Electromagnetic Modeling of Microstrip Patch Antennas: A Neural Network Approach

This paper aims to elucidate a design procedure that can help solve both the synthesis and analysis parts of an electromagnetic microstrip patch antenna problem. Unlike numerical conventional modeling methods or maybe even analytical methods, artificial neural network are way faster and easier to perform due to their robust correlation and high accuracy. They are also computationally inexpensive when compared to tradition analytical and conventional methods alike. Due to their brevity and highly flexible nature, they are easy to train and work on and can be expanded or modelled in different ways as per the use case owing to their plasticity. Conventional methods are not only time consuming put also require high computational capacity unlike the neural network perspective which has been proposed here. Hence, this paper gives a brief overview of the same and also includes a brief background and basic working of artificial neural networks, activation functions, Radial Basis functions, etc. from both the analysis and synthesis perspective from a RF designers viewpoint. Unlike previous soft computing neural network approaches like back propagation and regression techniques to achieve the same purpose, this paper uses a custom designed Radial Basis Network due to its proven history of computational inexpensiveness and high accuracy and validates its accuracy through the results shown. Circuit design and yield optimizations using a method of moments(MoM) based electromagnetic simulation software was also done to verify accuracy and to feed the ground truth to the network. This neural network was implemented in python due to its vast resource library and ease of implementation.