Design and Optimization of Microstrip Patch Antennas using Artificial Neural Networks

Abstract

Over the past one decade, there is a rapid growth in the development of various applications involving wireless communication. The performance of all such wireless systems depends on the design and proper functioning of the antenna. Microstrip antennas are preferred for majority of these applications. This is because of their inherent advantages, such as small size, planner structure and ease of fabrication. However, for all modern wireless applications, the design of microstrip antenna has become challenging, as several performance parameters, such as return-loss, gain, cross-polarization, side-lobes, etc. are to be optimized simultaneously. Conventionally, for design and analysis of microstrip antennas, methods such as, Finite Element Method (FEM), Full-wave Method of Moment (MoM), Finite Di erence Time Domain (FDTD), etc. are in use. However, these techniques su er from a serious drawback of high computation time and high computational resources. As alternative to these conventional methods, recently, the use of soft computing techniques for design and analysis of Microstrip antennas has increased. The most recognized soft computing techniques are : (i)Arti cial Neural Net- works (ANNs), (ii)Genetic Algorithm (GA), (iii)Fuzzy Logic Models (FLM), (iv)Partial Swarm Techniques (PST). Out of all these techniques, ANNs are most preferred technique for design and optimization of microstrip patch antennas. ANNs are non- linear mapping structures based on the function of the human brain. In the present dissertation report, the analysis and synthesis of rectangular, circular, equilateral triangular and elliptical microstrip patch antennas using ANNs is presented. ANNs based various CAD models have been developed are presented in the report. The feed forward back propagation and radial basis functions have been used to form network architectures of ANN based CAD models. The analysis model consists of antenna geometrical parameters as inputs and resonant frequency and return loss as outputs. The synthesis process is exactly reverse of analysis process. The synthesis model gives patch dimensions as output for a given set of performance parameters. In order to train these ANN models, the training data are obtained theoretically as well as through microstrip antenna full-wave solver. For all CAD models, the results of testing data are compared with the theoretical / simulated results and are thoroughly summarized in the report.