Performance Analysis of Machine Learning Algorithms for Microwave Low-Pass Filter Design in Modern Communication Systems

The increasing reliance on wireless networks for data transmission has increased the demand for microwave filters that are characterized by high efficiency, small size, and low cost. This study presents an improved design of a Butterworth low-pass filter that can be used in 5G networks at a cutoff frequency of 3.6 GHz. The design was developed using Advanced Design System (ADS) simulation software using an FR4 dielectric substrate. Modern communications require accurate designs to be provided in less time and effort, and achieve better performance. Therefore, several machine learning (ML) algorithms, such as artificial neural networks (ANNs), decision trees (DTs), linear regression (LRs), and support vector machines (SVMs), were used to optimize LPF designs. A dataset was created that included parameters of the length and width of filter transmission lines at different frequencies. The ML algorithms were then trained by generating their code in Python. The results demonstrate significant improvements in the various algorithms' prediction accuracy and computational efficiency. The ANN (Model 1) achieved the lowest average error of 0.17%, but with the longest training time, while the SVM provided a balance between accuracy and training time with an error of 0.21%. In contrast, DT achieved the fastest training times, but with a higher average error of 2.63%. This approach significantly reduced the need for manual tuning and simulation iterations. These results highlight the potential of machine learning techniques to optimize microwave filter designs, providing engineers with flexible and high-precision tools for modern communications applications