| Titre : | Design and Optimization of a Sixth-Order Microstrip Filter Using HFSS and 1D-CNN |
| Auteurs : | KOUIDRI Mohamed El Amine, Auteur ; AGAB RAMZI, Auteur ; BOUDKHIL Abdelhakim, Directeur de thèse |
| Type de document : | texte manuscrit |
| Editeur : | Université de Saïda – Dr. Moulay Tahar – Faculté des Mathématiques, de l’Informatique et des Télécommunications, 2025/2026 |
| Format : | 81ص |
| Accompagnement : | CD |
| Note générale : |
Modern wireless communication systems require compact, selective, and low-loss mi-
crowave components capable of operating in increasingly crowded frequency spectra. Bandpass filters are among the most important blocks in radio-frequency front-ends because they select the useful frequency band, reject adjacent-channel interference, and protect subsequent stages from unwanted signals. In practical microwave engineering, microstrip filters are attractive because of their planar implementation, low fabrication cost, compact size, and compatibility with printed circuit technology. The performance of a microwave filter depends strongly on the interaction between resonators, coupling gaps, substrate properties, losses, and input/output feeding. For high-selectivity responses, cross-coupled resonator topologies are widely used because they can generate finite transmission zeros and improve stopband rejection without increasing the filter order. Nevertheless, the diagnosis and tuning of such filters remain challenging because small geometrical deviations may significantly modify the return loss, insertion loss, bandwidth, and transmission-zero locations. The coupling matrix provides a compact and physically interpretable representation of a coupled-resonator filter. It links the electromagnetic behavior of the structure to a set of coefficients representing external coupling, main-path coupling, cross-coupling, and resonator detuning. Once extracted, the coupling matrix can be used to identify physical causes of performance degradation and guide the tuning process. However, classical extraction methods such as Cauchy approximation, vector fitting, and itera- tive optimization may require careful initialization, phase de-embedding, and expert intervention. This thesis investigates a learning-based alternative inspired by one-dimensional convolutional autoencoders. The proposed approach treats the extraction of the cou- pling matrix as an inverse regression problem from complex S-parameter responses. ANSYS HFSS is used to simulate the microstrip filter and export the complex S11 and S21 responses. These data are then converted into four channels: ℜ(S11 ), ℑ(S11 ), ℜ(S21 ), and ℑ(S21 ). A convolutional autoencoder is trained to predict the physically meaningful coupling coefficients while reconstructing the input response. The predicted matrix is finally validated by recomputing the S-parameters from the coupling-matrix 1 |
| Langues: | Français |
| Index. décimale : | BUC-M 008579 |
| Catégories : |
Master's Degree in Telecommunications Specialty: Networks and Telecommunications |
| Mots-clés: | microstrip filter, bandpass filter, coupling matrix, S-parameters, HFSS, 1D-CNN, convolutional autoencoder, transmission zeros, microwave filter diagnosis. |
| Résumé : |
This Master’s thesis presents the design, simulation, and intelligent diagnosis of a
sixth-order cross-coupled microstrip bandpass filter operating around 0.85 GHz with a target bandwidth of 40 MHz. The work combines full-wave electromagnetic simulation using ANSYS HFSS with coupling-matrix modeling and a one-dimensional convolu- tional autoencoder approach for coupling-coefficient extraction. The theoretical part introduces microwave filters, microstrip technology, S-parameters, return loss, insertion loss, coupled resonators, transmission zeros, equivalent circuits, and coupling-matrix theory. Classical extraction methods such as Cauchy approximation, vector fitting, and optimization are reviewed to motivate the proposed learning-based approach. The intelligent model uses the real and imaginary parts of S11 and S21 as four input channels and predicts a 17-coefficient coupling vector associated with the selected filter topol- ogy. The predicted matrix is validated physically by recomputing the S-parameters from the coupling matrix and comparing them with the target or HFSS response. The results show that the method is promising for rapid microwave filter diagnosis, es- pecially on synthetic matrix-generated data. However, direct extraction from HFSS data reveals discrepancies caused by phase-shift effects, dataset limitations, non-ideal electromagnetic phenomena, and insufficient physics-informed constraints. The thesis concludes that improved preprocessing, larger HFSS-enriched datasets, and stronger physics-informed loss functions are necessary to achieve more reliable coupling-matrix extraction for practical microstrip filter tuning and diagnosis. |
Exemplaires
| Code-barres | Cote | Support | Localisation | Section | Disponibilité |
|---|---|---|---|---|---|
| aucun exemplaire |
Documents numériques (1)
BUC-M 008579 Adobe Acrobat PDF |
