Abhilasha Mishra et al. / (IJCSE) International Journal on Computer Science and Engineering Vol. 02, No. 03, 2010, 626-629 Effect of Normalized Scale on Design of Rectangular Microstrip Antenna by using FFBP Abhilasha Mishra 1 , Vaishali Bhagile 2 , Smita Kasar 3 , P.M.Patil 4 1. Research Student, Department of Electronics, North Maharashtra University,Jalgaon,India. 2. Professor, College of Computer Science and Information Techonology , Aurangabad, India. 3. Lecturer, Department of Computer Science and Engineering, JNEC Aurangabad, India. 4. Professor and Head of Electronics Engineering at Vishwakarma Institute of Technology, Pune, India. Abstract— Nonlinear neural optimization networks are used for the designing of MSA. The paper presents NN model for computing resonant frequency of RMSA with normalized scale. NN model with normalised scale computes very fast. The normalized data has been applied to NN to get more efficient NN model. This model is compared with the NN model without normalized scale which gives very accurate result. Keywords- NN (Neural Network), ANN (Artificial Neural Network), FFBP (Feed Forward Back Propogation), RMSA (Rectangular Microstrip Antenna), EM. (Electromagnetic). I. INTRODUCTION Microstrip Antenna has wide range of applications from communication systems to biomedical systems. MSA consists of a metallic radiating patch on one side of thin dielectric substrate and the other side is ground plane. The resonant frequency of MSA needs to be determined accurately as these MSAs have narrow BW and can operate effectively in the vicinity of resonant frequency [1]. Various methods are available to calculate patch dimensions and resonant frequencies of different geometries of MSA. Mainly two approaches are available for calculation, namely: analytical and numerical methods. The main objective of the present paper is to develop a simple NN model using FFBP algorithm to calculate resonant frequency of RMSA with normalized scale. Once the model is developed, it can be used in place of computationally intensive physics or EM models [2]. II. NEURAL NETWORK DEVELOPMENT Neural network technology is an emerging technology in the microwave area for microwave modeling, simulation, optimization and design [5]. Multilayer Perceptron (MLP), Radial Basis Function (RBF), Knowledge Based Neural Network (KBNN), Wavelet network & Recurrent Neural Network (RNN) are commonly used NN structures. Selection of NN structure and training algorithm are two major issues in developing NN model. The most important and time consuming step in model development is NN training. The microwave behavior is learned through this process. The NN model uses the measured or simulated data for training. Training is an optimization process in the weight space and is often done using optimization –based training algorithm such as backpropogation (BP). Training algorithms are an internal part of neural n/w model development. Any alternative structure may still fail to give a better model, unless trained by a suitable training algorithm [3]. The proper training algorithm manages to reduce the training time by achieving better accuracy. A distinct advantage of neural computation is that, after proper training, it completely bypasses the repeated use of complex iterative processes for new design presented to it. III. NEURAL NETWORK MODEL i) For design of RMSA a FFBP algorithm is used which consists of three layers: input, hidden & output layer. ii) The input and output must be selected which best suits the requirement of the job. iii) Sufficient amount of data must be generated for training either by measurement, simulation or calculation using equations. iv) Execution of algorithm starts the training process of NN model. v) After executing the algorithm the output is compared with the required target. If the outcome of algorithm gives better result then the NN model is tested with new data to calculate the accuracy of the NN model vi) The selection of number of neurons used for hidden layers and value of moment count (mc) in algorithm plays very important role in providing higher accuracy for the developed NN model. vii) If the accuracy is not achieved then by changing the number of neurons for hidden layers and or changing number of epochs, appropriate NN model can be developed as explained in fig.4. viii) If NN model outcome shows over learning/ under learning of target data then the neurons are deleted, or else they are added. In this way by trial and error method the accurate NN model can be developed. IV. DESIGN PROBLEM FOR MSA In the paper [6], the analysis design by FFBP has been presented by considering the ANN model with inputs as ISSN : 0975-3397 626