ETJOURNALoFENGINEERING&TECHNOLOGY

Autumn 2010coNcoIt)

IC])NNNZ(J)(J)Neural Network Based

CAD Model for the Design ofRectangular Patch Antennae

Vandana Vikas Thakare' Pramod Singhal'

'Dept. 0/ Electronics Engineering,Madhav Institute of Technology &Science, near Gola Ka Mandir;

Gwalior. India

INTRODUCTIONAbstract

A new method of calculation of patch dimensions of arectangular microstrip patch antennae using ArtificialNeuralNetwork (ANN)has been adopted in thispaper.An ANN model has been developed and tested forrectangular patch antenna design. It transforms thedata containing the dielectric constant (tr), thicknessof the substrate (h), and antenna's dominant-moderesonant frequency (fr) to the patch dimensions i.e.length (I) and width (w) of the patch. The resultsobtained using ANNs are compared with thesimulation findings. The simulated values areobtained by designing the micros trip antenna usingIE3Delectromagnetic simulator. TheANNs results aremore in agreement with simulation findings.

'Dept. 0/ Electronics andInstrumentation EngineeringAnand Engineering CollegeKeetham, Agra, India

Microstrip antennae due to their many

attractive features have drawn the attention ofindustries for an ultimate solution for wirelessantenna. The existing era of wirelesscommunication has led to the design of efficient,wide band, low cost and small volume antennaewhich can readily be incorporated into a broadspectrum of systems [1,2,3]. This needs veryaccurate calculation of various designparameters of microstrip patch antennae. Patchdimensions of rectangular microstrip antenna isa vital parameter in deciding the utility of amicrostrip antennae. The rectangular microstripantennae are made up of a rectangular patchwith dimensions width (W) and length (L) over aground plane with a substrate thickness (H) anddielectric constants tr.

In this paper, an attempt has been made toexploit the capability of artificial neural networksto calculate the resonating frequency ofrectangular microstrip patch antenna. Neuralnetworks have recently gained attention as fastand flexible vehicles to EM /microwavemodeling, simulations and optimization.Recently CAD approach based on neuralnetworks has been introduced in the microwavecommunity for modeling of passive and activemicrowave component [4, 5, 6 and 7] andmicrowave circuit design. A number of papers[8 to 14] indicate how ANN can be usedefficiently to calculate different designparameters of micro strip antennae. In this work, ••••••••..•..••1the authors extend the work on the use of the~~~

Keywords: Microstrip antenna, bandwidth,simulation,modelling, neural networks, CAD.

artificial neural network (ANN) technique in placeof conventional numerical techniques for themicrostrip antenna design.

II. DESIGN AND DATA GENERATION

As an example, microstrip line feed rectangularpatch microstrip antenna as shown is designedto resonate at 8 Ghz frequency with dielectricconstant (Er)=2 , substrate thickness h=1 mm,L=12.6mm, W=15.3mm. The length and thewidth of the patch are calculated by the givenrelationships. (1), (2), (3) and (4).

W Vo rT ...(1)=2/,~~

where "o is the free space velocity ofthe light.

L= }-:;- 2M ... (2)2/, £~ff

(£ ~+0.3) W +0.264I1L =0.412' h( ) ... (3)h (£~ff-0.258) : + 0.8

where Sl: is extension in length due to fringingeffects and effective dielectric constant is givenby

£ +1 £ -1[ h ]-"2£ =-'-+-'- 1+12-~ff 2 2 W

With the calculated values of various designparameters, the patch antenna is designed for 8GHz resonating frequency. The exact position offeed point can be determined by using IE3DElectromagnetic Simulator. Microstrip antennawith microstrip line feed is shown in Figure 1. Thewidth Wo of microstrip line is taken as 0.5 mmand the feed length is 2mm. The patch isenergized electromagnetically using 50 ohmmicrostrip feed line. The geometry is as shown inFigure1.

... (4)

Figure 1Microstrip Line Feed Microstrip Antenna

IE3D software from Zeland Corporation hasbeen used to calculate the return loss (S11) andhence the cut-off frequencies of the antenna.IE3D software has been used to generate data in

Autumn 2010

the form of cut-off frequencies for 1.5_ Er_ 3.5,1mm _h _2.5 mm, 11mm L 13.5mm and13mm _W _17mm. The generated data werethen arranged in six matrices. The four matricescontaining the values of Eph, f1 and f2 are usedas the input to the network. The other twomatrices containing the corresponding values ofLand Ware the outputs of the network.

---r dB[S(1,1)]

3.------- 13o t-'••.........l--<;~'---" ~ 0

1 j1~ ~!g-12

1

1-12!g-15 -15-18 j -18-21 1 I -21-241 -24-271 1 -27-30 LI---------------'1 -30

6 6.5 7 7.5 8 8.5 9Frequency (GHz)

Figure 2 The return loss (S 11) in dB versesresonating frequency of micros trip antenna

Figure 2.0 shows the return loss (S11) versusfrequency curve for the given physicaldimension. The length, width, substratethickness h and dielectric constant (Er), werevaried for the specified range to see the effect onthe microstrip antenna bandwidth. It wasobserved that antenna performance could becontrolled by varying these parameters to alarge extent.

III. APPLYING NEUROCOMPUTINGTECHNIQUE

In the present work, we have adopted the sameoptimization strategy that is used for error backpropagation algorithm [14, 15 and 16] .Out ofthe 245 data generated, 230 were used fortraining and the rest were used for testing thetrained neural network. The optimized values ofdifferent parameters, which are obtained by trialand error for the training of the network, are: 1)the number of input nodes = 4; 2) the number ofneurons in the hidden layer = 40; 3) the numberof output node = 2; 4) the learning rateparameter = 0:002; 5) and the learning rateadaption = 1:5. Three layer network as shownin figure 3.0 is proposed i.e. input layer, hidden

ETJOURNALoFENGINEERING&TECHNOLOGY

Inputr=. ( Layer 1 Layer 2(

p,R

"---../ "-•...._--_ ...•.•) '-•...._--_ ...•.•)

al = fl(IWI.lp+bl) a3 =f3 (LWl) ai +b3)

Figure 3 Three layer feed forward artificial neuralnetwork.

The various inputs to the network are dielectricconstant (Er) ,substrate thickness h and thecut-off frequencies f1 and f2 and correspondingoutputs are length L and width W of the patch.The training time is 45 seconds and trainingperforms in 1500 epochs. The networkarchitecture preferred in the research work isfeed forward back propagation network. Theback propagation algorithm is used to train theANN which learns using gradient descentmethod. During the training process the neuralnetwork automatically adjusts its weights andthreshold values so that the error betweenpredicted and sampled outputs is minimized.The adjustments are computed by the backpropagation algorithm. The training algorithmused is trainlm .The error goal is .001 andlearning rate is 0.1. The other networkparameters used were noise factor of 0.004 andmomentum factor of 0.075.The transfer functionpreferred is tansig and purelin in thearchitecture. The feed forward backpropagation network architecture is simulatingthe antenna design with high efficiency andaccuracy.

IV. RESULT

A rectangular microstrip patch antenna hasbeen considered. The simulated data of theantenna has been taken from the IE3DElectromagnetic Simulator. Table 1.0 depictsthe comparison of results among ANNs andsimulated values. Network testing was

Autumn 2010

Layer 3 performed for those 15input combinations whichare not included in the setof training data and foundsatisfactory as depicted inthe table 1.0

V. CONCLUSIONS

'\ '\

A neural network-basedCAD model is developedfor the design of arectangular patch antenna,which is robust both fromthe angle of time ofcomputation and accuracy.A distinct advantage of

neuro computing is that, after proper training, aneural network completely bypasses therepeated use of complex iterative processes fornew cases presented to it. The single networkstructure can predict the results for patchdimensions provided that the values of Er f1 , f2and h are in the domain oftraining values.

"-•...._--_ ....••.)

Table 1.0 Comparison of results of IE3D and ANNIMENTforthe calculation of patch dimensions

Er h f1 f2 W(IE3D W(ANN L(IE3D) L(ANN)mm GHz GHz mm mm mm mm

2 1 7.92 8.244 15.3 15.23 12.6 12.34

2.2 1.4 7.32 7.69 15.6 15.42 12.8 12.42

2.5 1.7 6.69 7.06 15.9 16 13.1 12.96

2.7 1.9 6.37 6.77 13.9 14.1 13.3 13.36

2.9 1 6.40 6.57 13.7 13.75 13.5 13.44

2.6 1.8 7.03 7.47 13.2 12.99 12.2 12.36

3 1.2 6.60 6.92 14.4 14.35 12.4 12.53

2.4 1.7 6.70 7.09 15.4 15.60 13.4 13.21

2.1 1.5 8.64 9.02 13 13.1 11 11.4

2.8 1.6 6.28 6.7 15.2 15.13 13.2 13.10

2.9 1.7 6.08 6.45 15.4 15.50 13.4 13.21

2 1.8 7.11 7.47 15.8 15.78 13.8 13.56

2.8 1.8 6.81 7.21 14.8 14.77 12 12.03

2.9 1.7 6.66 7.05 14.6 14.75 12.2 12.14

2.6 1.7 7.28 7.72 13 12.89 11.8 11.72 I~"=lr'ffilf

-

REFERENCES

1. CA Balanis, Antenna Theory, John Wiley & Sons,Inc., 1997.

2. D.M. Pozar, Microstrip Antenna, Proc. IEEE, Vol.80, pp.79-81 ,

3. James.J.R.Hall,P.S.;Wood,c" Microstrip antennatheory and design" Stevenage UK: Peter Peregrines(IEEE),1981.

4. Q. J. Zhang, K. C. Gupta, Neural Networks for RFand Microwave Design, Artech House Publishers,2000.

5. A. H. Zaabab, Q.J. Zhang, M. Nakhla," Analysis andOptimization of Microwave Circuits & Devices UsingNeural Network Models", IEEE MTT-S Digest 1994,pp393-396

6. R. K. Mishra, Member, IEEE, and A. Patnaik ANNTechniques in Microwave technology

7. P.M. Watson, K.C. Gupta, and R.L. Mahajan,Development of knowledge based artificial neuralnetwork models for microwave components, IEEEIntI. Microwave W p . Digest, pp. 9-12, Baltimore,MD,1998.

8. Nurhan rURKER, Filiz G"UNES, rulay YILDIRIM,"Artificial NeuralDesign of Microstrip Antennas",TurkJ Elec Engin,VoI.14,NO.3 2006,© TUBITAK

9. F. Peik, G. Coutts, R.R. Mansour ,COM DEV,Cambridge, ON, Canada," Application of neuralnetworks in microwave circuit modelling", Electricaland computer Engineering, 1998, IEEE CanadianConference, vol-2, 24-28 May 1998, pages:928-931

10. S. Devi , D.C. Panda, S.S. Pattnaik, "A novel methodof using artificial neural networks to calculate input

Autumn 2010

impedance of circular microstrip antenna",Antennae and Propagation Society InternationalSymposium, vol. 3, pp. 462-465,16-21 June2002.

11. J.Lakshmi Narayana, Dr. K. Sri Rama Krishna,Dr.L.Pratap Reddy, Design of Microstrip AntennaeUsing Artificial Neural Networks, proceedings ofInternational Conference on ComputationalIntelligence and Multimedia Applications 2007 ,pp.332-334

12. A. Patnaik, R.K. Mishra, G.K. Patra, S.K. Dash, Anartificial neural network model for effectivedielectric constant of microstrip line, IEEE Trans. onAntennae Propagat., vol. 45, no. 11, p. 1697, Nov. 11997.

13. D. Karaboga, K. Guney, S. Saqiroqlu, M. Erler,"Neural computation of resonant frequency ofelectrically thin and thick rectangular microstripantennas", Microwaves, Antennae andPropagation, IEEE Proceedings-Vol. 146, No.2, pp.

155 - 159, April 1999 January, 1992.

14. Simon Haykin, Neural Networks, second editionpHI.

15. Mohamad H. Hassoun, Fundamentals of ArtificialNeural Networks, New Delhi, Prentice Hall of India,1999, Chapter 8.

16. M. Naser-Moghaddasi, Pouya DerakhshanBarjoei, Alireza Naghsh "Heuristic Artificial NeuralNetwork for Analysing and synthesizingRectangular Microstrip Antenna" IJCSNSInternational Journal of Computer Science andNetwork Security, vol7 No.12. December 2007.

000