Optimizing Q-Learning for Automated Cavity Filter Tuning: Leveraging PCA and Neural Networks
This paper presents a reinforcement learning-based approach to automate the tuning of a 6thorder combline bandpass filter, operating at 941 MHz, using a Q-learning algorithm. To reduce complexity, only two tuning screws are considered in the optimization. One of the main challenges in this process l...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
EDP Sciences
2025-01-01
|
| Series: | EPJ Web of Conferences |
| Online Access: | https://www.epj-conferences.org/articles/epjconf/pdf/2025/11/epjconf_cofmer2025_01006.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850269531991179264 |
|---|---|
| author | Aghanim Amina Otman Oulhaj Oukaira Aziz Lasri Rafik |
| author_facet | Aghanim Amina Otman Oulhaj Oukaira Aziz Lasri Rafik |
| author_sort | Aghanim Amina |
| collection | DOAJ |
| description | This paper presents a reinforcement learning-based approach to automate the tuning of a 6thorder combline bandpass filter, operating at 941 MHz, using a Q-learning algorithm. To reduce complexity, only two tuning screws are considered in the optimization. One of the main challenges in this process lies in the nonlinear relationship between screw positions and the filter’s frequency response, making conventional tuning methods difficult and inefficient. Additionally, while intelligent algorithms can assist in tuning, they often require large volumes of simulated data, leading to high computational costs. However, reducing the dataset size can compromise accuracy, as important frequency response information may be lost. To overcome these limitations, PCA is applied to minimize the dimensionality of the S11 response data, keeping only the most relevant information while improving computational efficiency. A feedforward neural network is employed to predict the PCA-reduced S-parameters, serving as a surrogate model that enables faster decision-making within the Q-learning framework. By integrating PCA at the data preprocessing stage, the number of frequency points is reduced from 401 to 20, significantly accelerating the Q-learning convergence process. The proposed approach, successfully reduces the tuning process from 1000 steps to just 45, ensuring faster and more precise optimization. |
| format | Article |
| id | doaj-art-86be0cfada9a499fbb5ffb0089ffb06c |
| institution | OA Journals |
| issn | 2100-014X |
| language | English |
| publishDate | 2025-01-01 |
| publisher | EDP Sciences |
| record_format | Article |
| series | EPJ Web of Conferences |
| spelling | doaj-art-86be0cfada9a499fbb5ffb0089ffb06c2025-08-20T01:53:06ZengEDP SciencesEPJ Web of Conferences2100-014X2025-01-013260100610.1051/epjconf/202532601006epjconf_cofmer2025_01006Optimizing Q-Learning for Automated Cavity Filter Tuning: Leveraging PCA and Neural NetworksAghanim Amina0Otman Oulhaj1Oukaira Aziz2Lasri Rafik3TED: AEEP, FPL, Abdelmalek Essaâdi UniversityTED: AEEP, FPL, Abdelmalek Essaâdi UniversityFaculty of Engineering, Electrical Engineering Department, Moncton UniversityTED: AEEP, FPL, Abdelmalek Essaâdi UniversityThis paper presents a reinforcement learning-based approach to automate the tuning of a 6thorder combline bandpass filter, operating at 941 MHz, using a Q-learning algorithm. To reduce complexity, only two tuning screws are considered in the optimization. One of the main challenges in this process lies in the nonlinear relationship between screw positions and the filter’s frequency response, making conventional tuning methods difficult and inefficient. Additionally, while intelligent algorithms can assist in tuning, they often require large volumes of simulated data, leading to high computational costs. However, reducing the dataset size can compromise accuracy, as important frequency response information may be lost. To overcome these limitations, PCA is applied to minimize the dimensionality of the S11 response data, keeping only the most relevant information while improving computational efficiency. A feedforward neural network is employed to predict the PCA-reduced S-parameters, serving as a surrogate model that enables faster decision-making within the Q-learning framework. By integrating PCA at the data preprocessing stage, the number of frequency points is reduced from 401 to 20, significantly accelerating the Q-learning convergence process. The proposed approach, successfully reduces the tuning process from 1000 steps to just 45, ensuring faster and more precise optimization.https://www.epj-conferences.org/articles/epjconf/pdf/2025/11/epjconf_cofmer2025_01006.pdf |
| spellingShingle | Aghanim Amina Otman Oulhaj Oukaira Aziz Lasri Rafik Optimizing Q-Learning for Automated Cavity Filter Tuning: Leveraging PCA and Neural Networks EPJ Web of Conferences |
| title | Optimizing Q-Learning for Automated Cavity Filter Tuning: Leveraging PCA and Neural Networks |
| title_full | Optimizing Q-Learning for Automated Cavity Filter Tuning: Leveraging PCA and Neural Networks |
| title_fullStr | Optimizing Q-Learning for Automated Cavity Filter Tuning: Leveraging PCA and Neural Networks |
| title_full_unstemmed | Optimizing Q-Learning for Automated Cavity Filter Tuning: Leveraging PCA and Neural Networks |
| title_short | Optimizing Q-Learning for Automated Cavity Filter Tuning: Leveraging PCA and Neural Networks |
| title_sort | optimizing q learning for automated cavity filter tuning leveraging pca and neural networks |
| url | https://www.epj-conferences.org/articles/epjconf/pdf/2025/11/epjconf_cofmer2025_01006.pdf |
| work_keys_str_mv | AT aghanimamina optimizingqlearningforautomatedcavityfiltertuningleveragingpcaandneuralnetworks AT otmanoulhaj optimizingqlearningforautomatedcavityfiltertuningleveragingpcaandneuralnetworks AT oukairaaziz optimizingqlearningforautomatedcavityfiltertuningleveragingpcaandneuralnetworks AT lasrirafik optimizingqlearningforautomatedcavityfiltertuningleveragingpcaandneuralnetworks |