Using Artificial Neural Network to Model Microstrip
Inset Fed Rectangular Patch Antenna
Vipul Sharma, Member, IEEE, (09412352583),Gaurav Kumar Gupta Dept. of Electronics & Communication Engineering, Gurukul Kangri University, Haridwar-249404 (India) vipul.s@rediffmail.com, g_gaurav28@yahoo.co.in
ABSTRACT: Artificial neural networks (ANNs) techniques are recently indicating a lot of promise in application for various microwaves engineering field. Such a use of ANNs for estimating input impedance of Rectangular microstrip patch antennas is presented in this paper. The performance of the neural networks based estimators has been compared with the theoretical values and excellent performance is indicated. Neural networks based estimation has the usual advantage of very fast & simultaneous response of all the outputs.
I. INTRODUCTION Microstrip Patch antennas are increasing in popularity for use in wireless application due to their low profile structure. Therefore they are extremely compatible for embedded antennas in handheld wireless device such as cellular phones, pagers etc as they can be easily integrated with microwave integrated circuits. The telemetry and communication antennas on missiles need to be thin and conformal and are often Microstrip patch antennas. In microstrip antennas the feed line impedance (50) is always the same as the resistance at the edge of the patch, which is usually a few hundred ohms depending on the patch dimension and the substrate used. As a result this input mismatch will affect the antenna performance because maximum power is not being transferred. Thus input impedance of rectangular microstrip patch antenna decides the matching between feed line and patch. Resistance of the patch varies as a cosine squared function along the length of the patch (According to Transmission line theory [1], [4]). Maximum value occurs at the edge of the patch where the voltage is maximum and current is minimum (typical values are in the 150-300 ohms). The minimum value (zero) occurs at the center of the patch (=L/2) where the voltage is zero and current is maximum. As the feed point moves from the edge towards the center of the patch, the resonant input impedance decreases monotonically and reaches zero at the center. When the value of feed point approaches the center of the patch (=L/2), the square cosine function varies very rapidly; therefore the input resistance also changes rapidly with the position of the feed point. To maintain very accurate values, a close tolerance must be preserved.
Top View
Figure 1. Inset Feed Patch
An error in feed point location of .01L would yield a deviation in resonant resistance of the order of 10. Many methods have been developed to achieve such high design accuracy. But these involve high cost experimental infrastructure and experimental time. In this paper we emphasis on the validity of neural networks for calculation of input impedance of rectangular microstrip patch antenna up to a much higher accuracy.
II. ANALYSIS USING NEURON MODEL: Neural network or artificial neural networks (ANNs) [2] are the interconnection of artificial neurons that tends to emulate the human brain. The network has the capability to learn because of the distributed intelligence or associative memory property. Neural networks are fast in a parallel and distributed manner compared with slow sequential computation in a conventional von-Neumann computer that uses a centralized CPU and central memory. Also the neural network has fault tolerant properties and provides noise immune computation due to distributed knowledge throughout the network. We have used ANNs to evaluate the accurate feed point impedance. A feed forward multilayer architecture is used. And the Back Propagation Algorithm is used to train the feed forward neural network.
Figure 2. Structure of a Feed Forward Three Layer (1-50-1) Neural Network Showing Back Propagation Training
A three layer1-30-1 network is used for the specific problem. About 10 numbers of patterns are taken for training the network. Other parameters considered in the network are as follows: Learning rate=0.1 20 patterns have been tested to see the accuracy of the neuron model. Figure 2. shows the network structure used.
III. RESULT A rectangular microstrip patch antenna having length (L=30.36 mm) and width (W=40.61 mm), substrate used is RT Duroid 6010 which has a dielectric constant of 10 and a thickness of 0.508 mm at a resonating frequency of 1.575 GHz is designed using transmission line theory [1]. And the result for the normalized input impedance thus obtained is used as the training data. The theoretical values obtained by the transmission line model are used to compare the results. Figure 3. reflects the comparison made among the results obtained from ANNs and theoretical values. Performance of the neural network based estimators was found to be excellent
IV. CONCLUSION
Feed Point Location (m)
Figure 3. Comparison of Normalized Input Impedance (Theoretical and ANNs result)
The Paper successfully demonstrates the validity of feed- forward neural networks for the estimation of normalized input impedance of Microstrip Patch Antenna varying with the feed point location. The neural Network estimates theoutput (in present case normalized input impedance) from the known feed point location. We used a Neural Model Developed in C++ as Artificial Neural Network and MATLAB to plot the results. After a large no. of training steps estimators accuracy was found to be excellent. ANN application was simple, inexpensive and more accurate. The estimation principal can be extended to more complex application in Microwave Engineering field such as finding radiation pattern. Estimator neural network chips can be easily integrated with DSP and/or ASIC chips.
REFERENCE [1] Keith R. Carver, Member, IEEE, James W. Mink, Member, IEEE, Microstrip Antenna Technology IEEE Transactions on Antennas and Propagation, Vol. AP-29, N0. 1, January 1981. [2] Shyam S Pattnaik, Dhruba C. Panda, and S.Devi, Radiation Resistance of Coax-Feed Rectangular Microstrip Patch Antenna With the use of Artificial Neural Networks Microwave and Optical Technology Letters, vol.34, No.1, July 5, 2002. [3] Shyam S Pattnaik, Dhruba C. Panda, and S.Devi, Input Impedance of Rectangular Microstrip Patch Antenna Using Artificial Neural Networks Microwave and Optical Technology Letters, vol.32, No.5, March 5, 2002. [4] K.R. Carver, A modal theory of expansion for the microstrip antenna in Dig, Int. Symp. Antennas Propagat. Soc., Seattle, WA, June 1979, pp. 101-104. [5] R. E. Munson, Conformal Microstrip Antennas and Microstrip Phased Arrays IEEE Transactions on Antennas and Propagation, Vol. AP-22, N0. 1, pp. 74-77 January 1974