Qi-Jun Zhang

This letter proposes a novel resonance degree discrimination algorithm for the systematic compact embedded neuro-transfer function (SC-TF) parametric modeling method. As the variation range of geometrical parameters expands, the electromagnetic (EM) responses of microwave components exhibit stronger fluctuations, which lead to the emergence of abnormal pole–residue pairs and inconsistent distributions of high- and low-resonance sub-TFs under different geometrical parameters. To address these issues, a three-step strategy is introduced, which incorporates preliminary pole–residue classification, sub-TF scaling, and adaptive pole–residue tracking. The proposed algorithm effectively improves the modeling accuracy and applicability range of the SC-TF method. The effectiveness of the proposed algorithm is verified through two microwave filter examples.

As microwave devices evolve toward miniaturization and high-power operation, multiphysics (MP) simulation becomes essential to accurately capture real-world performance. However, MP-based optimization is more challenging than pure electromagnetic (EM) optimization due to its high computational cost, and the direct MP optimization process is easy to become trapped in local minima when the initial design is poor. While feature-assisted surrogate optimization has proven effective in solving this issue, its efficiency can be further improved. This letter proposes a novel sensitivity-based feature-assisted MP surrogate optimization technique. For the first time, adjoint sensitivity information is incorporated into the training of the MP-based feature surrogate, which significantly enhances sample efficiency and model accuracy. Furthermore, a scale-balanced objective function is formulated to eliminate dimensional disparities and accelerate convergence. The effectiveness of the proposed technique is validated through two microwave filter design examples.

The vector fitting (VF) technique is popularly used in neural network-transfer function (neuro-TF) modeling methods for electromagnetic (EM) behaviors of microwave filters. However, VF often produces nonsmooth and discontinuous transfer function (TF) coefficients under large geometrical variations, which degrade training stability and model accuracy. To address this issue, this article proposes a novel autoencoder-based neural network-TF (AENN-TF) parametric modeling method for EM behavior of microwave filters. In the proposed method, an advanced autoencoder (AE)-based feature parameter extraction technique is incorporated to develop a smooth relationship of TF parameters with respect to geometrical variables. This smooth relationship improves the neural network training accuracy, consequently improving the overall AENN-TF modeling accuracy for EM behavior. Novel derivative formulations are derived for the proposed model structure in order to perform fast gradient-based overall model training. Furthermore, we develop a novel four-step training framework with detailed modeling processes for the proposed AENN-TF modeling method. Compared to existing neuro-TF methods with the VF technique, the proposed AENN-TF method achieves better accuracy and robustness. The superior performance of the proposed method over existing neuro-TF methods is further demonstrated through two microwave application examples.

Space mapping (SM) accelerates full-wave electromagnetic (EM) optimization, but its efficiency depends strongly on the quality of the coarse model. In reduced-order modeling (ROM)-based SM, a reduced basis constructed at a nominal design may lose approximation capability under geometric deviations, thereby restricting the validity region of the reduced-order coarse model (ROCM). To address this issue, this article presents a common projection subspace (CPS)-based reduced-order coarse-modeling approach for ROM-based SM. By constructing a CPS from reduced bases extracted at multiple neighboring geometries, the proposed method improves the local modeling validity of the ROCM over the sampled neighborhood within each SM iteration while preserving high computational efficiency. Combined with mesh-deformation techniques, the method enables efficient coarse-model updating without repeated remeshing. Three representative microwave design examples, together with comparisons against direct EM optimization, mesh SM (MSM), aggressive SM (ASM), and nominal-basis ROM-SM, demonstrate that the proposed coarse-modeling strategy can reduce the number of SM iterations and total CPU cost in the tested cases while maintaining comparable final design accuracy.

When the performances of microwave waveguide filters are obviously influenced by thermal effects, designers have to consider and analyze the multiphysics coupling effect in devices. Most existing parametric modeling techniques based on pure electromagnetic (EM) are not suitable for multiphysics scenarios. Among the available approaches, vector fitting-based neuro-transfer functions (neuro-TFs) represent one of the techniques compatible with multiphysics parametric modeling due to their data-driven characteristics. However, vector fitting-based neuro-TF inherently suffers from discontinuities in the transfer function parameters caused by order changing. For pure EM problems, physics-driven transfer functions can be derived using finite element methods (FEMs) and model order reduction (MOR), which significantly improves the continuity of the extracted transfer function parameters and thereby enhances modeling accuracy. However, in the case of multiphysics problems, it is currently infeasible to extract physics-driven transfer function parameters. In this article, we propose an implicit iterative formulation for extracting multiphysics transfer function parameters based on FEM and MOR in the EM-thermal-mechanical multiphysics. The continuity of the multiphysics transfer function parameters is ensured through mesh deformation, and an iterative convergence criterion is established. Based on the extracted multiphysics transfer function parameters, a pole-zero neuro-TF method for multiphysics parametric modeling is developed. The proposed method is validated through two high-power filter examples. Compared with existing conventional neuro-TF and multiphysics modeling methods, the proposed multiphysics parametric modeling approach achieves significantly higher model accuracy.

Neuro-transfer function (Neuro-TF) has emerged as a prominent approach for electromagnetic (EM) parametric modeling, though traditional implementations for TF parameter extraction relying on vector fitting often suffer from parameter discontinuity. In this letter, we propose a novel framework integrating 1-D convolutional autoencoder (1D-CAE) with adjoint sensitivity analysis to establish high-fidelity parametric models for microwave filters. The proposed 1D-CAE-based TF parameter extraction captures intricate frequency-dependent relationships within S-parameters through its hierarchical convolutional layers, achieving efficient dimensionality reduction while preserving essential EM characteristics. To accelerate the training process, an adjoint sensitivity-enhanced multilayer perceptron (MLP) and EM sensitivity are used to map geometric parameters to the extracted TF parameters. The final hybrid architecture combines the sensitivity-informed MLP with the 1D-CAE decoder to form an end-to-end surrogate that directly translates geometric parameters into EM responses. Compared to conventional neuro-TF approaches, the proposed method achieves better model accuracy with fewer TF parameters. Two microwave filter modeling examples are provided for demonstration.

Recent research has demonstrated the exponential potential of hybrid quantum–classical algorithms (HQAs) in solving electromagnetic (EM) problems. However, the optimization objective of HQAs typically focuses on the magnitude of electric fields in EM problems, neglecting phase information. Consequently, the resulting quantum solution is inherently normalized to unity and subject to a global phase offset. This limitation prevents the calculation of the correct magnitude and phase of S-parameters. To address this challenge, we propose a block-encoding-based HQA (BEHQA) integrating a specialized quantum method for S-parameter extraction with magnitude and phase. The proposed method employs the specialized quantum method to reconstruct both the magnitude scaling and the phase offset, ensuring correct S-parameter extraction. Besides, HQAs fundamentally operate on unitary matrices, necessitating the explicit decomposition of nonunitary finite element method (FEM) matrices into many unitary operators. To mitigate the overhead associated with matrix decomposition, we introduce the block-encoding to embed the FEM matrix directly as a sub-block within a unitary operator. This strategy eliminates the need for decomposing FEM matrices to unitary operators, thereby reducing circuit complexity. For efficient optimization, we also derive a novel cost function specifically designed for the proposed BEHQA. The algorithm employs a strongly entangling ansatz (SEA) and gradient-based optimization to minimize the loss function. The effectiveness of BEHQA is demonstrated through its application to two microwave components, showcasing its feasibility for quantum-driven EM simulations.

The simulation-inserted optimization (SIO) technology accelerates design optimizations by executing simulation and optimization simultaneously. To broaden the applicability of the classical SIO, the adaptive augmented Lagrangian method (AALM) is proposed in this letter to introduce penalty terms and to achieve a wider feasible domain of design variables. Meanwhile, this article has also proposed a complex quasi-Newton method (CQNM) to avoid converting complex parameters into real-valued equivalents, thereby reducing matrix dimensionality and lowering computational time. Consequently, the proposed method expands the convergence region, improves convergence efficiency, accelerates optimization, and reduces the memory footprint. The AALM-CQNM is verified through two EM-based microwave applications of passive components.

This letter presents an ultra-wideband fractional-N PLL in 130-nm SiGe BiCMOS, achieving a continuous 67:1 tuning range from 125 MHz to 8.4 GHz with exceptionally low phase noise. The architecture overcomes the fundamental tuning-range-versus-phase-noise tradeoff through two key innovations. First, a high-isolation multi-core voltage-controlled oscillator (HIMC-VCO) with a novel signal combiner partitions the fundamental 2.0–4.2-GHz range across three cores, preserving a high tank Q-factor by preventing intercore interference. Second, a novel active-feedback push–push doubler (AF-PPD) extends the frequency coverage to 8.4 GHz. It employs a cross-coupled feedback network to synthesize a negative impedance, simultaneously enhancing conversion gain and achieving wideband input matching. The prototype demonstrates a phase noise of −152.9dBc/Hz at a 10-MHz offset and an integrated rms jitter of 146.8 fs and achieves a tuning-range-aware FoM_T of −251.8dB.