Dielectric Resonator Antenna (DRA) has become very popular in the field of wireless communication and a lot of research has been done in this field in the last few decades. But no closed form relationship is still available relating the dimensions of the DRA or those of feed structures to the performance parameters of the antenna. Various models and numerical techniques are used to predict the performance parameters of the antenna. The disadvantages are that the first approach gives only approximate values and the second one is a time-consuming process. On the other hand, Artificial Neural Network (ANN) is a mathematical model which is used to predict the performance parameters of any complex nonlinear system. Once the model is ready, the ANN predicts very quickly the performance parameters for a given set of input parameters. So two ANN models are developed for DRAs operating at two different center frequencies of the ISM bands. The ISM bands with center frequency of 2.45 GHz (2.4-2.5 GHz) and 5.8 GHz (5.725-5.875 GHz) are chosen as both of them have widespread application, and considering their importance, the DRAs are separately made to operate in these two bands. Both the models are first trained using two different datasets collected through simulations using the software ANSYS HFSSTM (v.13). The outputs of the ANN models are then compared with other two test datasets, and in every case, the error found is negligibly small. The outputs of the neural networks are also verified experimentally. Each of the networks is trained with five input parameters, the three dimensions of the rectangular DRA (length, width and height) along three axes, probe offset position and the probe height. The output parameters consist of resonant frequency, gain and bandwidth.

Dielectric Resonator Antenna (DRA) has become very popular as an effective radiating element in the field of wireless communication. High radiation efficiency because of absence of conductor loss is the main reason for its popularity. Apart from that, it also has a number of other advantages like low loss, light weight, compact size and good compatibility with other structures. Huge application of DRA demands more and more accurate prediction of parameters like resonant frequency, bandwidth, gain, return loss, radiation efficiency. Various models have been used to predict these parameters like waveguide model (Henry et al., 1998), circuit model (Sharshar et al., 2008). But these models give only approximate results. Better results are obtained when numerical techniques like finite element method or method of moments are used. But these are extremely time and resource-consuming processes. On the other hand, Artificial Neural Network (ANN) is a mathematical model used for mapping the input-output parameters of any nonlinear system. So an attempt is made to develop an ANN model for a probe-fed rectangular DRA which can predict its performance parameters with negligible processing time and high degree of accuracy for a given set of antenna input parameters. The application of ANN in the field of microwave engineering, specially to analysis of patch antennas and ring-shaped DRA are already reported (Patnaik et al., 1997; Patnaik and Mishra, 1998; karaboga et al., 1999; Patnaik and Mishra, 2000; Panda et al., 2002; Ouchar et al., 2005; Günes et al., 2006; Narayana et al., 2007; and Pelosi et al., 2011). However, no such attempt is found in open literature for analysis of the simplest form of DRA, i.e., Rectangular DRA (RDRA).

Electrical and Electronics Engineering Journal, Artificial Neural Network (ANN), Dielectric Resonator Antenna (DRA)