Optimization algorithm of neural network on RF MEMS switch for wireless and mobile reconfigurable antenna applications

Abstract

A neural network based feed-forward back-propagation multi layered-perceptron algorithm is presented for the validation of changing the width of upper beam, dielectric and anchor arm length of RF switch designed specifically for reconfigurable antenna. These physical dimensions are varied and design structure is optimizing for low power consumption and to achieve acceptable level of isolation and insertion loss. The algorithm takes these new data samples during training from finite element method (FEM) based simulation tool Ansoft-HFSS. Results of the artificial neural network (ANN) are compared with those of the electromagnetic solver. The developed procedures allows the optimisation solutions of the design to be carried out by replacing repeated electromagnetic simulations whilst still retaining an excellent accuracy as compared with finite element modelling. This procedure requires less simulation time especially for the designing problem, where there is a need of reliable and fully functional methods. The calculated insertion loss and isolation results are in very good agreement with the experimental results reported elsewhere.