Li Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Li Liu has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Sanshan Sun
College of Physics and Electronic Engineering, Sichuan Normal University, Chengdu, China
14 shared publications - Youzhi Xiong
College of Physics and Electronic Engineering, Sichuan Normal University, Chengdu, China
14 shared publications - Ning Wei
National Key Laboratory of Science and Technology on Communications, University of Electronic Science and Technology of China, Chengdu, China
12 shared publications - Zhongpei Zhang
University of Electronic Science and Technology of China
12 shared publications - Songjie Yang
University of Electronic Science and Technology of China
6 shared publications - Sun Mao
Sichuan Normal University
4 shared publications
Latest Publications
- 23 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. PP (00189545) , 1-14
Participation Priority-Based Model Parameter Diffusion Mechanism for Non-IID Data in FL-Empowered Cellular Networks
Benefited from its distributed nature, Federated Learning (FL) preserves data privacy and reduces bandwidth pressure associated with transmitting large volumes of data over wireless links. However, in FL-empowered cellular systems, key challenges including device heterogeneity, geographically imbalanced data distributions, and dynamic bandwidth demands of parameter transmission severely degrade model performance. To address these issues, we propose a participation priority-based model parameter diffusion mechanism within a hierarchical federated edge learning (HFEL) architecture, with aim to improve the convergence efficiency of global model. In this mechanism, a cloud server collects the data proportion and historical training performance of training devices through gNBs and calculates a participation priority for each device. During the edge aggregation performed by gNBs, the devices that have completed training are selected based on this priority to promptly diffuse their trained parameters to the gNBs that encounter difficulties in obtaining timely and sufficient local models, which guarantees harmonious and high-quality edge models for further global aggregation on the cloud server. To realize this mechanism efficiently, we formulate a total-latency minimization problem for each global aggregation round under accuracy, energy, and delay constraints, and decompose it into device-side resource allocation and network-wide diffusion decision subproblems that are solved alternately. Analytical and experimental results show that, even with computing-constrained devices and Non-IID data, the proposed mechanism allows the global model to approach the accuracy and loss of an ideal HFEL system with modest energy and time overhead.
- 23 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. PP (00189545) , 1-14
Participation Priority-Based Model Parameter Diffusion Mechanism for Non-IID Data in FL-Empowered Cellular Networks
Benefited from its distributed nature, Federated Learning (FL) preserves data privacy and reduces bandwidth pressure associated with transmitting large volumes of data over wireless links. However, in FL-empowered cellular systems, key challenges including device heterogeneity, geographically imbalanced data distributions, and dynamic bandwidth demands of parameter transmission severely degrade model performance. To address these issues, we propose a participation priority-based model parameter diffusion mechanism within a hierarchical federated edge learning (HFEL) architecture, with aim to improve the convergence efficiency of global model. In this mechanism, a cloud server collects the data proportion and historical training performance of training devices through gNBs and calculates a participation priority for each device. During the edge aggregation performed by gNBs, the devices that have completed training are selected based on this priority to promptly diffuse their trained parameters to the gNBs that encounter difficulties in obtaining timely and sufficient local models, which guarantees harmonious and high-quality edge models for further global aggregation on the cloud server. To realize this mechanism efficiently, we formulate a total-latency minimization problem for each global aggregation round under accuracy, energy, and delay constraints, and decompose it into device-side resource allocation and network-wide diffusion decision subproblems that are solved alternately. Analytical and experimental results show that, even with computing-constrained devices and Non-IID data, the proposed mechanism allows the global model to approach the accuracy and loss of an ideal HFEL system with modest energy and time overhead.
- 8 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 30 (10897798) , 332-336
Multi-Agent Reinforcement Learning-Based Joint Optimization for Movable Antenna and RIS-Assisted Systems
This letter investigates a multi-user multi-input single-output (MISO) system assisted by movable antennas (MAs) and the reconfigurable intelligent surface (RIS). Specifically, the positions of MAs and beamforming matrix at the base station (BS) together with the phase shifts at the RIS are jointly optimized to maximize the sum spectral efficiency (SE) over all users. To address the non-convex problem with coupled variables, we formulate the problem as a Markov decision process (MDP) and propose a multi-agent reinforcement learning (MARL)-based method which entails multiple agents for optimizing MA positions and one agent for optimizing BS beamforming and RIS phase shift matrix. Simulation results demonstrate that the proposed algorithm can effectively learn preferable strategies from wireless environments compared to other learning-based benchmarks, yielding significant improvements in sum SE in comparison with fixed-position antennas.
- 8 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 30 (10897798) , 332-336
Multi-Agent Reinforcement Learning-Based Joint Optimization for Movable Antenna and RIS-Assisted Systems
This letter investigates a multi-user multi-input single-output (MISO) system assisted by movable antennas (MAs) and the reconfigurable intelligent surface (RIS). Specifically, the positions of MAs and beamforming matrix at the base station (BS) together with the phase shifts at the RIS are jointly optimized to maximize the sum spectral efficiency (SE) over all users. To address the non-convex problem with coupled variables, we formulate the problem as a Markov decision process (MDP) and propose a multi-agent reinforcement learning (MARL)-based method which entails multiple agents for optimizing MA positions and one agent for optimizing BS beamforming and RIS phase shift matrix. Simulation results demonstrate that the proposed algorithm can effectively learn preferable strategies from wireless environments compared to other learning-based benchmarks, yielding significant improvements in sum SE in comparison with fixed-position antennas.
- 28 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (24) , 55822-55837
Rotatable and Movable Antenna Enhanced Multiuser Communications: Rotation and Position Optimization
Movable antenna (MA) is a promising technology that can enhance communication performance by properly adjusting the antenna position within a local region at transceivers. To further explore the potential of an antenna array, this article proposes a new rotatable and MA (RMA) architecture where the antenna array at a base station (BS) not only employs multiple MAs but also is capable of being rotated along its yaw, pitch, and roll angles. In this context, we first characterize the wireless channel with respect to different rotation angles and antenna positions and formulate an optimization problem to maximize the downlink sum rate under practical system constraints. Subsequently, we solve the nonconvex problem for single-user and multiuser scenarios, respectively. In particular, for the single-user case, we optimize the rotation angles and antenna positions to maximize the user’s rate and propose a gradient ascent (GA) algorithm based on the alternating optimization (AO) framework. For the multiuser scenario with the purpose of maximizing the sum rate of all users, we make the original problem more tractable by exploiting the Lagrangian dual transform and fractional programming (FP) techniques. On this basis, a GA-based algorithm is also proposed to jointly optimize the rotation angles and MAs’ positions together with the precoding matrix at the BS in an iterative manner. Finally, numerical results show that the RMA architecture can improve the sum rate by using the proposed algorithm to adjust rotation angles and antenna positions, compared to the element-level MA, rotatable antenna, and fixed-position antenna. Moreover, the proposed optimization algorithm outperforms its counterparts in achieving a tradeoff between performance and computational complexity.
- 28 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (24) , 55822-55837
Rotatable and Movable Antenna Enhanced Multiuser Communications: Rotation and Position Optimization
Movable antenna (MA) is a promising technology that can enhance communication performance by properly adjusting the antenna position within a local region at transceivers. To further explore the potential of an antenna array, this article proposes a new rotatable and MA (RMA) architecture where the antenna array at a base station (BS) not only employs multiple MAs but also is capable of being rotated along its yaw, pitch, and roll angles. In this context, we first characterize the wireless channel with respect to different rotation angles and antenna positions and formulate an optimization problem to maximize the downlink sum rate under practical system constraints. Subsequently, we solve the nonconvex problem for single-user and multiuser scenarios, respectively. In particular, for the single-user case, we optimize the rotation angles and antenna positions to maximize the user’s rate and propose a gradient ascent (GA) algorithm based on the alternating optimization (AO) framework. For the multiuser scenario with the purpose of maximizing the sum rate of all users, we make the original problem more tractable by exploiting the Lagrangian dual transform and fractional programming (FP) techniques. On this basis, a GA-based algorithm is also proposed to jointly optimize the rotation angles and MAs’ positions together with the precoding matrix at the BS in an iterative manner. Finally, numerical results show that the RMA architecture can improve the sum rate by using the proposed algorithm to adjust rotation angles and antenna positions, compared to the element-level MA, rotatable antenna, and fixed-position antenna. Moreover, the proposed optimization algorithm outperforms its counterparts in achieving a tradeoff between performance and computational complexity.
- 23 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Signal Processing
- Vol. 19 (6) , 1055-1071
Cell-Free Massive MIMO Networks With Low-Resolution Quantization: Downlink Performance Analysis and Optimization
In this paper, we consider the downlink data transmission of a cell-free massive multiple-input multiple-output (MIMO) network with capacity-constrained fronthauls and low-resolution quantization. Primarily, we provide a data-assisted channel estimator by exploiting uplink pilots and detected data in the context of time division duplex (TDD) mode. Taking into consideration the imperfect channel state information (CSI) and the impact of quantization error, we then derive the theoretical expressions of the downlink spectral efficiency (SE) for two precoders, i.e., the maximum ratio transmission (MRT) and partial minimum mean-square error (PMMSE) precoders. Thirdly, based on the SE expression and power consumption model, we formulate and solve an optimization problem that maximizes the downlink energy efficiency (EE) under the constraints of transmit power and fronthaul capacity. Particularly, by leveraging the Dinkelbach transform, Lagrangian dual transform and fractional programming (FP) techniques, we propose an alternating maximization (AM) algorithm to achieve the quantization bit allocation (BA) and downlink power allocation (PA) schemes simultaneously. Finally, simulation results validate the accuracy of the analytical SEs and demonstrate that the proposed AM algorithm has benefits for green communications by improving system EE and saving more power.
- 23 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Signal Processing
- Vol. 19 (6) , 1055-1071
Cell-Free Massive MIMO Networks With Low-Resolution Quantization: Downlink Performance Analysis and Optimization
In this paper, we consider the downlink data transmission of a cell-free massive multiple-input multiple-output (MIMO) network with capacity-constrained fronthauls and low-resolution quantization. Primarily, we provide a data-assisted channel estimator by exploiting uplink pilots and detected data in the context of time division duplex (TDD) mode. Taking into consideration the imperfect channel state information (CSI) and the impact of quantization error, we then derive the theoretical expressions of the downlink spectral efficiency (SE) for two precoders, i.e., the maximum ratio transmission (MRT) and partial minimum mean-square error (PMMSE) precoders. Thirdly, based on the SE expression and power consumption model, we formulate and solve an optimization problem that maximizes the downlink energy efficiency (EE) under the constraints of transmit power and fronthaul capacity. Particularly, by leveraging the Dinkelbach transform, Lagrangian dual transform and fractional programming (FP) techniques, we propose an alternating maximization (AM) algorithm to achieve the quantization bit allocation (BA) and downlink power allocation (PA) schemes simultaneously. Finally, simulation results validate the accuracy of the analytical SEs and demonstrate that the proposed AM algorithm has benefits for green communications by improving system EE and saving more power.
- 10 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (11) , 18249-18265
RIS-Aided Cell-Free Massive MIMO Systems With Low-Resolution ADCs: Uplink Performance Analysis and Optimization
This article investigates the uplink performance of reconfigurable intelligent surface (RIS)-aided cell-free (CF) massive multiple-input-multiple-output (mMIMO) systems over spatially correlated Rayleigh fading channels. We consider multiple RISs and low-resolution analog-to-digital converters (ADCs) to improve the system energy efficiency (EE). We first provide an aggregated channel estimation technique with less pilot overhead. By exploiting the statistical channel state information (CSI), we further optimize the RISs’ phase shifts with the goal of minimizing the total normalized mean square error (NMSE) of the estimated aggregated channels. Subsequently, we derive the closed-form expression of the uplink spectral efficiency (SE) for quantization-aware minimum mean-square error (MMSE) combining. Third, based on the closed-form SE expression and power consumption model, we formulate and solve an optimization problem that maximizes the uplink EE under the constraints of transmit power and total ADC quantization bits. Specifically, by leveraging the Dinkelbach transform, Lagrangian dual transform, and fractional programming (FP) techniques, an alternating optimization (AO)-based algorithm is proposed to jointly obtain the bit allocation (BA) scheme among all access points (APs) and the uplink power control (PC) strategy for all users. Finally, numerical results validate the correctness of the closed-form SE expression and show the effectiveness of the proposed optimization methods for phase shift design and EE maximization.
- 10 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (11) , 18249-18265
RIS-Aided Cell-Free Massive MIMO Systems With Low-Resolution ADCs: Uplink Performance Analysis and Optimization
This article investigates the uplink performance of reconfigurable intelligent surface (RIS)-aided cell-free (CF) massive multiple-input-multiple-output (mMIMO) systems over spatially correlated Rayleigh fading channels. We consider multiple RISs and low-resolution analog-to-digital converters (ADCs) to improve the system energy efficiency (EE). We first provide an aggregated channel estimation technique with less pilot overhead. By exploiting the statistical channel state information (CSI), we further optimize the RISs’ phase shifts with the goal of minimizing the total normalized mean square error (NMSE) of the estimated aggregated channels. Subsequently, we derive the closed-form expression of the uplink spectral efficiency (SE) for quantization-aware minimum mean-square error (MMSE) combining. Third, based on the closed-form SE expression and power consumption model, we formulate and solve an optimization problem that maximizes the uplink EE under the constraints of transmit power and total ADC quantization bits. Specifically, by leveraging the Dinkelbach transform, Lagrangian dual transform, and fractional programming (FP) techniques, an alternating optimization (AO)-based algorithm is proposed to jointly obtain the bit allocation (BA) scheme among all access points (APs) and the uplink power control (PC) strategy for all users. Finally, numerical results validate the correctness of the closed-form SE expression and show the effectiveness of the proposed optimization methods for phase shift design and EE maximization.
- 30 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 28 (8) , 1924-1928
Separate Channel Estimation and Data Detection for Semi-Passive RIS-Empowered Massive MIMO
This letter investigates the channel estimation and data detection for semi-passive reconfigurable intelligent surface (RIS) assisted millimeter wave massive multiple-input-multiple-output systems. We first achieve the coarse user-to-RIS and RIS-to-base station channels by leveraging partial active elements at the RIS. We then develop a double-layer bilinear generalized approximate message passing (DL-BiGAMP) algorithm to refine the individual channels and recover data symbols. The numerical results demonstrate that the proposed DL-BiGAMP technique can improve the performance of the channel estimation and data detection compared with other counterparts.
- 30 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 28 (8) , 1924-1928
Separate Channel Estimation and Data Detection for Semi-Passive RIS-Empowered Massive MIMO
This letter investigates the channel estimation and data detection for semi-passive reconfigurable intelligent surface (RIS) assisted millimeter wave massive multiple-input-multiple-output systems. We first achieve the coarse user-to-RIS and RIS-to-base station channels by leveraging partial active elements at the RIS. We then develop a double-layer bilinear generalized approximate message passing (DL-BiGAMP) algorithm to refine the individual channels and recover data symbols. The numerical results demonstrate that the proposed DL-BiGAMP technique can improve the performance of the channel estimation and data detection compared with other counterparts.
- 22 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 28 (3) , 642-646
Data-Aided Channel Estimation and Combining for Cell-Free Massive MIMO With Low-Resolution ADCs
This paper aims to alleviate the impact of pilot contamination stemming from sharing pilots in cell-free massive multiple-input multiple-output (MIMO) networks with low-resolution analog-to-digital converters (ADCs). Toward this goal, we treat the decided data symbols as additional pilots and develop a data-aided channel estimator to enhance the performance of channel estimation and data detection. Furthermore, by exploiting the statistical characteristic of the estimated channel state information (CSI), we derive the analytical achievable uplink spectral efficiency (SE) for minimum mean-squared error (MMSE) combining. Numerical results show that, in comparison with the pilot-based channel estimators, the data-aided one can obtain more accurate CSI and in turn facilitate the data detection.
- 22 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 28 (3) , 642-646
Data-Aided Channel Estimation and Combining for Cell-Free Massive MIMO With Low-Resolution ADCs
This paper aims to alleviate the impact of pilot contamination stemming from sharing pilots in cell-free massive multiple-input multiple-output (MIMO) networks with low-resolution analog-to-digital converters (ADCs). Toward this goal, we treat the decided data symbols as additional pilots and develop a data-aided channel estimator to enhance the performance of channel estimation and data detection. Furthermore, by exploiting the statistical characteristic of the estimated channel state information (CSI), we derive the analytical achievable uplink spectral efficiency (SE) for minimum mean-squared error (MMSE) combining. Numerical results show that, in comparison with the pilot-based channel estimators, the data-aided one can obtain more accurate CSI and in turn facilitate the data detection.
- 16 May 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 72 (10) , 13723-13728
Performance Analysis of Cell-Free Massive MIMO Network With Large-Scale-Fading Decoding and Low-Resolution ADCs
Concentrating on cell-free massive multiple-input multiple-output (MIMO) networks with low-resolution analog-to-digital converters (ADCs), we derive the theoretical achievable uplink spectral efficiency (SE) for the minimum mean-squared error (MMSE)-based large-scale fading decoding (LSFD). Moreover, we investigate the bit allocation (BA) among access points (APs) to maximize the asymptotic sum SE under the constraint of total ADC quantization bits. Simulation results confirm that our theoretical analysis is accurate and validate that the proposed BA technique is preferable to the random and fixed counterparts.
- 16 May 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 72 (10) , 13723-13728
Performance Analysis of Cell-Free Massive MIMO Network With Large-Scale-Fading Decoding and Low-Resolution ADCs
Concentrating on cell-free massive multiple-input multiple-output (MIMO) networks with low-resolution analog-to-digital converters (ADCs), we derive the theoretical achievable uplink spectral efficiency (SE) for the minimum mean-squared error (MMSE)-based large-scale fading decoding (LSFD). Moreover, we investigate the bit allocation (BA) among access points (APs) to maximize the asymptotic sum SE under the constraint of total ADC quantization bits. Simulation results confirm that our theoretical analysis is accurate and validate that the proposed BA technique is preferable to the random and fixed counterparts.
- 26 December 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 27 (2) , 721-725
Joint Effective Channel Estimation and Data Detection for RIS-Aided Massive MIMO Systems With Low-Resolution ADCs
This paper investigates the uplink data detection for reconfigurable intelligent surface (RIS)-assisted massive multiple-input multiple-output (MIMO) systems with low-resolution quantization at the base stations (BSs). By exploiting the statistical characteristics of the user-to-RIS and RIS-to-BS channels and the central limit theorem (CLT), we first demonstrate that the components in the effective channel are approximately Gaussian distributed under propagation in sub-6 GHz bands. Based on the bilinear generalized approximate message passing (BiG-AMP) algorithm, we then provide a quantization-aware approach to estimate the effective channel and recover transmit symbols. Finally, numerical results show that the revised BiG-AMP technique can improve the performance of channel estimation and data detection with affordable computational complexity.
- 26 December 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 27 (2) , 721-725
Joint Effective Channel Estimation and Data Detection for RIS-Aided Massive MIMO Systems With Low-Resolution ADCs
This paper investigates the uplink data detection for reconfigurable intelligent surface (RIS)-assisted massive multiple-input multiple-output (MIMO) systems with low-resolution quantization at the base stations (BSs). By exploiting the statistical characteristics of the user-to-RIS and RIS-to-BS channels and the central limit theorem (CLT), we first demonstrate that the components in the effective channel are approximately Gaussian distributed under propagation in sub-6 GHz bands. Based on the bilinear generalized approximate message passing (BiG-AMP) algorithm, we then provide a quantization-aware approach to estimate the effective channel and recover transmit symbols. Finally, numerical results show that the revised BiG-AMP technique can improve the performance of channel estimation and data detection with affordable computational complexity.