Makine öğrenimine dayalı 6 ghz altında 5g iletişimi için mikroşerit yama anteninin tasarımı ve optimizasyonu
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Open Access Color
OpenAIRE Downloads
OpenAIRE Views
Abstract
Sub-6 GHz frekans aralığında çalışan beşinci nesil (5G) kablosuz sistemlere yönelik artan talep, küçük boyutlu, geniş bantlı ve yüksek verimli anten çözümlerini gerekli kılmaktadır. Mikroşerit yama antenler (MYA), düşük profilli yapıları ve kolay entegrasyon özellikleriyle avantaj sağlasa da bant genişliği açısından sınırlıdır. Bu tez, 5G uygulamaları için girintili beslemeli yarıklı dikdörtgen MYA'ların tasarım ve optimizasyonuna yönelik makine öğrenmesi tabanlı bir yöntem önermektedir. Geri dönüş kaybı (S₁₁) ile anten parametreleri arasındaki ilişkiyi incelemek üzere 200.000'in üzerinde HFSS benzetimi yapılmıştır. Gradient Boosting modeli, yüksek doğruluk ve genelleme başarısı elde etmiş ve bant genişliği farkındalıklı optimizasyonda kullanılmıştır. Anten 3.56 GHz ve 6.03 GHz'de çift bantlı çalışmış, -18 dB geri dönüş kaybı, %3.2 ve %4.2 kesirsel bant genişliği ile %84 verimlilik göstermiştir. Önerilen ML destekli tasarım yaklaşımı, tasarım süresini önemli ölçüde azaltmakta ve sürdürülebilir, ölçeklenebilir anten tasarımı için güçlü bir potansiyel sunmaktadır.
The significant surge in demand for fifth-generation wireless systems in the sub-6 GHz frequency range requires small, wideband, high-efficiency antennas. Microstrip patch antennas are low profile and easy to incorporate, but are limited in bandwidth and efficiency. This thesis studies a machine-learning-based methodology for the design and optimization of inset-fed slotted rectangular MPAs for fifth-generation applications. More than 200,000 HFSS simulations were used to study the non-linear relationship between the return loss and the antenna design characteristics. A reproducible machine learning pipeline with integrated feature engineering, cross-validation, and hyperparameter tuning ensured robust prediction. Gradient Boosting achieved excellent accuracy and generalization, and was used for bandwidth-aware MPA optimization. The optimized MPA displayed matched bands at 3.56 and 6.03 GHz, an S11 of -18 dB, fractional bandwidths of 3.2% and 4.2%, and an efficiency of 84%. Validation with HFSS simulation showed a near match. The proposed ML-guided design procedure achieves an order-of-magnitude reduction in design time and demonstrates the potential for sustainable, scalable antenna design.
The significant surge in demand for fifth-generation wireless systems in the sub-6 GHz frequency range requires small, wideband, high-efficiency antennas. Microstrip patch antennas are low profile and easy to incorporate, but are limited in bandwidth and efficiency. This thesis studies a machine-learning-based methodology for the design and optimization of inset-fed slotted rectangular MPAs for fifth-generation applications. More than 200,000 HFSS simulations were used to study the non-linear relationship between the return loss and the antenna design characteristics. A reproducible machine learning pipeline with integrated feature engineering, cross-validation, and hyperparameter tuning ensured robust prediction. Gradient Boosting achieved excellent accuracy and generalization, and was used for bandwidth-aware MPA optimization. The optimized MPA displayed matched bands at 3.56 and 6.03 GHz, an S11 of -18 dB, fractional bandwidths of 3.2% and 4.2%, and an efficiency of 84%. Validation with HFSS simulation showed a near match. The proposed ML-guided design procedure achieves an order-of-magnitude reduction in design time and demonstrates the potential for sustainable, scalable antenna design.
Description
Keywords
Elektrik ve Elektronik Mühendisliği, Electrical and Electronics Engineering
Turkish CoHE Thesis Center URL
Fields of Science
Citation
WoS Q
Scopus Q
Source
Volume
Issue
Start Page
End Page
78
