Changwei Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Changwei Zhang has a total of 57 co-authors. Here are the top co-authors by number of shared publications:
- Hongbo Zhu
Nanjing University of Posts and Telecommunications
7 shared publications - Ruochen Huang
School of Biomedical Engineering and Informatics, Nanjing Medical University, Nanjing, China
6 shared publications - Cheng Zhang
Pervasive Communications Center, Purple Mountain Laboratories, Nanjing, China
5 shared publications - Xinghua Sun
School of Electronics and Communication Engineering, Sun Yat-sen University, Guangzhou, China
5 shared publications - Yongming Huang
National Mobile Communications Research Laboratory, Southeast University, Nanjing, China
5 shared publications - Yifan Xu
Nanjing Medical University, Nanjing, Jiangsu, China
4 shared publications
Latest Publications
- Open Access
- 3 November 2026
- Published by MDPI AG in Electronics
- Vol. 15 (21) , 5106
Joint Jamming Avoidance and Suppression with Intelligent Detection Threshold Adaptation for Emergency Wireless Communications
As wireless communication technologies continue to evolve, malicious jamming becomes a major factor that limits system performance and communication quality, especially in emergency communication scenarios where reliable links are essential for timely information delivery and rescue coordination. Conventional anti-jamming transmission methods often suffer from two issues. In the detection stage, the jamming threshold is difficult to determine. In the data transmission stage, spectrum resources are not efficiently utilized. To address these limitations, this paper jointly considers jamming detection and data transmission, and proposes a joint jamming avoidance and suppression method based on intelligent detection threshold adaptation. Specifically, reinforcement learning is first used to iteratively optimize the detection threshold according to transmission outcomes, which improves the accuracy and robustness of jamming detection. Then, based on the detection results, the jamming status is dynamically evaluated, then jamming avoidance and jamming suppression are jointly employed during data transmission, which effectively improves spectrum utilization. Simulation results show that, under both continuous-band jamming and comb-spectrum jamming, the proposed scheme nearly doubles the system sum rate and also achieves significantly higher detection accuracy than conventional methods, which demonstrates strong robustness and adaptability for emergency communication systems.
- 3 November 2026
- Published by MDPI AG in Electronics
- Vol. 15 (21) , 5106
Open AccessJoint Jamming Avoidance and Suppression with Intelligent Detection Threshold Adaptation for Emergency Wireless Communications
As wireless communication technologies continue to evolve, malicious jamming becomes a major factor that limits system performance and communication quality, especially in emergency communication scenarios where reliable links are essential for timely information delivery and rescue coordination. Conventional anti-jamming transmission methods often suffer from two issues. In the detection stage, the jamming threshold is difficult to determine. In the data transmission stage, spectrum resources are not efficiently utilized. To address these limitations, this paper jointly considers jamming detection and data transmission, and proposes a joint jamming avoidance and suppression method based on intelligent detection threshold adaptation. Specifically, reinforcement learning is first used to iteratively optimize the detection threshold according to transmission outcomes, which improves the accuracy and robustness of jamming detection. Then, based on the detection results, the jamming status is dynamically evaluated, then jamming avoidance and jamming suppression are jointly employed during data transmission, which effectively improves spectrum utilization. Simulation results show that, under both continuous-band jamming and comb-spectrum jamming, the proposed scheme nearly doubles the system sum rate and also achieves significantly higher detection accuracy than conventional methods, which demonstrates strong robustness and adaptability for emergency communication systems.
- Open Access
- 10 July 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e85290
Large Language Model–Assisted Annotation Framework for Cross-Platform Analysis of Online Autism Communities: Implications for Parent Education and Digital Support
Background: Online health communities (OHCs) are important channels for families of children with autism spectrum disorder to obtain health information and psychosocial support. Differences between an open forum platform and the physician-patient consultation platforms may shape caregiver decisions, yet comparative evidence from China remains limited. A large language model (LLM) provides a scalable approach for systematic content annotation in large OHC datasets. Objective: This study proposes and validates a standardized LLM-assisted annotation framework under a unified classification schema and compares topic distributions and poster identities across an open forum platform (Baidu Tieba) and physician-patient consultation platforms (Chunyu Doctor and Haodf). Methods: We implemented an LLM-assisted annotation framework. A unified taxonomy of topics and poster identities was first developed through human open coding. Poster identities in Baidu Tieba were annotated through a double-blind manual procedure. For topic classification, interannotator and human-LLM agreement were evaluated on a manually labeled subset to benchmark models of varying sizes. The best-performing LLM was selected for full-dataset topic annotation, followed by statistical and cross-platform analysis. Results: When metrics were arithmetically averaged across all annotation tasks, the best-performing LLM achieved agreement levels comparable to human annotation (accuracy=79.18%, SD 0.20%; κ=0.736, SD 0.003; F1-score=0.727, SD 0.006), approaching interannotator agreement (accuracy=81.65%; κ=0.767; F1-score=0.758), demonstrating strong stability and scalability. Full-dataset analysis yielded 3 main findings. First, model performance increased with parameter scale but plateaued beyond 14B, indicating diminishing marginal returns from further scaling. Second, clear cross-platform differences in poster identity were observed: the open forum platform was dominated by family members of patients (caregivers: 2377/3516, 67.61%), with substantial participation from commercial rehabilitation practitioners (commercial posters: 427/3516, 12.14%), resulting in a more heterogeneous participation structure. Third, topic distributions reflected both shared high demand for resource-related information and differentiated help-seeking pathways: both platform types demonstrated consistently high demand for resource recommendation and evaluation; the open forum platform was primarily characterized by diagnosis-related discussions (1183/7535, 15.70%), whereas the physician-patient consultation platforms were centered on intervention-related consultation (2864/7687, 37.26%). Conclusions: The LLM-assisted annotation framework proposed in this study enables reliable large-scale annotation of OHC data while maintaining high human-LLM agreement and operational stability. Midsized models (eg, 14B) demonstrated favorable cost-performance efficiency. The findings reveal 2 key aspects: the open forum platform exhibits a complex participation structure, and the influence of commercially affiliated actors should not be overlooked; users on both platform types show sustained demand for resource-related information but follow different help-seeking pathways, emphasizing diagnostic exploration and professional intervention, respectively. These results suggest that platform structure and governance mechanisms may shape caregivers’ information access and decision-making. The framework provides a transparent, reproducible, and cost-effective approach for OHC research. All data were deidentified and handled in accordance with relevant platform policies and ethical standards.
- 10 July 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e85290
Open AccessLarge Language Model–Assisted Annotation Framework for Cross-Platform Analysis of Online Autism Communities: Implications for Parent Education and Digital Support
Background: Online health communities (OHCs) are important channels for families of children with autism spectrum disorder to obtain health information and psychosocial support. Differences between an open forum platform and the physician-patient consultation platforms may shape caregiver decisions, yet comparative evidence from China remains limited. A large language model (LLM) provides a scalable approach for systematic content annotation in large OHC datasets. Objective: This study proposes and validates a standardized LLM-assisted annotation framework under a unified classification schema and compares topic distributions and poster identities across an open forum platform (Baidu Tieba) and physician-patient consultation platforms (Chunyu Doctor and Haodf). Methods: We implemented an LLM-assisted annotation framework. A unified taxonomy of topics and poster identities was first developed through human open coding. Poster identities in Baidu Tieba were annotated through a double-blind manual procedure. For topic classification, interannotator and human-LLM agreement were evaluated on a manually labeled subset to benchmark models of varying sizes. The best-performing LLM was selected for full-dataset topic annotation, followed by statistical and cross-platform analysis. Results: When metrics were arithmetically averaged across all annotation tasks, the best-performing LLM achieved agreement levels comparable to human annotation (accuracy=79.18%, SD 0.20%; κ=0.736, SD 0.003; F1-score=0.727, SD 0.006), approaching interannotator agreement (accuracy=81.65%; κ=0.767; F1-score=0.758), demonstrating strong stability and scalability. Full-dataset analysis yielded 3 main findings. First, model performance increased with parameter scale but plateaued beyond 14B, indicating diminishing marginal returns from further scaling. Second, clear cross-platform differences in poster identity were observed: the open forum platform was dominated by family members of patients (caregivers: 2377/3516, 67.61%), with substantial participation from commercial rehabilitation practitioners (commercial posters: 427/3516, 12.14%), resulting in a more heterogeneous participation structure. Third, topic distributions reflected both shared high demand for resource-related information and differentiated help-seeking pathways: both platform types demonstrated consistently high demand for resource recommendation and evaluation; the open forum platform was primarily characterized by diagnosis-related discussions (1183/7535, 15.70%), whereas the physician-patient consultation platforms were centered on intervention-related consultation (2864/7687, 37.26%). Conclusions: The LLM-assisted annotation framework proposed in this study enables reliable large-scale annotation of OHC data while maintaining high human-LLM agreement and operational stability. Midsized models (eg, 14B) demonstrated favorable cost-performance efficiency. The findings reveal 2 key aspects: the open forum platform exhibits a complex participation structure, and the influence of commercially affiliated actors should not be overlooked; users on both platform types show sustained demand for resource-related information but follow different help-seeking pathways, emphasizing diagnostic exploration and professional intervention, respectively. These results suggest that platform structure and governance mechanisms may shape caregivers’ information access and decision-making. The framework provides a transparent, reproducible, and cost-effective approach for OHC research. All data were deidentified and handled in accordance with relevant platform policies and ethical standards.
- 12 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Wireless Communications Letters
- Vol. 15 (21622337) , 3304-3308
Dual-Timescale Resource Allocation for eMBB and URLLC Slices in Cell-Free RAN: A Hybrid SNC-DDQN Framework
In cell-free radio access networks (RANs), the co-existence of enhanced mobile broadband (eMBB) and ultra-reliable low-latency communication (URLLC) slices poses significant challenges for resource allocation under heterogeneous quality-of-service (QoS) constraints. To overcome the low spectral efficiency of static slicing and the high complexity of conventional deep reinforcement learning (DRL), we propose a dual-timescale hybrid framework integrating stochastic network calculus (SNC) with a double deep Q-network (DDQN). By leveraging SNC and signal-to-interference-plus-noise ratio (SINR) statistics, the action space is restricted to QoS-feasible resource allocations, substantially reducing learning complexity. Simulation results demonstrate that the proposed scheme satisfies eMBB rate and URLLC latency requirements, improves spectral efficiency by 39% compared to hard slicing, and reduces training time by 85% relative to standard DDQN.
- 12 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Wireless Communications Letters
- Vol. 15 (21622337) , 3304-3308
Dual-Timescale Resource Allocation for eMBB and URLLC Slices in Cell-Free RAN: A Hybrid SNC-DDQN Framework
In cell-free radio access networks (RANs), the co-existence of enhanced mobile broadband (eMBB) and ultra-reliable low-latency communication (URLLC) slices poses significant challenges for resource allocation under heterogeneous quality-of-service (QoS) constraints. To overcome the low spectral efficiency of static slicing and the high complexity of conventional deep reinforcement learning (DRL), we propose a dual-timescale hybrid framework integrating stochastic network calculus (SNC) with a double deep Q-network (DDQN). By leveraging SNC and signal-to-interference-plus-noise ratio (SINR) statistics, the action space is restricted to QoS-feasible resource allocations, substantially reducing learning complexity. Simulation results demonstrate that the proposed scheme satisfies eMBB rate and URLLC latency requirements, improves spectral efficiency by 39% compared to hard slicing, and reduces training time by 85% relative to standard DDQN.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Wireless Communications Letters
- Vol. 15, 5432-5436
Anti-Jamming Hybrid Beamforming for Millimeter-Wave Multi-Subarray Systems
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Wireless Communications Letters
- Vol. 15, 5432-5436
Anti-Jamming Hybrid Beamforming for Millimeter-Wave Multi-Subarray Systems
- 25 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 3879-3893
Cloud–Edge–End CF–MMIMO for UAV Swarms: Integrated Sensing and Robust Interference Detection in Perception Networks
With the rapid advancement of sixth-generation (6G) communications, uncrewed aerial vehicle (UAV) swarms have emerged as a key application scenario. However, their high mobility and complex electromagnetic environment pose significant communication challenges, such as frequent handovers and severe co-channel interference. Although conventional cell-free massive MIMO (CF-mMIMO) systems offer ubiquitous coverage, their high backhaul overhead and complex baseband processing limit practical deployment. To address these challenges, this paper proposes a novel cloud-edge-end architecture-based CF-mMIMO communication system to support robust communication in UAV swarm perception networks. Our designed scalable architecture leverages distributed edge units (EDUs) and user-centric distributed units (UCDUs) for collaborative processing, enabling distributed localized processing and user-centric coordination, thus supporting infinite collaboration expansion. Furthermore, we develop a hybrid signal detection scheme that combines EDU-level local minimum mean squared error estimation (MMSE) detection with UCDU-level global signal aggregation, achieving performance close to that of a centralized receiver while significantly reducing backhaul overhead. To address the problem of dynamic interference, we also propose a spatiotemporal-frequency forward continuous mean excision (FCME)-based interference detection framework. This framework effectively suppresses both narrowband and wideband interference by dynamically identifying interfered frequency bands and beams. Simulation results demonstrate that the proposed method outperforms conventional approaches in terms of signal combining and interference suppression, providing a reliable communication link for UAV swarm communications.
- 25 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 3879-3893
Cloud–Edge–End CF–MMIMO for UAV Swarms: Integrated Sensing and Robust Interference Detection in Perception Networks
With the rapid advancement of sixth-generation (6G) communications, uncrewed aerial vehicle (UAV) swarms have emerged as a key application scenario. However, their high mobility and complex electromagnetic environment pose significant communication challenges, such as frequent handovers and severe co-channel interference. Although conventional cell-free massive MIMO (CF-mMIMO) systems offer ubiquitous coverage, their high backhaul overhead and complex baseband processing limit practical deployment. To address these challenges, this paper proposes a novel cloud-edge-end architecture-based CF-mMIMO communication system to support robust communication in UAV swarm perception networks. Our designed scalable architecture leverages distributed edge units (EDUs) and user-centric distributed units (UCDUs) for collaborative processing, enabling distributed localized processing and user-centric coordination, thus supporting infinite collaboration expansion. Furthermore, we develop a hybrid signal detection scheme that combines EDU-level local minimum mean squared error estimation (MMSE) detection with UCDU-level global signal aggregation, achieving performance close to that of a centralized receiver while significantly reducing backhaul overhead. To address the problem of dynamic interference, we also propose a spatiotemporal-frequency forward continuous mean excision (FCME)-based interference detection framework. This framework effectively suppresses both narrowband and wideband interference by dynamically identifying interfered frequency bands and beams. Simulation results demonstrate that the proposed method outperforms conventional approaches in terms of signal combining and interference suppression, providing a reliable communication link for UAV swarm communications.
- Open Access
- 4 October 2025
- Published by JMIR Publications Inc. in JMIR Preprints
Large Language Model–Assisted Annotation Framework for Cross-Platform Analysis of Online Autism Communities: Implications for Parent Education and Digital Support (Preprint)
BACKGROUND Online health communities (OHCs) are important channels for families of children with autism spectrum disorder to obtain health information and psychosocial support. Differences between an open forum platform and the physician-patient consultation platforms may shape caregiver decisions, yet comparative evidence from China remains limited. A large language model (LLM) provides a scalable approach for systematic content annotation in large OHC datasets. OBJECTIVE This study proposes and validates a standardized LLM-assisted annotation framework under a unified classification schema and compares topic distributions and poster identities across an open forum platform (Baidu Tieba) and physician-patient consultation platforms (Chunyu Doctor and Haodf). METHODS We implemented an LLM-assisted annotation framework. A unified taxonomy of topics and poster identities was first developed through human open coding. Poster identities in Baidu Tieba were annotated through a double-blind manual procedure. For topic classification, interannotator and human-LLM agreement were evaluated on a manually labeled subset to benchmark models of varying sizes. The best-performing LLM was selected for full-dataset topic annotation, followed by statistical and cross-platform analysis. RESULTS When metrics were arithmetically averaged across all annotation tasks, the best-performing LLM achieved agreement levels comparable to human annotation (accuracy=79.18%, SD 0.20%; κ=0.736, SD 0.003; F1-score=0.727, SD 0.006), approaching interannotator agreement (accuracy=81.65%; κ=0.767; F1-score=0.758), demonstrating strong stability and scalability. Full-dataset analysis yielded 3 main findings. First, model performance increased with parameter scale but plateaued beyond 14B, indicating diminishing marginal returns from further scaling. Second, clear cross-platform differences in poster identity were observed: the open forum platform was dominated by family members of patients (caregivers: 2377/3516, 67.61%), with substantial participation from commercial rehabilitation practitioners (commercial posters: 427/3516, 12.14%), resulting in a more heterogeneous participation structure. Third, topic distributions reflected both shared high demand for resource-related information and differentiated help-seeking pathways: both platform types demonstrated consistently high demand for resource recommendation and evaluation; the open forum platform was primarily characterized by diagnosis-related discussions (1183/7535, 15.70%), whereas the physician-patient consultation platforms were centered on intervention-related consultation (2864/7687, 37.26%). CONCLUSIONS The LLM-assisted annotation framework proposed in this study enables reliable large-scale annotation of OHC data while maintaining high human-LLM agreement and operational stability. Midsized models (eg, 14B) demonstrated favorable cost-performance efficiency. The findings reveal 2 key aspects: the open forum platform exhibits a complex participation structure, and the influence of commercially affiliated actors should not be overlooked; users on both platform types show sustained demand for resource-related information but follow different help-seeking pathways, emphasizing diagnostic exploration and professional intervention, respectively. These results suggest that platform structure and governance mechanisms may shape caregivers’ information access and decision-making. The framework provides a transparent, reproducible, and cost-effective approach for OHC research. All data were deidentified and handled in accordance with relevant platform policies and ethical standards. CLINICALTRIAL
- 4 October 2025
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessLarge Language Model–Assisted Annotation Framework for Cross-Platform Analysis of Online Autism Communities: Implications for Parent Education and Digital Support (Preprint)
BACKGROUND Online health communities (OHCs) are important channels for families of children with autism spectrum disorder to obtain health information and psychosocial support. Differences between an open forum platform and the physician-patient consultation platforms may shape caregiver decisions, yet comparative evidence from China remains limited. A large language model (LLM) provides a scalable approach for systematic content annotation in large OHC datasets. OBJECTIVE This study proposes and validates a standardized LLM-assisted annotation framework under a unified classification schema and compares topic distributions and poster identities across an open forum platform (Baidu Tieba) and physician-patient consultation platforms (Chunyu Doctor and Haodf). METHODS We implemented an LLM-assisted annotation framework. A unified taxonomy of topics and poster identities was first developed through human open coding. Poster identities in Baidu Tieba were annotated through a double-blind manual procedure. For topic classification, interannotator and human-LLM agreement were evaluated on a manually labeled subset to benchmark models of varying sizes. The best-performing LLM was selected for full-dataset topic annotation, followed by statistical and cross-platform analysis. RESULTS When metrics were arithmetically averaged across all annotation tasks, the best-performing LLM achieved agreement levels comparable to human annotation (accuracy=79.18%, SD 0.20%; κ=0.736, SD 0.003; F1-score=0.727, SD 0.006), approaching interannotator agreement (accuracy=81.65%; κ=0.767; F1-score=0.758), demonstrating strong stability and scalability. Full-dataset analysis yielded 3 main findings. First, model performance increased with parameter scale but plateaued beyond 14B, indicating diminishing marginal returns from further scaling. Second, clear cross-platform differences in poster identity were observed: the open forum platform was dominated by family members of patients (caregivers: 2377/3516, 67.61%), with substantial participation from commercial rehabilitation practitioners (commercial posters: 427/3516, 12.14%), resulting in a more heterogeneous participation structure. Third, topic distributions reflected both shared high demand for resource-related information and differentiated help-seeking pathways: both platform types demonstrated consistently high demand for resource recommendation and evaluation; the open forum platform was primarily characterized by diagnosis-related discussions (1183/7535, 15.70%), whereas the physician-patient consultation platforms were centered on intervention-related consultation (2864/7687, 37.26%). CONCLUSIONS The LLM-assisted annotation framework proposed in this study enables reliable large-scale annotation of OHC data while maintaining high human-LLM agreement and operational stability. Midsized models (eg, 14B) demonstrated favorable cost-performance efficiency. The findings reveal 2 key aspects: the open forum platform exhibits a complex participation structure, and the influence of commercially affiliated actors should not be overlooked; users on both platform types show sustained demand for resource-related information but follow different help-seeking pathways, emphasizing diagnostic exploration and professional intervention, respectively. These results suggest that platform structure and governance mechanisms may shape caregivers’ information access and decision-making. The framework provides a transparent, reproducible, and cost-effective approach for OHC research. All data were deidentified and handled in accordance with relevant platform policies and ethical standards. CLINICALTRIAL
- 25 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (20) , 41608-41624
Federated-Learning-Enabled Cross-Modal Semantic Communication for 6G
In view of super-large scale access and dynamic connectivity requirements in 6G, the number of users and data is increasing exponentially, which makes it difficult to achieve sustainable development of communications. Meanwhile, with the development of cross-modal processing, semantic information is highly considered accurate and intelligent, which is expected to bring new ideas for 6G. Therefore, in this study, we propose a novel framework of cross-modal semantic communications to improve the efficiency of the multi-modal processing, especially the tactile modal. Meanwhile, we have redesigned the significant aspects of artificial intelligence (AI), such as encoding, transmission, and processing. As federated learning (FL) inherently supports multiple privacy-preserving and security measures, we also introduce FL to assist AI training in cross-modal semantic communications, including the expansion of multi-modal data, cross-modal semantic extraction, comprehensive decision-making, and privacy protection. First, in encoding aspects, we propose a hybrid coding for haptic signal coding by deep learning (DL) according to the semantic association between material identification and tactile code optimization. Second, in transmission aspects, we propose a modal-aware resource allocation for the fairness optimization between transmission requirements and network resources with deep reinforcement learning (DRL). Third, in signal processing, low-rank signal reconstruction and immersive quality of experience (QoE) evaluation by DL and machine learning (ML) are provided to improve and quantify users’ experience. In addition, the computational experiments of those key technologies are shown individually and the specification of their probability is discussed separately. Finally, future research topics related to the above issues are suggested.
- 25 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (20) , 41608-41624
Federated-Learning-Enabled Cross-Modal Semantic Communication for 6G
In view of super-large scale access and dynamic connectivity requirements in 6G, the number of users and data is increasing exponentially, which makes it difficult to achieve sustainable development of communications. Meanwhile, with the development of cross-modal processing, semantic information is highly considered accurate and intelligent, which is expected to bring new ideas for 6G. Therefore, in this study, we propose a novel framework of cross-modal semantic communications to improve the efficiency of the multi-modal processing, especially the tactile modal. Meanwhile, we have redesigned the significant aspects of artificial intelligence (AI), such as encoding, transmission, and processing. As federated learning (FL) inherently supports multiple privacy-preserving and security measures, we also introduce FL to assist AI training in cross-modal semantic communications, including the expansion of multi-modal data, cross-modal semantic extraction, comprehensive decision-making, and privacy protection. First, in encoding aspects, we propose a hybrid coding for haptic signal coding by deep learning (DL) according to the semantic association between material identification and tactile code optimization. Second, in transmission aspects, we propose a modal-aware resource allocation for the fairness optimization between transmission requirements and network resources with deep reinforcement learning (DRL). Third, in signal processing, low-rank signal reconstruction and immersive quality of experience (QoE) evaluation by DL and machine learning (ML) are provided to improve and quantify users’ experience. In addition, the computational experiments of those key technologies are shown individually and the specification of their probability is discussed separately. Finally, future research topics related to the above issues are suggested.
- 21 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (15) , 31221-31234
Hybrid Multicast/Unicast/D2D Transmission for Downlink Cell-Free Massive MIMO IoT Systems
This article concentrates on the synergetic effect of the hybrid multicast/unicast/device-to-device (D2D) transmission for downlink cell-free massive multiple-input-multiple-output (MIMO) Internet-of-Things (IoT) systems. By leveraging on the acquired imperfect channel state information (CSI) both for multicast, unicast, D2D links, particularly, we first derive the closed-form solutions for the multicast, unicast, and D2D transmission links, respectively. After that, based on the practical power consumption model, the achievable sum energy efficiency (EE) analysis is conducted as well by exploiting the achievable sum rate. At last, extensive simulation results are provided to validate the performance of the proposed framework. The obtained results are given to provide promising preliminary insights on the potential of deploying multicast/unicast/D2D in the cell-free massive MIMO topology. It is revealed by the above analysis that some guiding rules of the practical deployment of future. It is noteworthy that when we select the 5, 6, and 8 bits, we can achieve the maximum sum EE, the tradeoff from the total power consumption to the sum rate, and the maximum sum rate, respectively.
- 21 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (15) , 31221-31234
Hybrid Multicast/Unicast/D2D Transmission for Downlink Cell-Free Massive MIMO IoT Systems
This article concentrates on the synergetic effect of the hybrid multicast/unicast/device-to-device (D2D) transmission for downlink cell-free massive multiple-input-multiple-output (MIMO) Internet-of-Things (IoT) systems. By leveraging on the acquired imperfect channel state information (CSI) both for multicast, unicast, D2D links, particularly, we first derive the closed-form solutions for the multicast, unicast, and D2D transmission links, respectively. After that, based on the practical power consumption model, the achievable sum energy efficiency (EE) analysis is conducted as well by exploiting the achievable sum rate. At last, extensive simulation results are provided to validate the performance of the proposed framework. The obtained results are given to provide promising preliminary insights on the potential of deploying multicast/unicast/D2D in the cell-free massive MIMO topology. It is revealed by the above analysis that some guiding rules of the practical deployment of future. It is noteworthy that when we select the 5, 6, and 8 bits, we can achieve the maximum sum EE, the tradeoff from the total power consumption to the sum rate, and the maximum sum rate, respectively.
- Open Access
- 12 May 2025
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 27, e65317
Evaluation and Bias Analysis of Large Language Models in Generating Synthetic Electronic Health Records: Comparative Study
Background: Synthetic electronic health records (EHRs) generated by large language models (LLMs) offer potential for clinical education and model training while addressing privacy concerns. However, performance variations and demographic biases in these models remain underexplored, posing risks to equitable health care. Objective: This study aimed to systematically assess the performance of various LLMs in generating synthetic EHRs and to critically evaluate the presence of gender and racial biases in the generated outputs. We focused on assessing the completeness and representativeness of these EHRs across 20 diseases with varying demographic prevalence. Methods: A framework was developed to generate 140,000 synthetic EHRs using 10 standardized prompts across 7 LLMs. The electronic health record performance score (EPS) was introduced to quantify completeness, while the statistical parity difference (SPD) was proposed to assess the degree and direction of demographic bias. Chi-square tests were used to evaluate the presence of bias across demographic groups. Results: Larger models exhibited superior performance but heightened biases. The Yi-34B achieved the highest EPS (96.8), while smaller models (Qwen-1.8B: EPS=63.35) underperformed. Sex polarization emerged: female-dominated diseases (eg, multiple sclerosis) saw amplified female representation in outputs (Qwen-14B: 973/1000, 97.3% female vs 564,424/744,778, 75.78% real; SPD=+21.50%), while balanced diseases and male-dominated diseases skewed the male group (eg, hypertension Llama 2-13 B: 957/1000, 95.7% male vs 79,540,040/152,466,669, 52.17% real; SPD=+43.50%). Racial bias patterns revealed that some models overestimated the representation of White (eg, Yi-6B: mean SPD +14.40%, SD 16.22%) or Black groups (eg, Yi-34B: mean SPD +14.90%, SD 27.16%), while most models systematically underestimated the representation of Hispanic (average SPD across 7 models is –11.93%, SD 8.36%) and Asian groups (average SPD across 7 models is –0.77%, SD 11.99%). Conclusions: Larger models, such as Yi-34B, Qwen-14B, and Llama 2 to 13 B, showed improved performance in generating more comprehensive EHRs, as reflected in higher EPS values. However, this increased performance was accompanied by a notable escalation in both gender and racial biases, highlighting a performance-bias trade-off. The study identified 4 key findings as follows: (1) as model size increased, EHR generation improved, but demographic biases also became more pronounced; (2) biases were observed across all models, not just the larger ones; (3) gender bias closely aligned with real-world disease prevalence, while racial bias was evident in only a subset of diseases; and (4) racial biases varied, with some diseases showing overrepresentation of White or Black populations and underrepresentation of Hispanic and Asian groups. These findings underline the need for effective bias mitigation strategies and the development of benchmarks to ensure fairness in artificial intelligence applications for health care.
- 12 May 2025
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 27, e65317
Open AccessEvaluation and Bias Analysis of Large Language Models in Generating Synthetic Electronic Health Records: Comparative Study
Background: Synthetic electronic health records (EHRs) generated by large language models (LLMs) offer potential for clinical education and model training while addressing privacy concerns. However, performance variations and demographic biases in these models remain underexplored, posing risks to equitable health care. Objective: This study aimed to systematically assess the performance of various LLMs in generating synthetic EHRs and to critically evaluate the presence of gender and racial biases in the generated outputs. We focused on assessing the completeness and representativeness of these EHRs across 20 diseases with varying demographic prevalence. Methods: A framework was developed to generate 140,000 synthetic EHRs using 10 standardized prompts across 7 LLMs. The electronic health record performance score (EPS) was introduced to quantify completeness, while the statistical parity difference (SPD) was proposed to assess the degree and direction of demographic bias. Chi-square tests were used to evaluate the presence of bias across demographic groups. Results: Larger models exhibited superior performance but heightened biases. The Yi-34B achieved the highest EPS (96.8), while smaller models (Qwen-1.8B: EPS=63.35) underperformed. Sex polarization emerged: female-dominated diseases (eg, multiple sclerosis) saw amplified female representation in outputs (Qwen-14B: 973/1000, 97.3% female vs 564,424/744,778, 75.78% real; SPD=+21.50%), while balanced diseases and male-dominated diseases skewed the male group (eg, hypertension Llama 2-13 B: 957/1000, 95.7% male vs 79,540,040/152,466,669, 52.17% real; SPD=+43.50%). Racial bias patterns revealed that some models overestimated the representation of White (eg, Yi-6B: mean SPD +14.40%, SD 16.22%) or Black groups (eg, Yi-34B: mean SPD +14.90%, SD 27.16%), while most models systematically underestimated the representation of Hispanic (average SPD across 7 models is –11.93%, SD 8.36%) and Asian groups (average SPD across 7 models is –0.77%, SD 11.99%). Conclusions: Larger models, such as Yi-34B, Qwen-14B, and Llama 2 to 13 B, showed improved performance in generating more comprehensive EHRs, as reflected in higher EPS values. However, this increased performance was accompanied by a notable escalation in both gender and racial biases, highlighting a performance-bias trade-off. The study identified 4 key findings as follows: (1) as model size increased, EHR generation improved, but demographic biases also became more pronounced; (2) biases were observed across all models, not just the larger ones; (3) gender bias closely aligned with real-world disease prevalence, while racial bias was evident in only a subset of diseases; and (4) racial biases varied, with some diseases showing overrepresentation of White or Black populations and underrepresentation of Hispanic and Asian groups. These findings underline the need for effective bias mitigation strategies and the development of benchmarks to ensure fairness in artificial intelligence applications for health care.