Hai Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Hai Li has a total of 34 co-authors. Here are the top co-authors by number of shared publications:
- Shujuan Hou
Beijing Institute of Technology
12 shared publications - Zhengyu Song
School of Electronic and Information Engineering, Beijing Jiaotong University, Beijing, China
5 shared publications Qin Zhang
4 shared publicationsBowen Dou
3 shared publications- Hao-Sen Chen
Institute of Advanced Structure Technology, Beijing Institute of Technology, Beijing, CHINA
3 shared publications - Lei Sun
Beijing Institute of Technology
3 shared publications
Latest Publications
- 15 September 2026
- Published by Elsevier in Applied Energy
- Vol. 427
Cross-domain state-of-health estimation for lithium-ion batteries via meta-transfer learning
- 15 September 2026
- Published by Elsevier in Applied Energy
- Vol. 427
Cross-domain state-of-health estimation for lithium-ion batteries via meta-transfer learning
- 1 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Green Communications and Networking
- Vol. 10, 2579-2592
Discrete Pulse OTFS Waveform Scheme: Transceiver Design and Parameter Optimization
Orthogonal time frequency space (OTFS) is a novel modulation scheme that effectively mitigates multipath fading and Doppler shifts. Moreover, certain scenarios such as forward collision warning and emergency signal transmission in vehicle-to-everything (V2X) systems impose stringent requirements on low-latency communications under high-speed conditions. In this paper, we propose a discrete pulse orthogonal time frequency space (DP-OTFS) waveform scheme that enables low-latency communication in multipath high-speed scenarios by realizing logical full-duplex communication. Additionally, we model the transmit-receive collision channel and multi-node collision channel, and propose a signal detection method based on the collision mask matrix in full-duplex mode. Regarding the bit error rate (BER) performance, we innovatively define the collision impact factor (CIF) by leveraging OTFS characteristics to more accurately quantify the actual impact of collisions on system performance. Furthermore, we formulate and solve an optimization problem aimed at minimizing BER caused by inter-signal conflicts through time-hopping sequence (THS) adjustment. Simulation results demonstrate that the proposed DP-OTFS waveform exhibits both full-duplex capability and robustness against multipath Doppler interference. Compared with OTFS, DP-OTFS enables simultaneous signal transmission and reception, thus significantly improving spectral efficiency and reducing communication latency. In contrast, conventional OTFS experiences severe BER degradation under transmit-receive and multi-node collision scenarios, which renders it unable to maintain reliable communication.
- 1 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Green Communications and Networking
- Vol. 10, 2579-2592
Discrete Pulse OTFS Waveform Scheme: Transceiver Design and Parameter Optimization
Orthogonal time frequency space (OTFS) is a novel modulation scheme that effectively mitigates multipath fading and Doppler shifts. Moreover, certain scenarios such as forward collision warning and emergency signal transmission in vehicle-to-everything (V2X) systems impose stringent requirements on low-latency communications under high-speed conditions. In this paper, we propose a discrete pulse orthogonal time frequency space (DP-OTFS) waveform scheme that enables low-latency communication in multipath high-speed scenarios by realizing logical full-duplex communication. Additionally, we model the transmit-receive collision channel and multi-node collision channel, and propose a signal detection method based on the collision mask matrix in full-duplex mode. Regarding the bit error rate (BER) performance, we innovatively define the collision impact factor (CIF) by leveraging OTFS characteristics to more accurately quantify the actual impact of collisions on system performance. Furthermore, we formulate and solve an optimization problem aimed at minimizing BER caused by inter-signal conflicts through time-hopping sequence (THS) adjustment. Simulation results demonstrate that the proposed DP-OTFS waveform exhibits both full-duplex capability and robustness against multipath Doppler interference. Compared with OTFS, DP-OTFS enables simultaneous signal transmission and reception, thus significantly improving spectral efficiency and reducing communication latency. In contrast, conventional OTFS experiences severe BER degradation under transmit-receive and multi-node collision scenarios, which renders it unable to maintain reliable communication.
- Open Access
- 25 February 2026
- Published by MDPI AG in Electronics
- Vol. 15 (5) , 947
Lightweight Adaptive Reinforcement Learning-Based TCP Congestion Control for Multi-Hop Ad Hoc Networks
Ad hoc networks are characterized by flexible deployment and multi-hop communication, which has facilitated their growing prevalence in diverse applications. However, the TCP protocol exhibits substantial performance degradation in multi-hop ad hoc networks with dynamic topologies. To address this issue, this paper proposes TCP-RLA, a lightweight adaptive reinforcement learning-based TCP congestion control algorithm. It predicts network state variations and leverages a deep Q-network (DQN) with a rule-assisted discrete action space to adaptively tune the congestion window. This design boosts convergence speed and reduces computational complexity, making it well-suited for resource-constrained ad hoc nodes. Simulation results demonstrate that, compared with two reinforcement learning-based algorithms (GVegas and Orca), TCP-RLA achieves an average throughput improvement of 36.1% and 43.3%, an average round-trip time (RTT) reduction of 13.1% and 47.9%, and an average packet loss rate (PLR) reduction of 33.3% and 50%, respectively.
- 25 February 2026
- Published by MDPI AG in Electronics
- Vol. 15 (5) , 947
Open AccessLightweight Adaptive Reinforcement Learning-Based TCP Congestion Control for Multi-Hop Ad Hoc Networks
Ad hoc networks are characterized by flexible deployment and multi-hop communication, which has facilitated their growing prevalence in diverse applications. However, the TCP protocol exhibits substantial performance degradation in multi-hop ad hoc networks with dynamic topologies. To address this issue, this paper proposes TCP-RLA, a lightweight adaptive reinforcement learning-based TCP congestion control algorithm. It predicts network state variations and leverages a deep Q-network (DQN) with a rule-assisted discrete action space to adaptively tune the congestion window. This design boosts convergence speed and reduces computational complexity, making it well-suited for resource-constrained ad hoc nodes. Simulation results demonstrate that, compared with two reinforcement learning-based algorithms (GVegas and Orca), TCP-RLA achieves an average throughput improvement of 36.1% and 43.3%, an average round-trip time (RTT) reduction of 13.1% and 47.9%, and an average packet loss rate (PLR) reduction of 33.3% and 50%, respectively.
- 19 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (22) , 47323-47333
Lightweight Real-World Image Super Resolution via Channel Redundancy for Edge IoT Devices
In real world scenarios, low-resolution images often suffer from complex and unknown distortions during acquisition by edge Internet of Things (IoT) devices. To achieve a balance between super resolution and distortion suppression, real world super resolution (RWSR) models need to possess both global and local feature modeling capabilities. Moreover, a lightweight design is necessary for edge deployment without compromising performance. Therefore, this article proposes a lightweight channel group rearrange transformer (CGRFormer), which exploits channel redundancy to reduce computational burden while enhancing local feature modeling capabilities. CGRFormer comprises two key components: 1) the channel group and rearrange operation (CGRO) and 2) the multiscale local enhancement feed forward network (MSLE-FFN). To address the issue of a heavy computational burden, CGRO divides the feature maps into active and inert groups. Heavy operations are applied only to the active group, significantly reducing overall complexity. A subsequent channel rearrange operation enhances information interaction between the two groups. To enhance local feature modeling, MSLE-FFN incorporates a multiscale convolutional group (MSCG) and a channel excitation module (CEM), enabling effective extraction and enhancement of local features at different receptive fields. Overall, CGRFormer effectively balances super resolution and distortion suppression while maintaining a lightweight framework. Experimental results on the commonly used test datasets show that our model is still competitive with the SOTA model.
- 19 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (22) , 47323-47333
Lightweight Real-World Image Super Resolution via Channel Redundancy for Edge IoT Devices
In real world scenarios, low-resolution images often suffer from complex and unknown distortions during acquisition by edge Internet of Things (IoT) devices. To achieve a balance between super resolution and distortion suppression, real world super resolution (RWSR) models need to possess both global and local feature modeling capabilities. Moreover, a lightweight design is necessary for edge deployment without compromising performance. Therefore, this article proposes a lightweight channel group rearrange transformer (CGRFormer), which exploits channel redundancy to reduce computational burden while enhancing local feature modeling capabilities. CGRFormer comprises two key components: 1) the channel group and rearrange operation (CGRO) and 2) the multiscale local enhancement feed forward network (MSLE-FFN). To address the issue of a heavy computational burden, CGRO divides the feature maps into active and inert groups. Heavy operations are applied only to the active group, significantly reducing overall complexity. A subsequent channel rearrange operation enhances information interaction between the two groups. To enhance local feature modeling, MSLE-FFN incorporates a multiscale convolutional group (MSCG) and a channel excitation module (CEM), enabling effective extraction and enhancement of local features at different receptive fields. Overall, CGRFormer effectively balances super resolution and distortion suppression while maintaining a lightweight framework. Experimental results on the commonly used test datasets show that our model is still competitive with the SOTA model.
- 13 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-14
Joint Distortion Estimation and Removal Network for Versatile Hybrid-Distorted Image Restoration
Although image restoration methods have achieved great success on single distortion, their performance declines when faced with hybrid distortion effects in real-world scenarios. Recently, some methods for hybrid distortions have been explored. Suppose the total type of distortion is N, the types of distortion that the image suffers may change from 1 to N in real scenarios. However, existing methods fail to simultaneously address variations in distortions. Toward this end, a general network architecture named joint distortion estimation and removal network (JDERNet) is proposed for versatile hybrid-distorted image restoration. The distortion information of the hybrid-distorted image is mathematically represented as a distortion vector, and a neural network is constructed to estimate it. The distortion vector is utilized as a prior to modulate the image reconstruction process. Specifically, a novel feature modulation convolution block (FMCB) is designed. In the FMCB, the distortion vector is employed to modulate the feature maps generated by the attention branch (AB) and the non-AB (NAB), thereby enhancing the restoration of images affected by various types of distortions. Experimental results indicate that the proposed JDERNet achieves state-of-the-art performance in restoring images with varying types of distortions, ranging from multiple hybrid distortions to single distortions.
- 13 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-14
Joint Distortion Estimation and Removal Network for Versatile Hybrid-Distorted Image Restoration
Although image restoration methods have achieved great success on single distortion, their performance declines when faced with hybrid distortion effects in real-world scenarios. Recently, some methods for hybrid distortions have been explored. Suppose the total type of distortion is N, the types of distortion that the image suffers may change from 1 to N in real scenarios. However, existing methods fail to simultaneously address variations in distortions. Toward this end, a general network architecture named joint distortion estimation and removal network (JDERNet) is proposed for versatile hybrid-distorted image restoration. The distortion information of the hybrid-distorted image is mathematically represented as a distortion vector, and a neural network is constructed to estimate it. The distortion vector is utilized as a prior to modulate the image reconstruction process. Specifically, a novel feature modulation convolution block (FMCB) is designed. In the FMCB, the distortion vector is employed to modulate the feature maps generated by the attention branch (AB) and the non-AB (NAB), thereby enhancing the restoration of images affected by various types of distortions. Experimental results indicate that the proposed JDERNet achieves state-of-the-art performance in restoring images with varying types of distortions, ranging from multiple hybrid distortions to single distortions.
- 20 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 29 (1) , 140-144
Joint Location and Beamforming Design for Energy Efficient STAR-RIS-Aided ISAC Systems
In this letter, we investigate a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) aided integrated sensing and communication (ISAC) system. Our object is to maximize the energy efficiency (EE) of the ISAC system by jointly optimizing the beamforming at the base station (BS), the transmitting/reflecting coefficient matrices of the STAR-RIS, as well as the deployment location of the STAR-RIS. Since the formulated problem is non-convex, we propose an alternative optimization scheme by decomposing the original problem into three sub-problems, where the sub-problems are solved based on semidefinite relaxation (SDR) and fractional programming. Simulation results demonstrate that the STAR-RIS aided ISAC system achieves a higher EE than conventional RIS scheme, and the location optimization for STAR-RIS can significantly improve the system EE. Furthermore, the EE declines rapidly when the radar beampattern gain threshold increases to a larger value.
- 20 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 29 (1) , 140-144
Joint Location and Beamforming Design for Energy Efficient STAR-RIS-Aided ISAC Systems
In this letter, we investigate a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) aided integrated sensing and communication (ISAC) system. Our object is to maximize the energy efficiency (EE) of the ISAC system by jointly optimizing the beamforming at the base station (BS), the transmitting/reflecting coefficient matrices of the STAR-RIS, as well as the deployment location of the STAR-RIS. Since the formulated problem is non-convex, we propose an alternative optimization scheme by decomposing the original problem into three sub-problems, where the sub-problems are solved based on semidefinite relaxation (SDR) and fractional programming. Simulation results demonstrate that the STAR-RIS aided ISAC system achieves a higher EE than conventional RIS scheme, and the location optimization for STAR-RIS can significantly improve the system EE. Furthermore, the EE declines rapidly when the radar beampattern gain threshold increases to a larger value.
- 30 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 73 (5) , 6383-6394
Short Term Charging Data Based Battery State of Health and State of Charge Estimation Using Feature Pyramid
Accurate battery states estimation is critical to the safe and stable operation of Li-ion batteries, and it is one of the fundamental functions of a battery management system (BMS). This paper proposes a novel deep learning framework called multi-timescale dual feature-based state estimation network (MFN) using extremely limited charging data to estimate three critical states simultaneously end-to-end: maximum capacity, the capacity at the beginning of charging, and the capacity at the end of charging. The core of the framework is the dual feature extraction module (DFM) and the multi-timescale information extraction module (MTM). First, DFM extracts independent and coupling features of external signals (voltage, current, temperature) in the time dimension, respectively, and then merges them. Based on this, MTM explores the health features of the external signals on different time scales and merges them. Finally, the three internal states are estimated using the mined depth features. Experimental results in the Oxford battery data show that the root mean square error (RMSE) for three internal states is less than 9.71 mAh, corresponding to 1.31% of the nominal capacity.
- 30 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 73 (5) , 6383-6394
Short Term Charging Data Based Battery State of Health and State of Charge Estimation Using Feature Pyramid
Accurate battery states estimation is critical to the safe and stable operation of Li-ion batteries, and it is one of the fundamental functions of a battery management system (BMS). This paper proposes a novel deep learning framework called multi-timescale dual feature-based state estimation network (MFN) using extremely limited charging data to estimate three critical states simultaneously end-to-end: maximum capacity, the capacity at the beginning of charging, and the capacity at the end of charging. The core of the framework is the dual feature extraction module (DFM) and the multi-timescale information extraction module (MTM). First, DFM extracts independent and coupling features of external signals (voltage, current, temperature) in the time dimension, respectively, and then merges them. Based on this, MTM explores the health features of the external signals on different time scales and merges them. Finally, the three internal states are estimated using the mined depth features. Experimental results in the Oxford battery data show that the root mean square error (RMSE) for three internal states is less than 9.71 mAh, corresponding to 1.31% of the nominal capacity.