Fully-Adaptive and Semi-Adaptive Frequency Sweep Algorithm Exploiting Loewner-State Model for EM Simulation of Multiport Systems
Shilpa T. N, Rakesh Sinha · 2025
Read Fully-Adaptive and Semi-Adaptive Frequency Sweep Algorithm Exploiting Loewner-State Model for EM Simulation of Multiport Systems by Shilpa T. N, Rakesh Sinha (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-09942.txt.
Introduction
Simulating how an electromagnetic device behaves across a band of frequencies can mean solving the physics at hundreds of points. Shilpa and Sinha present frequency sweep algorithms that pick the points on their own. They build compact state models from Loewner matrices and run two models of slightly different order or spacing. The gap between the two models serves as an error signal that steers where to sample next.
One version starts from just the band edges and adapts fully; the other first estimates how many starting points the structure's electrical size demands. The loop stops when the model matches the simulation within a set tolerance. Against two known sampling methods, the new algorithms ran faster, hit the accuracy target with fewer samples, and stayed close to full-wave results across four test cases.
Catalog record
| CREATOR | Shilpa T. N, Rakesh Sinha |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 69 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-09942
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Fully-Adaptive and Semi-Adaptive Frequency Sweep Algorithm Exploiting Loewner-State Model for EM Simulation of Multiport Systems Shilpa T. N, Rakesh Sinha Abstract This paper employs a fully adaptive and semi-adaptive frequency sweep algorithm using the Loewner matrix-based state model for the electromagnetic simulation. The proposed algorithms use two Loewner matrix models with different or the same orders with small frequency perturbation for adaptive frequency selection. The error between the two models is calculated in each iteration, and the next frequency points are selected to minimize maximum error. With the help of memory, the algorithm terminates when the error between the model and the simulation result is reached within the specified error tolerance. In the fully adaptive frequency sweep algorithm, the method starts with the minimum and maximum frequency of simulation. In the semi-adaptive algorithm, a novel approach has been proposed to determine the initial number of frequency points necessary for system interpolation based on the electrical size of the structure. The proposed algorithms have been compared with the Stoer–Bulirsch algorithm and Pradovera’s minimal sampling algorithm for electromagnetic simulation. Four examples are presented using MATLAB R2024b. The results show that the proposed methods offer better performance in terms of speed, accuracy and the requirement of the minimum number of frequency samples. The proposed method shows remarkable consistency with full-wave simulation data, and the algorithm can be effectively applicable to electromagnetic simulations. I Introduction Mathematical modelling [ 1 ] , [ 2 ] , [ 3 ] , [ 4 ] , [ 5 ] of system response or system function (transfer function) plays an important role in electrical and electronics engineering in understanding the system behaviour for given input or excitation. One of the most popular models is rational polynomial approximation [ 4 ] of the transfer function. Both applied mathematicians [ 6 ] , [ 7 ] , [ 8 ] , [ 9 ] and engineers [ 4 ] , [ 10 ] , [ 11 ] played important roles in the development of the rational polynomial model of functions using interpolation techniques. Some of the well-known models are rational polynomial [ 4 ] , sum of exponential [ 12 ] , [ 13 ] , pole-residue expansion (partial fraction expansion) [ 14 ] , state-space [ 8 ] , [ 11 ] , barycentric polynomial [ 6 ] , continued fraction [ 15 ] , etc. Various techniques have been developed to fit the data of electromagnetic and circuit parameters in mathematical models. The most popular techniques are vector fitting (VF) [ 14 ] , matrix pencil [ 12 ] , [ 13 ] , asymptotic waveform evaluation [ 16 ] , Stoer–Bulirsch Algorithm [ 17 ] , [ 18 ] , Cauchy’s Technique [ 19 ] , [ 20 ] , [ 21 ] , least square fitting, Frequency Domain Prony Method [ 22 ] , Loewner Matrix Method [ 8 ] [ 11 ] , AAA (Adaptive Antoulas-Anderson) algorithm [ 9 ] , [ 23 ] , [ 24 ] , [ 25 ] , Lagrange’s interpolation [ 26 ] , Thiele Interpolation [ 15 ] etc. For electrically large electromagnetic structures or higher-order electric circuits, it takes a significant amount of time to simulate the circuit or geometry at each frequency sample. In broadband simulation, the number of frequency samples ( N ) increases depending on the frequency range ( f_{max}-f_{min} ) and resolution \Delta f (difference between two successive frequency samples). Modelling the simulation (electromagnetic or circuit) results using a few frequency samples n and interpolating the results to N frequency points is the subject of fast frequency sweep (FFS). If the n frequency points are selected adaptively to reduce the error between the actual result H(s) and predicted model \widetilde{H}(s) , the method will be called adaptive frequency sweep (AFS). Please note that all AFS can be called FFS, but not all FFS are necessarily adaptive in nature or AFS. Most of the commercial EM simulators use the asymptotic waveform evaluation (AWE) [ 16 ] , [ 27 ] , [ 28 ] , [ 29 ] , [ 30 ] , [ 31 ] algorithm for AFS. The AWE was initially developed for time domain analysis of electrical circuits [ 16 ] . Later, it was applied for the Method of Moments (MoM) and finite element method (FEM) analysis of electromagnetic scattering [ 30 ] and microwave circuit [ 32 ] , [ 31 ] analysis. In the AWE technique, a Taylor series expansion/ Pade approximation is applied to the matrix equation (of the MoM or FEM) around a specific frequency. The coefficients of the Taylor series/ Pade are derived from the frequency derivatives of the matrices evaluated at the expansion frequency. These coefficients are then utilized to estimate the system’s frequency response across a specified frequency range. In asymptotic waveform evaluation, explicit moment matching was performed to obtain the dominating poles using the Pade approximation. However, the AWE approach is mathematically unstable when approximating higher-order moments. The equivalent circuit-based technique is a hybrid process; it first solves the EM portion of the system using fundamental methods such as the Method of Moments (MoM) or the Finite Element Method (FEM) and then approximates the circuit portion using Modified nodal analysis or other circuit approximation [ 33 ] . However, this method has a downside, which becomes clear when dealing with RF devices with many ports. In such cases, the equivalent circuit model may become rather complex, and calculating and extracting the characteristic parameters for each element inside the model can be time-consuming [ 34 ] . One of the early works on rational polynomial-based AFS was presented in [ 35 ] , where segmented rational models of the EM transfer function were obtained using a few samples in each segment. Rational polynomial base AFS has recently been explored in [ 36 ] , where interpolating and testing frequency points are selected using the successive arithmetic mean of adjacent frequency points. The use of local rational interpolation (or segmented rational interpolation) has been demonstrated in [ 37 ] for modelling measurement results and in [ 38 ] for adaptive frequency sampling. A continued fraction rational polynomial based Stoer–Bulirsch (S–B) algorithm [ 39 , sec. 2.2] has been used in [ 17 ] , [ 18 ] for AFS. The adaptive frequency points were selected to minimize the error between two competitive Thiele-type rational polynomials. The vector fitting [ 14 ] algorithm-based AFS algorithm was proposed in [ 40 ] , where adaptive frequency points were selected to minimize the error between two competitive pole-residue type rational polynomials. The segmented non-rational Lagrange interpolation and Spline Interpolation based AFS proposed in [ 26 ] and [ 41 ] . Scalar rational polynomial-based AFS techniques presented in [ 35 ] , [ 36 ] , [ 38 ] , [ 17 ] , [ 40 ] may not be suitable for large-scale multiport networks. As the speed-up factors of rational interpolation-based AFS algorithms have not been reported, it is difficult to say which algorithms work better in terms of speed-up factor. Recently, Hongliang Li et al. [ 34 ] introduced an FFS method for miniature passive RF