Jeng-Shin Sheu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Jeng-Shin Sheu has a total of 21 co-authors. Here are the top co-authors by number of shared publications:
Hong‐Rui Chang
Department of Computer Science and Information Engineering National Yunlin University of Science and Technology 123, University Road Section 3 Douliou Yunlin Taiwan
2 shared publicationsJenn‐Kaie Lain
Department of Electronic Engineering National Yunlin University of Science & Technology Douliou Yunlin Taiwan
2 shared publicationsWern‐Ho Sheen
Department of Communications Engineering National Chung Cheng University 168, University Road Section 1, Minhsiung Chiayi Taiwan
2 shared publicationsBing‐Jia Li
Department of Computer Science and Information Engineering National Yunlin University of Science & Technology Douliou Yunlin Taiwan
1 shared publicationChao‐Kai Wu
Department of Communications Engineering National Chung Cheng University 168, University Road Section 1, Minhsiung Chiayi Taiwan
1 shared publicationCheng-Hsuan Lu
1 shared publication
Latest Publications
- Open Access
- 1 January 2025
- Published by Institution of Engineering and Technology (IET) in IET Communications
- Vol. 19 (1)
Optimising Energy Harvesting and Throughput for UAV‐Assisted IRS Systems With Adaptive Energy Harvesting
Integrating intelligent reflecting surfaces (IRS) with unmanned aerial vehicles (UAVs) presents a promising approach for future energy‐efficient wireless communications. This paper proposes an adaptive framework that dynamically balances energy harvesting (EH) efficiency and system throughput by adjusting the required EH efficiency based on the UAV's power levels and communication needs. Utilising real‐coded genetic algorithm (RCGA), the framework effectively tackles challenges posed by multi‐user interference (MUI) and imperfect channel estimation (CE). Our results demonstrate that the RCGA‐based approach outperforms deep reinforcement learning (DRL) methods, delivering superior energy harvesting and throughput in realistic conditions. The adaptive EH strategy not only optimises throughput performance but also ensures efficient UAV energy management, particularly in dynamic and energy‐constrained environments, making it a robust solution for sustained UAV operations in dynamic and energy‐constrained environments.
- 1 January 2025
- Published by Institution of Engineering and Technology (IET) in IET Communications
- Vol. 19 (1)
Open AccessOptimising Energy Harvesting and Throughput for UAV‐Assisted IRS Systems With Adaptive Energy Harvesting
Integrating intelligent reflecting surfaces (IRS) with unmanned aerial vehicles (UAVs) presents a promising approach for future energy‐efficient wireless communications. This paper proposes an adaptive framework that dynamically balances energy harvesting (EH) efficiency and system throughput by adjusting the required EH efficiency based on the UAV's power levels and communication needs. Utilising real‐coded genetic algorithm (RCGA), the framework effectively tackles challenges posed by multi‐user interference (MUI) and imperfect channel estimation (CE). Our results demonstrate that the RCGA‐based approach outperforms deep reinforcement learning (DRL) methods, delivering superior energy harvesting and throughput in realistic conditions. The adaptive EH strategy not only optimises throughput performance but also ensures efficient UAV energy management, particularly in dynamic and energy‐constrained environments, making it a robust solution for sustained UAV operations in dynamic and energy‐constrained environments.
- Open Access
- 18 August 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 88854-88868
Joint Beamforming, Power Control, and Interference Coordination: A Reinforcement Learning Approach Replacing Rewards With Examples
In this paper, we consider the problem of multi-cell interference coordination by joint beamforming and power control. Recent efforts have explored the use of reinforcement learning (RL) methods to tackle this complex optimization problem. Typically, a decentralized multi-agent framework is adopted, wherein each base station operates as an independent RL agent. This distributed coordination has gained attention because designing a reward function that effectively captures the condition of the entire cellular network is challenging for single-agent RL models. However, the distributed approach introduces unique challenges, particularly the non-stationarity of the multi-agent environment, as agents continually adapt their policies to interact with one another. The non-stationary environment necessitates information exchange among agents, as local observations of each agent are insufficient to fully capture the true state of the environment. Unfortunately, this information exchange incurs a significant overhead, thereby limiting data transmission capabilities. To address these challenges, we propose a novel single-agent RL approach that eliminates the need for information exchange and the conventional reward function. Instead, we leverage success examples to guide the learning process. Simulation results show that the proposed approach outperforms the existing multi-agent method and theoretical algorithm in terms of sum rates. Additionally, our approach ensures a uniform quality of service while maximizing the overall sum rate.
- 18 August 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 88854-88868
Open AccessJoint Beamforming, Power Control, and Interference Coordination: A Reinforcement Learning Approach Replacing Rewards With Examples
In this paper, we consider the problem of multi-cell interference coordination by joint beamforming and power control. Recent efforts have explored the use of reinforcement learning (RL) methods to tackle this complex optimization problem. Typically, a decentralized multi-agent framework is adopted, wherein each base station operates as an independent RL agent. This distributed coordination has gained attention because designing a reward function that effectively captures the condition of the entire cellular network is challenging for single-agent RL models. However, the distributed approach introduces unique challenges, particularly the non-stationarity of the multi-agent environment, as agents continually adapt their policies to interact with one another. The non-stationary environment necessitates information exchange among agents, as local observations of each agent are insufficient to fully capture the true state of the environment. Unfortunately, this information exchange incurs a significant overhead, thereby limiting data transmission capabilities. To address these challenges, we propose a novel single-agent RL approach that eliminates the need for information exchange and the conventional reward function. Instead, we leverage success examples to guide the learning process. Simulation results show that the proposed approach outperforms the existing multi-agent method and theoretical algorithm in terms of sum rates. Additionally, our approach ensures a uniform quality of service while maximizing the overall sum rate.
- Open Access
- 24 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 19849-19862
Performance Improvement on Traditional Chinese Task-Oriented Dialogue Systems With Reinforcement Learning and Regularized Dropout Technique
The development of conversational voice assistant applications has been in full swing around the world. This paper aims to develop traditional Chinese multi-domain task-oriented dialogue (TOD) systems. It is typically implemented using pipeline approach, where submodules are optimized independently, resulting in inconsistencies with each other. Instead, this paper implements end-to-end multi-domain TOD models using pre-trained deep neural networks (DNNs). This allows us to integrate all the submodules into one single DNN model to solve the inconsistencies. Data shortages are common in conversational natural language processing (NLP) tasks using DNN models. In this regard, dropout regularization has been widely used to improve overfitting caused by insufficient training dataset. However, the randomness it introduces leads to non-negligible discrepancies between training and inference. On the other hand, pre-trained language models have successfully provided effective regularization for NLP tasks. An inherent disadvantage is that fine-tuning the pre-trained language model suffers from exposure bias and loss-evaluation mismatch. To this end, we propose a reinforcement learning (RL) approach to address both issues. Furthermore, we adopt a method called regularized dropout (R-Drop) to improve the inconsistency in dropout layers of DNNs. Experimental results show that both our proposed RL approach and the R-Drop technique can significantly improve the joint target accuracy (JGA) score and combined score of traditional Chinese TOD system in tasks of dialogue state tracking (DST) and end-to-end sentence prediction, respectively.
- 24 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 19849-19862
Open AccessPerformance Improvement on Traditional Chinese Task-Oriented Dialogue Systems With Reinforcement Learning and Regularized Dropout Technique
The development of conversational voice assistant applications has been in full swing around the world. This paper aims to develop traditional Chinese multi-domain task-oriented dialogue (TOD) systems. It is typically implemented using pipeline approach, where submodules are optimized independently, resulting in inconsistencies with each other. Instead, this paper implements end-to-end multi-domain TOD models using pre-trained deep neural networks (DNNs). This allows us to integrate all the submodules into one single DNN model to solve the inconsistencies. Data shortages are common in conversational natural language processing (NLP) tasks using DNN models. In this regard, dropout regularization has been widely used to improve overfitting caused by insufficient training dataset. However, the randomness it introduces leads to non-negligible discrepancies between training and inference. On the other hand, pre-trained language models have successfully provided effective regularization for NLP tasks. An inherent disadvantage is that fine-tuning the pre-trained language model suffers from exposure bias and loss-evaluation mismatch. To this end, we propose a reinforcement learning (RL) approach to address both issues. Furthermore, we adopt a method called regularized dropout (R-Drop) to improve the inconsistency in dropout layers of DNNs. Experimental results show that both our proposed RL approach and the R-Drop technique can significantly improve the joint target accuracy (JGA) score and combined score of traditional Chinese TOD system in tasks of dialogue state tracking (DST) and end-to-end sentence prediction, respectively.
- Open Access
- 2 November 2021
- Published by MDPI AG in Applied Sciences
- Vol. 11 (21) , 10298
Performance Comparison for Single-User and Multi-User Network MIMO Cellular Systems with Power Management
Cellular mobile systems aim at aggressive spectrum reuse to achieve high spectral efficiency. Unfortunately, this leads to unacceptable interference near cell borders. To control this, network multi-input multiple-output (MIMO) can be adopted to improve coverage and cell-edge throughput through multi-cell cooperation. With network MIMO, multiple geographically separated base stations (BSs) cooperatively serve their cell-edge users (CEUs) using their antennas, acting together as a network of distributed antenna array. It can be single-user (SU) or multi-user (MU) network MIMO by coordinating channel allocation in adjacent cells. In this paper, we make a capacity comparison of SU- and MU-network MIMO. In network MIMO, a collaborative BS simultaneously serves its own cell-center users (CCUs) and CEUs, and the CEUs of other partner BSs under a power constraint. As a result, power management among three types of users (intra-cell CCUs/CEUs, inter-cell CEUs) becomes necessary. Accordingly, we propose power management methods to help raise the signal strength of inter-cell CEUs and in the meantime gratify the performance of intra-cell users. Simulation results show that MU-network MIMO with superposition coding offers much better CEU capacity than SU-network MIMO. As for the CCU capacity, MU-network MIMO is generally better than SU-network MIMO.
- 2 November 2021
- Published by MDPI AG in Applied Sciences
- Vol. 11 (21) , 10298
Open AccessPerformance Comparison for Single-User and Multi-User Network MIMO Cellular Systems with Power Management
Cellular mobile systems aim at aggressive spectrum reuse to achieve high spectral efficiency. Unfortunately, this leads to unacceptable interference near cell borders. To control this, network multi-input multiple-output (MIMO) can be adopted to improve coverage and cell-edge throughput through multi-cell cooperation. With network MIMO, multiple geographically separated base stations (BSs) cooperatively serve their cell-edge users (CEUs) using their antennas, acting together as a network of distributed antenna array. It can be single-user (SU) or multi-user (MU) network MIMO by coordinating channel allocation in adjacent cells. In this paper, we make a capacity comparison of SU- and MU-network MIMO. In network MIMO, a collaborative BS simultaneously serves its own cell-center users (CCUs) and CEUs, and the CEUs of other partner BSs under a power constraint. As a result, power management among three types of users (intra-cell CCUs/CEUs, inter-cell CEUs) becomes necessary. Accordingly, we propose power management methods to help raise the signal strength of inter-cell CEUs and in the meantime gratify the performance of intra-cell users. Simulation results show that MU-network MIMO with superposition coding offers much better CEU capacity than SU-network MIMO. As for the CCU capacity, MU-network MIMO is generally better than SU-network MIMO.
- Open Access
- 1 October 2019
- Published by Institution of Engineering and Technology (IET) in IET Communications
- Vol. 13 (17) , 2629-2638
Beam‐based uplink multi‐user detection for mmWave communications
To keep up with the dramatic proliferation of mobile traffic, the fifth‐generation (5G) systems adopt the millimetre‐wave (mmWave) frequency band together with a large‐scale antenna array. Although the use of large‐scale antenna array can effectively solve the problem of severe signal attenuation at high frequencies, it also derives other issues. For example, the system complexity on spatial domain becomes computationally prohibitive, because the system has to estimate all the channels of each antenna. Also, the ever‐growing data rates make the frequency selectivity of channels worst in the spatial domain. Recently, utilising the sparse characteristics of the mmWave band, beam‐domain signal processing has drawn increasing interest for future 5G systems. In this study, the authors design multi‐user data detection for the mmWave communication on beam domain. To effectively reduce the design complexity, they exploit the salient feature of channel sparsity to propose beam‐channel selection and user grouping. Results indicate that the beam‐channel selection is a critical step in the design of 5G systems since it can significantly decrease the complexity while maintaining the throughput. User grouping allows us to independently design detectors with lower complexity. Results show that the designed detectors can achieve a good trade‐off between throughput and complexity.
- 1 October 2019
- Published by Institution of Engineering and Technology (IET) in IET Communications
- Vol. 13 (17) , 2629-2638
Open AccessBeam‐based uplink multi‐user detection for mmWave communications
To keep up with the dramatic proliferation of mobile traffic, the fifth‐generation (5G) systems adopt the millimetre‐wave (mmWave) frequency band together with a large‐scale antenna array. Although the use of large‐scale antenna array can effectively solve the problem of severe signal attenuation at high frequencies, it also derives other issues. For example, the system complexity on spatial domain becomes computationally prohibitive, because the system has to estimate all the channels of each antenna. Also, the ever‐growing data rates make the frequency selectivity of channels worst in the spatial domain. Recently, utilising the sparse characteristics of the mmWave band, beam‐domain signal processing has drawn increasing interest for future 5G systems. In this study, the authors design multi‐user data detection for the mmWave communication on beam domain. To effectively reduce the design complexity, they exploit the salient feature of channel sparsity to propose beam‐channel selection and user grouping. Results indicate that the beam‐channel selection is a critical step in the design of 5G systems since it can significantly decrease the complexity while maintaining the throughput. User grouping allows us to independently design detectors with lower complexity. Results show that the designed detectors can achieve a good trade‐off between throughput and complexity.
- Open Access
- 22 October 2018
- Published by Institution of Engineering and Technology (IET) in IET Communications
- Vol. 12 (19) , 2493-2501
Pilot decontamination techniques based on beam‐domain channel characteristics in millimetre‐wave sparse channels
The upcoming 5G system is expected to utilise large‐scale antenna arrays to increase the users’ data rate and system capacity. With a large number of antennas, channel estimation becomes a very critical design issue due to the adverse effects of pilot contamination. Some problems in the existing channel estimation methods studied in the literature include: (i) the complexity of signal processing on the spatial domain is very high. (ii) Channels are quite frequency selective in the spatial domain. (iii) It is not easy to obtain the second‐order statistics of channels. In this study, the channel estimation issue is investigated using the signal processing in the beam domain. It is due to the salient channel sparsity at millimetre‐wave frequencies that only a few beam channels need to be estimated. The authors exploit these beam channels to propose some pilot decontamination techniques: basic pilot allocation, beam‐set non‐overlapping allocation, intra‐cell interference mitigation allocation and maximum beam power based allocation (MBPBA). Both large pilot reuse factor and beam‐domain power control are also presented to supplement the proposed techniques. Simulation results show that the best method is MBPBA in terms of overhead and normalised square error.
- 22 October 2018
- Published by Institution of Engineering and Technology (IET) in IET Communications
- Vol. 12 (19) , 2493-2501
Open AccessPilot decontamination techniques based on beam‐domain channel characteristics in millimetre‐wave sparse channels
The upcoming 5G system is expected to utilise large‐scale antenna arrays to increase the users’ data rate and system capacity. With a large number of antennas, channel estimation becomes a very critical design issue due to the adverse effects of pilot contamination. Some problems in the existing channel estimation methods studied in the literature include: (i) the complexity of signal processing on the spatial domain is very high. (ii) Channels are quite frequency selective in the spatial domain. (iii) It is not easy to obtain the second‐order statistics of channels. In this study, the channel estimation issue is investigated using the signal processing in the beam domain. It is due to the salient channel sparsity at millimetre‐wave frequencies that only a few beam channels need to be estimated. The authors exploit these beam channels to propose some pilot decontamination techniques: basic pilot allocation, beam‐set non‐overlapping allocation, intra‐cell interference mitigation allocation and maximum beam power based allocation (MBPBA). Both large pilot reuse factor and beam‐domain power control are also presented to supplement the proposed techniques. Simulation results show that the best method is MBPBA in terms of overhead and normalised square error.
- 23 April 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 67 (8) , 7191-7203
On the Design of Downlink Multiuser Transmission for a Beam-Group Division 5G System
To accommodate the ever-increasing data traffic, the 5G system has adopted the enormous swathes of unlicensed spectrum at millimeter-wave band. The millimeter-wave band however exhibits orders-of-magnitude more path loss than the microwave band. The upcoming 5G system thus necessitates the use of large-scale antenna arrays to compensate for the notable path loss with beamforming gain. Nevertheless, the large number of antenna elements also poses some challenges. For example, it is impractical to assign one dedicated transceiver unit (TXRU) for each individual element and the complexity of channel estimation on spatial (antenna) domain is prohibitively high. In this paper, we design low complexity multi-user transmission on beam domain at millimeter-wave band. Specifically, we propose the beam channel set selection and user grouping schemes to significantly lower the complexity of channel estimation and precoder design by exploiting the salient feature of channel sparsity. Moreover, 2D TXRU virtualization together with beam sweeping strategies is presented as the TXRU number is less than the number of columns of planar antenna array. Simulation results show that the proposed beam-group division 5G system can significantly reduce the complexity and keep the system throughput.
- 23 April 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 67 (8) , 7191-7203
On the Design of Downlink Multiuser Transmission for a Beam-Group Division 5G System
To accommodate the ever-increasing data traffic, the 5G system has adopted the enormous swathes of unlicensed spectrum at millimeter-wave band. The millimeter-wave band however exhibits orders-of-magnitude more path loss than the microwave band. The upcoming 5G system thus necessitates the use of large-scale antenna arrays to compensate for the notable path loss with beamforming gain. Nevertheless, the large number of antenna elements also poses some challenges. For example, it is impractical to assign one dedicated transceiver unit (TXRU) for each individual element and the complexity of channel estimation on spatial (antenna) domain is prohibitively high. In this paper, we design low complexity multi-user transmission on beam domain at millimeter-wave band. Specifically, we propose the beam channel set selection and user grouping schemes to significantly lower the complexity of channel estimation and precoder design by exploiting the salient feature of channel sparsity. Moreover, 2D TXRU virtualization together with beam sweeping strategies is presented as the TXRU number is less than the number of columns of planar antenna array. Simulation results show that the proposed beam-group division 5G system can significantly reduce the complexity and keep the system throughput.
- Open Access
- 1 December 2017
- Published by Institution of Engineering and Technology (IET) in IET Optoelectronics
- Vol. 11 (6) , 259-264
LED non‐linearity mitigation techniques for optical OFDM‐based visible light communications
The visible light communication (VLC) utilising the light emitting diodes (LEDs) as transmitters has gained much interest for the indoor wireless access due to its abundance of unregulated spectrum. However, the LED non‐linearities cause a detrimental non‐linear distortion especially for the adoption of optical orthogonal frequency‐division multiplexing (O‐OFDM), which is known to suffer high peak‐to‐average power ratio (PAPR). Specifically, the authors consider the direct current biased O‐OFDM (DCO‐OFDM) to realise the O‐OFDM signals. This study aims to develop the mitigation techniques for LED non‐linearities at transmitter through the combined advantages of PAPR reduction and the technique of linearising the non‐linear LED characteristics. The orthogonal cover code sequence is exploited to design a transform precoding to reduce the PAPR. An uncomplicated pre‐distorter at transmitter is adopted to linearise the LED non‐linearity. In addition, the authors propose a dynamic DC biasing method to avoid the clipping noise in DCO‐OFDM. As a whole, the proposed techniques at VLC transmitter can improve the LED non‐linearities even for large input signals.
- 1 December 2017
- Published by Institution of Engineering and Technology (IET) in IET Optoelectronics
- Vol. 11 (6) , 259-264
Open AccessLED non‐linearity mitigation techniques for optical OFDM‐based visible light communications
The visible light communication (VLC) utilising the light emitting diodes (LEDs) as transmitters has gained much interest for the indoor wireless access due to its abundance of unregulated spectrum. However, the LED non‐linearities cause a detrimental non‐linear distortion especially for the adoption of optical orthogonal frequency‐division multiplexing (O‐OFDM), which is known to suffer high peak‐to‐average power ratio (PAPR). Specifically, the authors consider the direct current biased O‐OFDM (DCO‐OFDM) to realise the O‐OFDM signals. This study aims to develop the mitigation techniques for LED non‐linearities at transmitter through the combined advantages of PAPR reduction and the technique of linearising the non‐linear LED characteristics. The orthogonal cover code sequence is exploited to design a transform precoding to reduce the PAPR. An uncomplicated pre‐distorter at transmitter is adopted to linearise the LED non‐linearity. In addition, the authors propose a dynamic DC biasing method to avoid the clipping noise in DCO‐OFDM. As a whole, the proposed techniques at VLC transmitter can improve the LED non‐linearities even for large input signals.
- 23 February 2015
- Published by Taylor & Francis in Journal of the Chinese Institute of Engineers
- Vol. 38 (5) , 583-593
On the joint preprocessing techniques for OFDM-based cooperative network MIMO systems
Next-generation (4G) systems are designed to support universal frequency reuse (UFR) and adopt orthogonal frequency-division multiplexing (OFDM) to achieve best use of their valuable spectra. Unfortunately, the implementation of UFR leads to unacceptable interference levels near cell borders. To control this, 4G systems adopt network multiple-input multiple-output (MIMO), an elaborate coordination among adjacent cells, to improve coverage and cell-edge throughput. Unfortunately, for multi-cell coordination, most studies on this subject have examined the corresponding information-theoretic capacities and only a few joint preprocessing schemes have been investigated for narrowband non-OFDM-based systems in flat fading channels. Instead of information-theoretic capacity, this paper derives various joint preprocessing techniques for OFDM-based network MIMO in wideband multipath fading channels. They include joint time-division multiple access (J-TDMA), joint zero-forcing (J-ZF), joint minimum mean square error (J-MMSE), and joint null-space decomposition (J-NSD). The J-TDMA scheme is free from co-channel interference (CCI) at the cost of spectral efficiency. The J-ZF scheme enables complete suppression of interference, but demands higher transmitter power for the precoding process. The J-MMSE yields a performance close to the CCI-free J-TDMA scheme, by making an acceptable compromise between interference suppression and transmitter power efficiency. In J-NSD, the CCI is completely removed, but signal transmission is confined to a reduced subspace.
- 23 February 2015
- Published by Taylor & Francis in Journal of the Chinese Institute of Engineers
- Vol. 38 (5) , 583-593
On the joint preprocessing techniques for OFDM-based cooperative network MIMO systems
Next-generation (4G) systems are designed to support universal frequency reuse (UFR) and adopt orthogonal frequency-division multiplexing (OFDM) to achieve best use of their valuable spectra. Unfortunately, the implementation of UFR leads to unacceptable interference levels near cell borders. To control this, 4G systems adopt network multiple-input multiple-output (MIMO), an elaborate coordination among adjacent cells, to improve coverage and cell-edge throughput. Unfortunately, for multi-cell coordination, most studies on this subject have examined the corresponding information-theoretic capacities and only a few joint preprocessing schemes have been investigated for narrowband non-OFDM-based systems in flat fading channels. Instead of information-theoretic capacity, this paper derives various joint preprocessing techniques for OFDM-based network MIMO in wideband multipath fading channels. They include joint time-division multiple access (J-TDMA), joint zero-forcing (J-ZF), joint minimum mean square error (J-MMSE), and joint null-space decomposition (J-NSD). The J-TDMA scheme is free from co-channel interference (CCI) at the cost of spectral efficiency. The J-ZF scheme enables complete suppression of interference, but demands higher transmitter power for the precoding process. The J-MMSE yields a performance close to the CCI-free J-TDMA scheme, by making an acceptable compromise between interference suppression and transmitter power efficiency. In J-NSD, the CCI is completely removed, but signal transmission is confined to a reduced subspace.