24 November 2021 in Journal of Scientific Computing
10 May 2024 in IEEE Transactions on Microwave Theory and Techniques
This article proposes a general and efficient filter topology reconfiguration method based on matrix completion theory, which allows for frequency-dependent couplings (FDCs) and nonresonating nodes (NRNs). The projected gradient of the objective function on the manifold of constraints is determined analytically, expressed as the stiff ordinary differential equation (ODE). This constructs the descent flow that can be numerically followed to solve the ODE with integration. Since all the gradient calculations are performed with advanced matrix operations instead of elementwise computation, the proposed synthesis algorithm is significantly efficient. Moreover, to cover more varied filter networks, the ODE expressions of the projected gradient are extended to the topologies incorporating FDCs and NRNs. Multiple synthesis examples are presented to demonstrate the efficiency of the approach. In particular, one experimental example is used to exhibit the parameter extraction application of the proposed method for post-production tuning.
7 October 2022 in Advances in Computational Mathematics
19 May 2025 in Journal of Scientific Computing
3 June 2025 in IEEE Transactions on Microwave Theory and Techniques
The conventional inverse eigenvalue problem (IEP)-based optimization model has proven effective for various coupling matrix (CM) reconfigurations. However, its mathematical foundations remain unclear, and it suffers from significant limitations, including high sensitivity to initial values and a tendency to converge to local optima, especially for large-scale problems. This study not only revisits and elucidates the mathematical theory underlying this approach but also introduces a novel IEP-based optimization model that uses a strictly diagonally dominant matrix and parameter reduction to reduce sensitivity to initial values. In addition, the “uphill move” strategy is integrated into the damped Levenberg–Marquardt (LM) optimization method to facilitate escape from local minima and accelerate convergence. Extensive numerical experiments validate the proposed optimization framework, demonstrating its rapid convergence and robustness, particularly in scenarios with large-scale problems and varying initial values.
24 April 2022 in Journal of Scientific Computing