total power consumption. We demonstrate the shortcomings of existing utilization-based and FLOPs-based models in estimating energy consumption, calling for the development of more accurate models to facilitate green AI designs for RAN.
- 12 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Vehicular Technology
- Vol. 7, 1559-1575
Open AccessMeasured Energy Consumption for AI-Based Precoding
This paper presents the first measurement-based study on the energy consumption of artificial intelligence (AI) algorithms for radio access networks (RAN), aiming to investigate whether the integration of AI leads to higher energy consumption than traditional non-AI algorithms. We use the multiuser precoding problem as a representative example, where AI is widely recognized for its potential in supporting the growing deployment of antennas for 6G systems. Our assessment examines three AI algorithms using the convolutional neural network (CNN), edge graph neural network (EGNN), and communication model-based graph neural network (MGNN), in comparison with two non-AI algorithms, specifically the weighted minimum mean square error (WMMSE) and regularized zero-forcing (RZF) algorithms. Through extensive measurements of energy and power consumption, we examine how efficient architecture designs of deep neural networks (DNNs) affect energy consumptions, and provide a quantitative comparison between measured results and estimates from existing power estimation models. The results indicate that the MGNN, which is the most efficient architecture among the three DNNs, consumes substantially less energy than the non-AI algorithms for large-scale systems, while the CNN consumes significantly more energy for both training and inference compared to the MGNN. Moreover, utilizing GPUs does not always lead to higher power consumption compared to CPUs, the power consumption of CPUs should be taken into account when GPUs are used, and DRAM contributes little to total power consumption. We demonstrate the shortcomings of existing utilization-based and FLOPs-based models in estimating energy consumption, calling for the development of more accurate models to facilitate green AI designs for RAN.
- 19 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Vehicular Technology Magazine
- Vol. 20 (4) , 36-45
AI-Native Management and Orchestration in O-RAN: Enabling the Path to Intelligent 6G Automation
On the path toward 6G, three trends, including radio access networks (RANs) cloudification, the convergence of communication and computing, as well as the adoption of artificial intelligence (AI), are reshaping the landscape of management and orchestration (M&O) for RAN. The RAN M&O should evolve itself to accommodate the requirements of the trend with intelligent automation for 6G. Toward this end, O-RAN has pioneered at this front with its proposed M&O framework called service management and orchestration (SMO). Building upon O-RAN, this paper further proposed an AI-native RAN SMO framework with the capability to support communication and computing convergence. The effectiveness and the superiority of the proposed framework was demonstrated in a vehicle-to-everything (V2X) scenario. Simulations showed that, thanks to the capability of joint communication and computing resource orchestration from the framework, AI-empowered algorithms could greatly improve the system performance in terms of latency of the computing tasks. In particular, under the arrival rate of 0.4 with a 500-B payload, our approach achieves over 35% latency reduction compared with benchmarks, and the gain increases to about 50% when the payload size varies within [350, 650] B.
- 19 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Vehicular Technology Magazine
- Vol. 20 (4) , 36-45
AI-Native Management and Orchestration in O-RAN: Enabling the Path to Intelligent 6G Automation
On the path toward 6G, three trends, including radio access networks (RANs) cloudification, the convergence of communication and computing, as well as the adoption of artificial intelligence (AI), are reshaping the landscape of management and orchestration (M&O) for RAN. The RAN M&O should evolve itself to accommodate the requirements of the trend with intelligent automation for 6G. Toward this end, O-RAN has pioneered at this front with its proposed M&O framework called service management and orchestration (SMO). Building upon O-RAN, this paper further proposed an AI-native RAN SMO framework with the capability to support communication and computing convergence. The effectiveness and the superiority of the proposed framework was demonstrated in a vehicle-to-everything (V2X) scenario. Simulations showed that, thanks to the capability of joint communication and computing resource orchestration from the framework, AI-empowered algorithms could greatly improve the system performance in terms of latency of the computing tasks. In particular, under the arrival rate of 0.4 with a 500-B payload, our approach achieves over 35% latency reduction compared with benchmarks, and the gain increases to about 50% when the payload size varies within [350, 650] B.
- Open Access
Learning Function-to-Function Mappings: A Fourier Neural Operator for Next-Generation MIMO Systems
Next-generation multiple-input multiple-output (MIMO) systems, characterized by extremely large-scale arrays, holographic surfaces, three-dimensional architectures, and flexible antennas, are poised to deliver unprecedented data rates, spectral efficiency and stability. However, these advancements introduce significant challenges for physical layer signal processing, stemming from complex near-field propagation, continuous aperture modeling, sub-wavelength antenna coupling effects, and dynamic channel conditions. Conventional model-based and deep learning approaches often struggle with the immense computational complexity and model inaccuracies inherent in these new regimes. This article proposes a Fourier neural operator (FNO) as a powerful and promising tool to address these challenges. The FNO learns function-to-function mappings between infinite-dimensional function spaces, making them exceptionally well-suited for modeling complex physical systems governed by partial differential equations based on electromagnetic wave propagation. We first present the fundamental principles of FNO, demonstrating its mesh-free nature and function-to-function ability to efficiently capture global dependencies in the Fourier domain. Furthermore, we explore a range of applications of FNO in physical-layer signal processing for next-generation MIMO systems. Representative case studies on channel modeling and estimation for novel MIMO architectures demonstrate the superior performance of FNO compared to state-of-the-art methods. Finally, we discuss open challenges and outline future research directions, positioning FNO as a promising technology for enabling the enormous potential of next-generation MIMO systems.
Open AccessLearning Function-to-Function Mappings: A Fourier Neural Operator for Next-Generation MIMO Systems
Next-generation multiple-input multiple-output (MIMO) systems, characterized by extremely large-scale arrays, holographic surfaces, three-dimensional architectures, and flexible antennas, are poised to deliver unprecedented data rates, spectral efficiency and stability. However, these advancements introduce significant challenges for physical layer signal processing, stemming from complex near-field propagation, continuous aperture modeling, sub-wavelength antenna coupling effects, and dynamic channel conditions. Conventional model-based and deep learning approaches often struggle with the immense computational complexity and model inaccuracies inherent in these new regimes. This article proposes a Fourier neural operator (FNO) as a powerful and promising tool to address these challenges. The FNO learns function-to-function mappings between infinite-dimensional function spaces, making them exceptionally well-suited for modeling complex physical systems governed by partial differential equations based on electromagnetic wave propagation. We first present the fundamental principles of FNO, demonstrating its mesh-free nature and function-to-function ability to efficiently capture global dependencies in the Fourier domain. Furthermore, we explore a range of applications of FNO in physical-layer signal processing for next-generation MIMO systems. Representative case studies on channel modeling and estimation for novel MIMO architectures demonstrate the superior performance of FNO compared to state-of-the-art methods. Finally, we discuss open challenges and outline future research directions, positioning FNO as a promising technology for enabling the enormous potential of next-generation MIMO systems.
- 30 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 40 (1) , 154-164
AI Empowered Modeling, Closed-Loop Optimization, and Field Trials of RAN Digital Twin
As mobile networks evolve, we are witnessing a significant shift from traditional communication services to advanced network applications. This transformation calls for efficient and reliable Radio Access Network (RAN) operations. RAN Digital Twin (DT) is such an enabling technology. This paper delves into the AI-empowered RAN DT technology, analyzing the services that RAN DTs can provide to enhance the performance and operational efficiency of both current 5G networks and upcoming 6G networks. A comprehensive framework is proposed, which includes four subsystems: feature application, data repository, model repository, and RAN DT management. By integrating several key enabling technologies such as data management, AI-empowered RAN modeling techniques, multi-objective RAN performance optimization methods and RAN DT performance management which refers to the optimization of the RAN DT’s own model performance, the proposed RAN Digital Twin system can predict network conditions and fine-tune network configuration in a way that enhances network reliability and user satisfaction. To validate these improvements, practical tests are carried out within real mobile networks, focusing on the joint optimization of network energy efficiency and user data rates. The results demonstrate significant improvements in both network energy consumption and user data rates through RAN DT optimization. These findings underscore the potential of RAN Digital Twins in revolutionizing network operations and achieving high-level network automation.
- 30 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 40 (1) , 154-164
AI Empowered Modeling, Closed-Loop Optimization, and Field Trials of RAN Digital Twin
As mobile networks evolve, we are witnessing a significant shift from traditional communication services to advanced network applications. This transformation calls for efficient and reliable Radio Access Network (RAN) operations. RAN Digital Twin (DT) is such an enabling technology. This paper delves into the AI-empowered RAN DT technology, analyzing the services that RAN DTs can provide to enhance the performance and operational efficiency of both current 5G networks and upcoming 6G networks. A comprehensive framework is proposed, which includes four subsystems: feature application, data repository, model repository, and RAN DT management. By integrating several key enabling technologies such as data management, AI-empowered RAN modeling techniques, multi-objective RAN performance optimization methods and RAN DT performance management which refers to the optimization of the RAN DT’s own model performance, the proposed RAN Digital Twin system can predict network conditions and fine-tune network configuration in a way that enhances network reliability and user satisfaction. To validate these improvements, practical tests are carried out within real mobile networks, focusing on the joint optimization of network energy efficiency and user data rates. The results demonstrate significant improvements in both network energy consumption and user data rates through RAN DT optimization. These findings underscore the potential of RAN Digital Twins in revolutionizing network operations and achieving high-level network automation.
- 15 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 38 (5) , 66-75
Enabling Mobile AI Agent in 6G Era: Architecture and Key Technologies
With the advent of mobile networks, we are witnessing an unprecedented shift in the landscape of mobile network services, evolving from traditional voice calls to advanced AI services. This paper delves into the intricacies of this evolution, particularly emphasizing the deep integration of AI Agents into 6G networks. Despite recent researches in using large language model (LLM) and artificial intelligence (AI) agent for network automation, the fundamental mobile AI agent use cases, their network requirements, potential network architecture and enabling technologies for supporting the pervasive AI agents in 6G ear are largely unexplored. In this article, we present an in-depth analysis of typical mobile AI agent use cases in 6G, consisting of AI agent-based 6G network automation, handheld personalized agents, connected robotics and autonomous systems, and wearable AI agent. Then, we elucidate a novel system architecture that supports identified use cases. The article also addresses core aspects of enabling technologies, including 6G agent and application agent collaboration, efficient model and memory management, coordinated agent-to-agent communication and support of multi-modal data transmission. Finally, two future research directions: security and privacy protection and AI agent tailored communication are discussed. This article lays a foundation for understanding the role of 6G in realizing the full potential of AI agents in various applications.
- 15 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 38 (5) , 66-75
Enabling Mobile AI Agent in 6G Era: Architecture and Key Technologies
With the advent of mobile networks, we are witnessing an unprecedented shift in the landscape of mobile network services, evolving from traditional voice calls to advanced AI services. This paper delves into the intricacies of this evolution, particularly emphasizing the deep integration of AI Agents into 6G networks. Despite recent researches in using large language model (LLM) and artificial intelligence (AI) agent for network automation, the fundamental mobile AI agent use cases, their network requirements, potential network architecture and enabling technologies for supporting the pervasive AI agents in 6G ear are largely unexplored. In this article, we present an in-depth analysis of typical mobile AI agent use cases in 6G, consisting of AI agent-based 6G network automation, handheld personalized agents, connected robotics and autonomous systems, and wearable AI agent. Then, we elucidate a novel system architecture that supports identified use cases. The article also addresses core aspects of enabling technologies, including 6G agent and application agent collaboration, efficient model and memory management, coordinated agent-to-agent communication and support of multi-modal data transmission. Finally, two future research directions: security and privacy protection and AI agent tailored communication are discussed. This article lays a foundation for understanding the role of 6G in realizing the full potential of AI agents in various applications.
- 17 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 38 (2) , 97-112
Communication and Computing Integrated RAN: A New Paradigm Shift for Mobile Network
The Radio Access Network (RAN) consists of numerous base stations that serve as access points for the end users to communicate with application services. Along with the increasingly powerful computing capability in base stations, and the stringent requirements from new services exemplified by augmented reality (AR) and virtual reality (VR), it is realistic and desirable to leverage the spare computing resource in the base stations to deploy such latency sensitive services. This concept, named Communication and Computing Integrated RAN (CCIR), is proposed in this article. CCIR is aimed to providing communication service and computing service in a holistic manner. This article presents design principles, logical architecture, and key enabling technologies of CCIR. Field trials on the computing resource migration among different base stations are conducted to demonstrate feasibility and effectiveness of CCIR. It is observed that the system performance is improved in terms of spectral efficiency and throughput due to the holistic orchestration and scheduling in CCIR.
- 17 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 38 (2) , 97-112
Communication and Computing Integrated RAN: A New Paradigm Shift for Mobile Network
The Radio Access Network (RAN) consists of numerous base stations that serve as access points for the end users to communicate with application services. Along with the increasingly powerful computing capability in base stations, and the stringent requirements from new services exemplified by augmented reality (AR) and virtual reality (VR), it is realistic and desirable to leverage the spare computing resource in the base stations to deploy such latency sensitive services. This concept, named Communication and Computing Integrated RAN (CCIR), is proposed in this article. CCIR is aimed to providing communication service and computing service in a holistic manner. This article presents design principles, logical architecture, and key enabling technologies of CCIR. Field trials on the computing resource migration among different base stations are conducted to demonstrate feasibility and effectiveness of CCIR. It is observed that the system performance is improved in terms of spectral efficiency and throughput due to the holistic orchestration and scheduling in CCIR.
- 28 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Selected Areas in Communications
- Vol. 42 (2) , 487-505
Validation of Current O-RAN Technologies and Insights on the Future Evolution
Entering the 5G era, the mobile network operators (MNO) are facing greater challenges in providing services cost effectively than any other previous generations. The potential solutions to this are lying on the emerging trend of deep convergence of information technology (IT), communication technology (CT) and data technology (DT). In particular, the O-RAN technology, the representation of such ICDT convergence and proposed by the O-RAN ALLIANCE in 2018, is transforming Radio Access Networks towards a new paradigm featuring openness, cloudification and intelligence. O-RAN has gained huge attention from both industry and academia since its inception. In this paper, we presented the recent endeavors from China Mobile, including our deployment scenarios, various test results from open fronthaul, cloud platform to the intelligent controller. Our rich and comprehensive tests have demonstrated the viability and superiority of current O-RAN technologies. Furthermore, we also provide our deep thinking on the O-RAN future evolution in order to better serve the emerging applications such as Metaverse, cloud extended-reality (XR), extensive enterprise private 5G verticals and so on.
- 28 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Selected Areas in Communications
- Vol. 42 (2) , 487-505
Validation of Current O-RAN Technologies and Insights on the Future Evolution
Entering the 5G era, the mobile network operators (MNO) are facing greater challenges in providing services cost effectively than any other previous generations. The potential solutions to this are lying on the emerging trend of deep convergence of information technology (IT), communication technology (CT) and data technology (DT). In particular, the O-RAN technology, the representation of such ICDT convergence and proposed by the O-RAN ALLIANCE in 2018, is transforming Radio Access Networks towards a new paradigm featuring openness, cloudification and intelligence. O-RAN has gained huge attention from both industry and academia since its inception. In this paper, we presented the recent endeavors from China Mobile, including our deployment scenarios, various test results from open fronthaul, cloud platform to the intelligent controller. Our rich and comprehensive tests have demonstrated the viability and superiority of current O-RAN technologies. Furthermore, we also provide our deep thinking on the O-RAN future evolution in order to better serve the emerging applications such as Metaverse, cloud extended-reality (XR), extensive enterprise private 5G verticals and so on.