Peng Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Peng Wang has a total of 24 co-authors. Here are the top co-authors by number of shared publications:
- Gert Frolund Pedersen
Aalborg University
11 shared publications - Ming Shen
Aalborg University
11 shared publications - Tao Hong
Xidian University
9 shared publications - Wen Jiang
Xidian University
9 shared publications - Zhaohui Wei
National Key Laboratory of Radar Detection and Sensing, Xidian University, Xian, China
8 shared publications - Zhenning Li
Chinese Academy of Sciences, Aerospace Information Research Institute, Beijing, China
5 shared publications
Latest Publications
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Microwave Theory and Techniques
- Vol. 74 (10) , 8779-8795
From Fixed to Arbitrary Frequency: A Scalable LLM-Enhanced Framework for Electromagnetic Device Design
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Microwave Theory and Techniques
- Vol. 74 (10) , 8779-8795
From Fixed to Arbitrary Frequency: A Scalable LLM-Enhanced Framework for Electromagnetic Device Design
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Microwave Theory and Techniques
- Vol. 74 (10) , 9112-9122
An Antenna Array Fault Diagnosis Method: RSNN Detection Based on Quasi-Far-Field Measurements
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Microwave Theory and Techniques
- Vol. 74 (10) , 9112-9122
An Antenna Array Fault Diagnosis Method: RSNN Detection Based on Quasi-Far-Field Measurements
- 19 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Antennas and Propagation
- Vol. 74 (2) , 2167-2172
The Gradual Interval Mapping Method Integration Framework for Multiple Metasurfaces Fast Design
This work presents a gradual interval mapping method (IMM) integration framework for the rapid design of multiple metasurfaces. By leveraging the deep neural networks, the gradual design framework efficiently predicts both S-parameters and geometric structures using interval folding lines as input. Unlike traditional and current approaches, the proposed method integrates IMM with a filling strategy, simplifying data processing and enhancing prediction accuracy. To validate its effectiveness, three types of frequency selective surface (FSS) are designed, one of which is fabricated and experimentally measured in a microwave chamber. Simulation and measurement results confirm the superior performance of the framework, highlighting its potential for intelligent electromagnetic device design. Finally, we conclude with a short discussion of the work, including its limitations and developability.
- 19 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Antennas and Propagation
- Vol. 74 (2) , 2167-2172
The Gradual Interval Mapping Method Integration Framework for Multiple Metasurfaces Fast Design
This work presents a gradual interval mapping method (IMM) integration framework for the rapid design of multiple metasurfaces. By leveraging the deep neural networks, the gradual design framework efficiently predicts both S-parameters and geometric structures using interval folding lines as input. Unlike traditional and current approaches, the proposed method integrates IMM with a filling strategy, simplifying data processing and enhancing prediction accuracy. To validate its effectiveness, three types of frequency selective surface (FSS) are designed, one of which is fabricated and experimentally measured in a microwave chamber. Simulation and measurement results confirm the superior performance of the framework, highlighting its potential for intelligent electromagnetic device design. Finally, we conclude with a short discussion of the work, including its limitations and developability.
- 5 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Microwave Theory and Techniques
- Vol. 73 (5) , 2475-2489
M-IFNN: A Multiple Inverse–Forward Requirement-Based Deep Learning Framework for Generated Design of Metasurfaces
This article presents a fully automated inverse–forward design of multiple electromagnetic (EM) structure designs by using a requirement-based multiple inverse–forward neural network (M-IFNN). Compared to the traditional design method, the proposed method can obtain the structure model, geometric parameters, and $S$ -parameters by just inputting sentences into a vector-extracted language model, which output to M-IFNN. In addition, we use the parallel-combined method to achieve inverse–forward output by only one neural network for the first time. To achieve the above target, first, we use three feature extraction algorithms to obtain the feature vector of three metasurfaces. Then, we use identification vector and filling methods to achieve multi-EM designs by one neural network. Subsequently, we use the parallel-combined method to integrate the inverse geometric parameters design and forward-parameters design into one model. Deep neural network (DNN) is chosen for M-IFNN because it has smaller storage memory and good results. The M-IFNN finally achieves a 100% correctness prediction of corresponding structures and a 0.41/0.034 mean absolute error (MAE) result for predicting inverse/forward geometric parameters/ $S$ -parameters. Three prototypes are fabricated and measured. The measurement results are in good agreement with the simulation results. The good validation results prove the feasibility of the proposed method. Additionally, several dialectical discussions about M-IFNN are at the end of the article.
- 5 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Microwave Theory and Techniques
- Vol. 73 (5) , 2475-2489
M-IFNN: A Multiple Inverse–Forward Requirement-Based Deep Learning Framework for Generated Design of Metasurfaces
This article presents a fully automated inverse–forward design of multiple electromagnetic (EM) structure designs by using a requirement-based multiple inverse–forward neural network (M-IFNN). Compared to the traditional design method, the proposed method can obtain the structure model, geometric parameters, and $S$ -parameters by just inputting sentences into a vector-extracted language model, which output to M-IFNN. In addition, we use the parallel-combined method to achieve inverse–forward output by only one neural network for the first time. To achieve the above target, first, we use three feature extraction algorithms to obtain the feature vector of three metasurfaces. Then, we use identification vector and filling methods to achieve multi-EM designs by one neural network. Subsequently, we use the parallel-combined method to integrate the inverse geometric parameters design and forward-parameters design into one model. Deep neural network (DNN) is chosen for M-IFNN because it has smaller storage memory and good results. The M-IFNN finally achieves a 100% correctness prediction of corresponding structures and a 0.41/0.034 mean absolute error (MAE) result for predicting inverse/forward geometric parameters/ $S$ -parameters. Three prototypes are fabricated and measured. The measurement results are in good agreement with the simulation results. The good validation results prove the feasibility of the proposed method. Additionally, several dialectical discussions about M-IFNN are at the end of the article.
- 29 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Antennas and Propagation
- Vol. 72 (5) , 4298-4310
Preprocessing-Based Fast Design of Multiple EM Structures With One Deep Neural Network
Deep learning plays a vital role in the design of electromagnetic (EM) structures. However, in current research, a single neural network typically supports only one structure design and requires a complex framework to accommodate multiple structure designs. This paper proposes using one neural-assist design for facilitating multiple EM structures. We employ two filling methods to control the vector length, an identification method to ensure accurate prediction results, and a random auxiliary vectors method to increase the data volume and reduce loss. Subsequently, we design a forward neural network (FNN) and an inverse neural network (INN) using the proposed method. The developed neural network is used to complete the dual-passband frequency selective surface (DP-FSS), space-time-coding digital metasurface element (STCDME), single/dual absorbing metasurface (SDAM), and dual-stopband frequency selective surface (DS-FSS) designs. The mean absolute error (MAE) loss values for the FNN/INN predictions and actual results for these four structures are 0.019/0.116, 0.035/0.602, 5.14/0.146, and 0.018/0.07, respectively. Finally, we design four structures with the well-training network, fabricate DP-FSS and DS-FSS, and measure them in an anechoic chamber. The measurement and simulation results are in good agreement. The proposed method significantly reduces the complexity of multiple EM structure designs, decreases the need for multiple neural networks, and simplifies the design framework, thereby contributing to the development of AI-assisted EM structures.
- 29 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Antennas and Propagation
- Vol. 72 (5) , 4298-4310
Preprocessing-Based Fast Design of Multiple EM Structures With One Deep Neural Network
Deep learning plays a vital role in the design of electromagnetic (EM) structures. However, in current research, a single neural network typically supports only one structure design and requires a complex framework to accommodate multiple structure designs. This paper proposes using one neural-assist design for facilitating multiple EM structures. We employ two filling methods to control the vector length, an identification method to ensure accurate prediction results, and a random auxiliary vectors method to increase the data volume and reduce loss. Subsequently, we design a forward neural network (FNN) and an inverse neural network (INN) using the proposed method. The developed neural network is used to complete the dual-passband frequency selective surface (DP-FSS), space-time-coding digital metasurface element (STCDME), single/dual absorbing metasurface (SDAM), and dual-stopband frequency selective surface (DS-FSS) designs. The mean absolute error (MAE) loss values for the FNN/INN predictions and actual results for these four structures are 0.019/0.116, 0.035/0.602, 5.14/0.146, and 0.018/0.07, respectively. Finally, we design four structures with the well-training network, fabricate DP-FSS and DS-FSS, and measure them in an anechoic chamber. The measurement and simulation results are in good agreement. The proposed method significantly reduces the complexity of multiple EM structure designs, decreases the need for multiple neural networks, and simplifies the design framework, thereby contributing to the development of AI-assisted EM structures.
- 10 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Antennas and Propagation
- Vol. 72 (6) , 4962-4975
Space-Time-Coding Digital Metasurface Element Design Based on State Recognition and Mapping Methods With CNN-LSTM-DNN
Space-time-coding digital metasurface has drawn worldwide attention with the ability to improve communication quality and change the direction of electromagnetic (EM) wave propagation in real-time. This article proposes a deep learning-assisted method to design the space-time-coding digital metasurface element (STCDME) with the state recognition and mapping technique methods. Compared to traditional pure EM simulation methods to simulate all states, the proposed method fully considers the relationship between different states of STCDME to accelerate the design. Firstly, we use the state recognition method to distinguish the S-parameters’ states. Subsequently, with the state mapping method, we use one state phase to predict the other states’ S-parameters. Simulation time is reduced by half with the proposed methods. Various algorithms are compared, and finally, the CNN-LSTM-DNN (CLD) hybrid algorithm is chosen for the proposed methods, which is also the first instance of using the CLD algorithm to design the EM structure. The proposed method is validated with two design examples. An element prototype is made and measured with the vector network analyzer. The measurement results agree with the simulation results. We then use the designed STCDME to complete the design of the beamforming array, which realizes the beamforming at 0°, 15°, 30°, and 45° angles. Finally, we briefly discuss the application/drawbacks of the proposed methods and the CLD model in EM fields.
- 10 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Antennas and Propagation
- Vol. 72 (6) , 4962-4975
Space-Time-Coding Digital Metasurface Element Design Based on State Recognition and Mapping Methods With CNN-LSTM-DNN
Space-time-coding digital metasurface has drawn worldwide attention with the ability to improve communication quality and change the direction of electromagnetic (EM) wave propagation in real-time. This article proposes a deep learning-assisted method to design the space-time-coding digital metasurface element (STCDME) with the state recognition and mapping technique methods. Compared to traditional pure EM simulation methods to simulate all states, the proposed method fully considers the relationship between different states of STCDME to accelerate the design. Firstly, we use the state recognition method to distinguish the S-parameters’ states. Subsequently, with the state mapping method, we use one state phase to predict the other states’ S-parameters. Simulation time is reduced by half with the proposed methods. Various algorithms are compared, and finally, the CNN-LSTM-DNN (CLD) hybrid algorithm is chosen for the proposed methods, which is also the first instance of using the CLD algorithm to design the EM structure. The proposed method is validated with two design examples. An element prototype is made and measured with the vector network analyzer. The measurement results agree with the simulation results. We then use the designed STCDME to complete the design of the beamforming array, which realizes the beamforming at 0°, 15°, 30°, and 45° angles. Finally, we briefly discuss the application/drawbacks of the proposed methods and the CLD model in EM fields.
- 29 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Antennas and Propagation
- Vol. 72 (2) , 1319-1328
Fast and Automatic Parametric Model Construction of Antenna Structures Using CNN–LSTM Networks
Deep learning-assisted antenna design methods such as surrogate models have gained significant popularity in recent years due to their potential to greatly increase design efficiencies by replacing the time-consuming full-wave electromagnetic (EM) simulations. A large number of training data with sufficiently diverse and representative samples (antenna structure parameters, scattering properties, etc.) is mandatory for these methods to ensure good performance. However, traditional antenna modeling methods relying on manual model construction and modification are time-consuming and cannot meet the requirement of efficient training data acquisition. Also, automatic model construction methods are rarely studied. In this paper, we pioneer investigation into the antenna model construction problem and firstly proposed a deep learning-assisted and image-based approach for achieving automatic model construction. Specifically, our method only needs an image of the antenna structure, usually available in scientific publications, as the input while the corresponding modeling codes (VBA language) are generated automatically. The proposed model mainly consists of two parts: Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM) networks. The former is used for capturing features of antenna structure images and the latter is employed to generate the modeling codes. Experimental results show that the proposed method can automatically achieve the antenna parametric model construction with an overhead of approximately fifty seconds, which is a significant time reduction than manual modeling. The proposed parametric model construction method lays the foundation for further data acquisition, tuning, analysis and optimization.
- 29 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Antennas and Propagation
- Vol. 72 (2) , 1319-1328
Fast and Automatic Parametric Model Construction of Antenna Structures Using CNN–LSTM Networks
Deep learning-assisted antenna design methods such as surrogate models have gained significant popularity in recent years due to their potential to greatly increase design efficiencies by replacing the time-consuming full-wave electromagnetic (EM) simulations. A large number of training data with sufficiently diverse and representative samples (antenna structure parameters, scattering properties, etc.) is mandatory for these methods to ensure good performance. However, traditional antenna modeling methods relying on manual model construction and modification are time-consuming and cannot meet the requirement of efficient training data acquisition. Also, automatic model construction methods are rarely studied. In this paper, we pioneer investigation into the antenna model construction problem and firstly proposed a deep learning-assisted and image-based approach for achieving automatic model construction. Specifically, our method only needs an image of the antenna structure, usually available in scientific publications, as the input while the corresponding modeling codes (VBA language) are generated automatically. The proposed model mainly consists of two parts: Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM) networks. The former is used for capturing features of antenna structure images and the latter is employed to generate the modeling codes. Experimental results show that the proposed method can automatically achieve the antenna parametric model construction with an overhead of approximately fifty seconds, which is a significant time reduction than manual modeling. The proposed parametric model construction method lays the foundation for further data acquisition, tuning, analysis and optimization.
- 15 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Microwave Theory and Techniques
- Vol. 72 (3) , 1467-1479
A Machine Learning Framework for the Design of STCDME Structures in RIS Applications
This article introduces a machine learning (ML) framework for the design of space–time-coding digital metasurface elements (STCDMEs), commonly used in reconfigurable intelligent surface (RIS)-based communication. It includes inverse design, forward design, and automodeling, which quickly achieve multistate electromagnetic (EM) structure designs, e.g., STCDME. The decision tree (DT) model is chosen for use with its lightweight, fast response, and highly accurate in EM structure small-scale data modeling. In addition, we present a new sensing STCDME design method, using gap technology based on ground structure. Using the proposed framework, we successfully design a sensing STCDME with a reflection phase within 180 $^{\circ}$ $\pm$ 30 $^{\circ}$ ranging from 6.66 to 7.3 GHz and a reflection coefficient larger than $-$ 2 dB, meeting RIS communication requirements. In the 8.27–9.5-GHz band, the structure’s transmission coefficient exceeds $-$ 3 dB, achieving EM wave transmission and sensing capabilities. The proposed framework offers a novel method for STCDME design, and the resulting sensing STCDME structure can be used for RIS sensing, contributing significantly to wireless communication and sensing applications.
- 15 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Microwave Theory and Techniques
- Vol. 72 (3) , 1467-1479
A Machine Learning Framework for the Design of STCDME Structures in RIS Applications
This article introduces a machine learning (ML) framework for the design of space–time-coding digital metasurface elements (STCDMEs), commonly used in reconfigurable intelligent surface (RIS)-based communication. It includes inverse design, forward design, and automodeling, which quickly achieve multistate electromagnetic (EM) structure designs, e.g., STCDME. The decision tree (DT) model is chosen for use with its lightweight, fast response, and highly accurate in EM structure small-scale data modeling. In addition, we present a new sensing STCDME design method, using gap technology based on ground structure. Using the proposed framework, we successfully design a sensing STCDME with a reflection phase within 180 $^{\circ}$ $\pm$ 30 $^{\circ}$ ranging from 6.66 to 7.3 GHz and a reflection coefficient larger than $-$ 2 dB, meeting RIS communication requirements. In the 8.27–9.5-GHz band, the structure’s transmission coefficient exceeds $-$ 3 dB, achieving EM wave transmission and sensing capabilities. The proposed framework offers a novel method for STCDME design, and the resulting sensing STCDME structure can be used for RIS sensing, contributing significantly to wireless communication and sensing applications.
- 14 April 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Antennas and Propagation
- Vol. 71 (7) , 5549-5557
Automated Antenna Design via Domain Knowledge-Informed Reinforcement Learning and Imitation Learning
In this study, we propose a novel approach for automating antenna design through the utilization of domain knowledge-informed reinforcement learning (RL) and imitation learning (IL). The proposed method allows the RL agent to learn from its interactions with the environment and to take actions that maximize the reward signal, resulting in the optimal policy. Despite the attractive advantages, the learning process of RL can be challenging and time-consuming, as the agent must explore a vast range of states, actions, and policies. To enhance the efficiency of the design process, we incorporate two key components into the proposed method. Firstly, we utilize domain knowledge, such as the scaling property of antenna, to initialize the antenna parameters. This not only provides interpretability to the design process but also reduces the solution space. Additionally, we employ IL to pretrain the decision-making network within the RL algorithm, which significantly mitigates the data inefficiency issue for RL and reduces the time required to explore feasible solutions. The proposed method has been evaluated using the design of a slot filtering antenna as a case study. Simulation results indicate that the proposed method can achieve the antenna design for different frequencies and bandwidth metrics with at least 50% fewer adjusting steps than the traditional reinforcement learning method, such as deep deterministic policy gradient (DDPG). The combination of RL and IL also results in a considerable performance boost compared to both pure IL and pure RL for antenna designs.
- 14 April 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Antennas and Propagation
- Vol. 71 (7) , 5549-5557
Automated Antenna Design via Domain Knowledge-Informed Reinforcement Learning and Imitation Learning
In this study, we propose a novel approach for automating antenna design through the utilization of domain knowledge-informed reinforcement learning (RL) and imitation learning (IL). The proposed method allows the RL agent to learn from its interactions with the environment and to take actions that maximize the reward signal, resulting in the optimal policy. Despite the attractive advantages, the learning process of RL can be challenging and time-consuming, as the agent must explore a vast range of states, actions, and policies. To enhance the efficiency of the design process, we incorporate two key components into the proposed method. Firstly, we utilize domain knowledge, such as the scaling property of antenna, to initialize the antenna parameters. This not only provides interpretability to the design process but also reduces the solution space. Additionally, we employ IL to pretrain the decision-making network within the RL algorithm, which significantly mitigates the data inefficiency issue for RL and reduces the time required to explore feasible solutions. The proposed method has been evaluated using the design of a slot filtering antenna as a case study. Simulation results indicate that the proposed method can achieve the antenna design for different frequencies and bandwidth metrics with at least 50% fewer adjusting steps than the traditional reinforcement learning method, such as deep deterministic policy gradient (DDPG). The combination of RL and IL also results in a considerable performance boost compared to both pure IL and pure RL for antenna designs.