YLYanhe Lyu

Yanhe Lyu

*This is an algorithmically generated author record based on publicly available information.

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


Yanhe Lyu has a total of 24 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Dual-Branch Tandem Model-based Generative Design of Miniaturized Broadband Metantennas for Wideband Wide-Angle Scanning Array

    by Yanhe Lyu, Dunzhou Tong, Zhi Ning Chen

    A miniaturized broadband metasurface antenna (MTA) is synthesized through a dual-branch tandem neural network (DB-TNN)–enabled generative design method for a wideband wide-angle beamscanning array. The proposed MTA consists of a 4 × 4 non-uniform dislocated meandering metasurface and a feeding slot coupled to a fan-shaped microstrip branch. Each metacell introduces capacitive loadings and extends current paths, shifting resonant frequency lower for a miniaturized MTA. A DB-TNN has been developed for the high-dimensional design space arising from these non-uniform MTAs. The DB-TNN integrates inverse-generation and forward-prediction modules in the dual-branch architecture to process sensitivity-differing variables separately for efficient training and physically consistent synthesis. After cascaded training, the inverse model generates four MTA designs for an operating bandwidth of 61% within a compact 0.34 × 0.34-wavelength aperture. A 1×10 linear-array prototype formed by these generated MTAs achieves beam scanning with scan loss remaining below 2.6 dB across the scanning angular range from –45° to 45° and a band ranging from 4.5 to 8 GHz.

  • Top Cited Papers

    Deep Learning-Enabled Generative Antenna Design: Review and Prospects

    by Zhi Ning Chen, Peiqin Liu, Yu Kuang, Yanhe Lyu, Bo Zhang, Danyu Yang, Yejun He, Jian-Ming Jin

    This paper explores generative antenna design (GAD) and optimization using neural network (NN)-based deep learning (DL) methods. We first outline traditional antenna design and the challenges of applying NN-DL synthesis, particularly with generative adversarial networks (GANs). While conventional approaches reach performance limits, applying NN-DL generative methods to designs with increased degrees of freedom offers new possibilities for enhanced performance, additional functionality, and knowledge discovery. We then review the historical development and diverse applications of NN-based antenna optimization, covering individual elements, multi-antennas, and periodic structures. Key algorithmic concepts such as conditional GANs and variational autoencoders (VAEs) are discussed, emphasizing the role of prior knowledge in GAD. Recent advancements demonstrate how DL-enabled synthesis can surpass phase shift limits, control dispersion, and enhance metalens antenna gain bandwidth by 52%. The approach also enables full-range amplitude and phase control, effectively reducing sidelobe levels to below −30 dB. Finally, we highlight the significance of prior knowledge in GAD, distinguishing GAD from conventional parametric optimization. We discuss the challenges, future trends, and broader impacts of NN-driven generative design in next-generation antenna and electromagnetic device engineering.


  • Generative Deep Learning-Enabled Multiobjective Synthesis of Dual-Polarized Wide-Angle Loop-Family Frequency-Selective Surfaces

    by Yanhe Lyu, Yang Hong, Zhi Ning Chen

    A generative deep learning (GDL)-enabled approach is developed for the multi-objective synthesis of dual-polarized, wide-angle loop-family frequency selective surfaces (FSSs) by proposing a multichannel conditional generative adversarial network (MC-cGAN). The initial FSS topology and pattern generation rules are derived to equip the high-degree-of-freedom pixelated elements with the increased capability to achieve the desired objectives. The FSS topology comprises five layers of resonant and non-resonant loop-family slots and four substrates in a compact period of 0.22 wavelengths. The pattern generation rules incorporate the initial topology's structural features while considering sensitive inter-layer mutual coupling. MC-cGAN utilizes dual channels to learn multi-layer FSS patterns with varying feature-sensitive regions and integrates weighted model updates guided by generation rules. After training, the generator synthesizes diverse designs by inputting performance indicators as condition vectors. The prototype of GDL-enabled FSSs experimentally realized a passband from 2.5 to 3 GHz with in-band insertion loss < 2.6 dB, pre-band suppression > 31 dB from 0.5 to 2.3 GHz, and a roll-off from −3 dB to −10 dB over a band narrower than 0.08 GHz simultaneously, under a dual-polarized incidence at the angle from 0° to 60°.


  • Synthesis and Design of Ultrathin Angular-Selective Surfaces

    by Hao Jiang, Yanjie Wu, Yinghao Zhang, Yanhe Lyu, Zhi Ning Chen, Quan Xue

    This work presents a systematic methodology for the synthesis and design of ultrathin angular-selective surfaces (ASSs). The proposed approach utilizes normalized surface impedance models of planar unit cells to establish equivalent circuit representations in the angular domain, thereby enabling the accurate prediction of angular responses and efficient mapping to physical geometries. Based on this framework, three ASS designs are developed using single-layer substrates, among which two prototypes are fabricated to realize all-pass and high-pass angular transmission characteristics. Measurements confirm that the all-pass ASS achieves broadband, wide-angle transmission, whereas the high-pass ASS provides superior angular filtering selectivity, with a maximum measured roll-off rate (RoR) of 2.6dB/°. All designs maintain an ultrathin thickness of less than $0.005\lambda _{0}$ and are compatible with low-cost fabrication processes. These results verify the effectiveness of the proposed method and provide a scalable design framework for diverse angular-domain filtering responses.


  • Synthesis and Design of Near-Ideal Elliptical Filtering Frequency-Selective Surfaces

    by Hao Jiang, Mei Qian, Yinghao Zhang, Peiqin Liu, Yanhe Lyu, Zhi Ning Chen, Quan Xue

    This work presents the design and synthesis of near-ideal elliptical filtering frequency selective surfaces (FSSs). We propose a symmetrical composite circuit topology with four cascaded resonators interleaved with three dielectric layers. Series resonance in the central dual-layer resonators induces capacitive and inductive couplings, generating an elliptical response in the lower sideband. Meanwhile, cascaded parallel circuits within these resonators produce another elliptical response in the upper sideband. By cascading these resonant behaviors in an improved higher-order topology, the proposed FSS achieves five in-band transmission poles (TPs) and multiple transmission zeros (TZs) on both sidebands. Both simulation and measurement results demonstrate that the proposed FSS exhibits a wide 1-dB passband, an ultra-thin profile ((IL), and scalable out-of-band rejection. The near-ideal elliptical response, characterized by a shape factor (SF) of 1.01, remains stable for incidence angles up to 50°. These results highlight the proposed design’s effectiveness in realizing superior filtering performance compared to conventional FSS structures, representing the state-of-the-art in elliptical FSSs to date.


  • TE–TM Balanced Wide-Angle Metacells for Low Scan-Loss Metalens Antenna Using Prior Knowledge-Guided Generative Deep Learning-Enabled Method

    by Yanhe Lyu, Theng Huat Gan, Zhi Ning Chen

    The miniaturized cage-like metacell is proposed for TE-TM balanced wide-angle transmission using the prior knowledge (PK)-guided generative deep learning (DL) method, enabling a low scan-loss metalens antenna. An initial metacell topology and pattern generation rules are proposed, guided by physical constraints and engineering experience, and efficiently construct a high-degree-of-freedom dataset for training a conditional deep convolutional generative adversarial network (cDCGAN). With a trained generator, diverse DL-enabled miniaturized cage-like metacells achieve a transmittance higher than 0.75 with fluctuations below 0.15 and a phase shift range of 295° with variations less than 15° at 10 GHz under TM and TE polarized incident waves from 0° to 45°. To verify the generative designs, a metalens antenna prototype consisting of proposed metacells shows a realized gain of 26.2 dBi with an aperture efficiency of 36.3%, and measured scan losses lower than 2.6 and 2.4 dB as TE and TM polarized beams scanning from −40° to 40° at 10 GHz.


  • Dual-Polarized Multiresonant Metantenna for Quasi-Independent Radiation and In-Band Scattering Manipulation

    by Yan-He Lv, Zhi Ning Chen, Ren Wang, Bing-Zhong Wang

    A dual-polarized multi-resonant metantenna (MRMTA) is proposed for quasi-independent radiation and multiple in-band x -polarized scattering manipulations using characteristic mode analysis. The MRMTA unit cells in radiation and scattering modes have different resonance behaviors around meandering edges and inner slots, respectively. Moreover, tuning the inner slots of MRMTA unit cells can alter the reflection phase distributions for MRMTA’s different scattering functions but leave a weak influence on the radiation mode resonance, which enables dual-polarized radiation quasi-independent of multiple in-band x -polarized scattering manipulations. To verify the quasi-independence, the MRMTA’s top layout is designed based on 2-bit coding for two examples of x -polarized in-band low scattering and scattered-beam deflection. Fed by three stacked substrate integrated waveguides, the MRMTAs with different layouts can reduce the radar cross section by 16 dB, steer scattered beams near the desired deflection direction, and maintain stable in-band dual-polarized radiation properties across experiments and simulations.


  • Anisotropic Complementary Metantenna for Low Sidelobe Radiation and Low In-Band Co-Polarized Scattering Using Characteristic Mode Analysis

    by Yan-He Lv, Ren Wang, Bing-Zhong Wang, Zhi Ning Chen

    An anisotropic complementary metantenna (ACMA) is proposed for the integration of low sidelobe radiation and low in-band co-polarized (IBCP) scattering with the aid of characteristic mode analysis (CMA). With the guidance of CMA, the proposed ACMA evolving from a conventional metantenna (MTA) not only generates a coding phase distribution for the low scattering but also restores the broadside radiation modes like a uniform MTA. With the IBCP conditions set to x-polarization and the band around 9 GHz, the ACMA is constructed by mixing two types of double-layer anisotropic complementary metasurface unit cells. Two desired characteristic modes of the ACMA are simultaneously excited by a substrate integrated waveguide cavity through dual bow-tie slots. Both the simulation and experiments demonstrate that the proposed ACMA achieves both the x-polarized broadside radiation with sidelobe level below -18.9 dB and the IBCP scattering levels below -17.8 dB over the frequency range from 8.87 to 9.78 GHz.


  • Knowledge-Based Neural Network for Thinned Array Modeling With Active Element Patterns

    by Yang Hong, Wei Shao, Yan-He Lv, Bing-Zhong Wang, Lin Peng, Baojun Jiang

    To speed up the process for designing finite thinned arrays, an efficient artificial neural network (ANN) model considering mutual coupling effects is proposed in this communication. Array elements are divided into three categories based on the active element pattern (AEP) technique, and sub-arrays are constructed to extract the AEP of each category of elements as training samples. The proposed model including parallel ANN branches takes into account the mutual coupling between elements and avoids the calculation of the radiation pattern of the entire array. To improve modeling performance, furthermore, a prior knowledge input technique is used to reduce the complexity of the input/output relationship that an ANN has to learn. The existing knowledge is obtained by back propagation ANN modeling. Once the knowledge-based model is well trained, it is repeatedly called by the genetic algorithm for the optimal solution as a substitute for the full-wave simulation. Numerical examples of a U-shaped slot thinned array and a dual-layer patch thinned array are utilized to verify the efficiency of the proposed scheme.