2010 International Conference on Computational Intelligence and Communication Systems

Performance Evaluation of ANN Models for the Analysis of Microstrip Band

Pass Filters
Vivek Singh Kushwah Geetam S. Tomar Fanindra Mohan Sharma
Amity School of Engineering & Technology M I R Labs R.J.I.T ,BSF Academy,
New Delhi-India Gwalior-India Tekanpur,Gwalior-India
Email:[email protected] [email protected] [email protected]

Abstract—This paper presents the design and analysis of Jj,j+1 are the characteristic admittances of J-inverters and
Microstrip Band pass Filter at 8 GHz. An artificial neural Y0 is the characteristic admittance of microstrip line.
network model to determine the Magnitude and Phase Assuming the capacitive gaps acts as a perfect, series
variations of scattering parameters (S-parameters) of these capacitance discontinuities of susceptance Bj,j+1, then
filters is also proposed for various frequencies. Performance of
the proposed model is evaluated in terms of average and Bj,j+1/Y0 = ( J j,j+1/Y0) /1-( J j,j+1/Y0)2
maximum estimated errors. Comparison of the EM simulated θj=π-1/2[tan-1(2Bj-1,j/Y0)+tan-1(2Bj,j+1/Y0)]radians (2)
Results with neural models is also presented.
Where the Bj,j+1 and θj are calculated at cutoff frequency.
Keyword- Microstrip Band Pass Filters, ANN models , IE3D The coupling gap sj,j+1 of the microstrip end –coupled
EM Simulation, S-parameters, Training Algorithms resonator filter can be so determined as to obtain the
I. INTRODUCTION series capacitance that satisfy the following equation
A Half- Wavelength resonator band pass filter has many Cgj,j+1=Bj,j+1/ω0 (3)
attractive features, such as compact size, easy fabrication,
Where ω0=2πfc is the angular frequency at the midband.
narrow bandwidth and low radiation loss. Therefore, it
The physical length of resonators are given by
acquires increased applications in microwave circuits [1].
This paper presents the design and analysis of Microstrip Lj=λgo/2π)θj-∆lje1-∆lje2 (4)
Band Pass filters at 8 GHz and proposes an artificial neural
network model to determine the Magnitude and Phase Where ∆lje1,e2 are the effective lengths of the shunt
variations of scattering parameters (S-parameters) of capacitance on the both ends of resonator. Because the
Microstrip Band Pass filters for various frequencies. shunt capacitance Cpj,j+1 are associated with the series
Performance of the proposed model is evaluated in terms of capacitances Cgj,j+1 as defined in equivalent circuit of
average and maximum estimated errors using different microstrip gap, they are also determined once Cgj,j+1 in
neural network training algorithms. equation (3) are solved for the required coupling gaps. the
effective lengths can then be found by
II. DESIGN OF END-COUPLED, HALF-
WAVELENGTH RESONATOR BANDPASS FILTER ∆lje1= ω0 Cpj,1-j/Y0 (λgo/2π)
The general configuration of three-pole (n=3) ∆lje2=ω0 Cpj,j+1/Y0 (λgo/2π) (5)
Band pass filters is illustrated in figure 1, where each The specifications for the filter under consideration are:
open-end microstrip resonator is approximately a half Cutoff frequency (fc ) = 8 GHz,
guided wavelength long at the mid band frequency of the Fractional bandwidth=0.028
band pass filter. The coupling between the resonators is Pass band ripple = 0.1 dB,
through the gap between the two adjacent open ends, and Source/load impedance (Zo) = 50 Ω
hence it is capacitive. In this case, the gap is represented A low pass prototype with Chebyshev response is
by the J-inverters, These J-inverters tend to reflect high chosen whose element values are
impedance levels to the ends of each of the half- g0 = g4 = 1
wavelength resonators and exhibits a shunt type g1 = g3 = 1.0316
resonance. Thus the filter operates like the shunt resonator g2 = 1.1474
whose general design equations are given as follows: The filter is assumed to be fabricated on a substrate of
J01/Y0 = √(π FBW/2g0g1) dielectric constant (εr=10.8) and of thickness 1.27 mm.
J j,j+1/Y0 = (π FBW /2)(1/√gj gj+1) j=1 to n-1 Width w0 of microstrip resonators are calculated which
Jn,n+1/Y0 =√(π FBW/2gngn+1) (1) is 25 mm. The effective dielectric constant (εre), guided
wavelength (λg) and characteristic impedance (Zc ) of
Where g0,g1……gn are the element of a ladder-type low the microstrip can be obtained by desired equations[1].
pass prototype with cutoff normalized at Ωc=1.0, and The Physical length(l1,l2,l3) of resonators and coupling
FBW is the fractional bandwidth of band pass filter. The

978-0-7695-4254-6/10 $26.00 © 2010 IEEE 36

DOI 10.1109/CICN.2010.18
gaps(s01,s12,s23,s34) at 8 GHz are found by using
equations(3)and (4).
l0.= 5 mm
l1=l3 = 5.48 mm
l2=5.86 mm
s01=s34=0.2
s12=s23=1
The layout of the designed filter is given in Figure 1.

Figure 1: Layout of a three-pole microstrip Bandpass filter

III. IMPLEMENTATION & RESULTS

IE3D Simulated results (Magnitude and Phase response)
are shown in figure 2 and 3. The results are satisfactory Figure 3: Phase response of Band pass filter at 8 GHz
and bandwidth is also as predicted.
From magnitude and phase response of Bandpass filter
at 8 GHz ,the magnitude and phase of S-Parameters at 8
GHz are given as.
Magnitude in db, S11= -21.01, S21= -2.093
Phase in Degrees,S11= -179.1, S21= -28.99

Now keeping all other parameter constants, changes the

length of Resonators( l1=l3 = 5.48 mm) and obtained the
different responses for different frequencies.
If l1=l3 = 5.5 mm, then fc=7.9GHz

Figure 2: Magnitude response of Band pass filter at 8 GHz

Figure 4: Magnitude response of Band pass filter at 7.9 GHz

37
 Figure 5: Phase response of Band pass filter at 7.9 GHz
Figure 7: Magnitude response of Band pass filter at 7.5 GHz

If l1=l3 = 5.9 mm, then fc=7.6GHz IE3D Simulated results for calculating the magnitude and
phase of S-Parameters are summarized in the following
table:

Table 1 : IE3D Simulated results

.
IV. ANN MODELS FOR THE ANALYSIS OF MICROSTRIP
BANDPASS FILTERS
ANN represents a every effective modeling technique,
especially for data sets having non-linear relationships.
While developing an ANN model, the architecture of
ANN and the algorithm are the two most important
factors. The ANN architecture used in this paper is
Figure 6: Magnitude response of Band pass filter at 7.6 GHz shown in Figure 8 which consists of an input layer, an
output layer and one hidden layer. It is utilizing the back
If l1=l3 = 5.98 mm, then fc=7.5GHz propagation training.

38
Figure 8: Neural model for calculating Magnitude\Phase of S-
Table 2 : ANN Trained results
parameters of a Microstrip BandPass Filter Training graph obtained after ANN training of samples
for Magnitudes of S-Parameters is shown in figure 9 .
The hidden layer incorporates nonlinear activation
functions, and allows modeling of complex input/output
relationships between multiple inputs and multiple
outputs. Inputs and output are connected by different
sets of weights. Training of the ANN model can be
accomplished by adjusting these weights to give the
required response.. ANN outputs are compared to the
known outputs and then the errors are computed.
Training process ends until errors are lower as much as
possible than the given prescribed values. In order to
develop an ANN model for this filter, a number of EM
simulations need to be performed first. In the ANN
model, the input parameters are Resonator length and
frequency and the output vector is represented in the
form of scattering parameters. The variation ranges of
input parameters are listed in Table 1. The training data
has been obtained in the EM simulation over a frequency
range of 7.15 to 8.4 GHz. S-Parameters obtained after
the training are given in following table:

Figure 9: ANN Training Graph Results for Magnitude of S-Parameters

Training graph obtained after ANN training of phase
samples of S-Parameters is shown in figure 10.

39
 suitable for Microstrip Band pass filter for magnitude and
phase response of S-parameters respectively.
REFERENCES
[1] Jia-Shen G. Hong & M.J. Lancaster, “Microstrip Filters for
RF/Microwave Applications” (1/e),John Wiley & Sons Inc.,2001.
[2] D.M.Pozar, “Microwave Engineering” (2/e), John Wiley, 2000.
[3] V. K. Devabhaktuni, M. Yagoub, and Q. J. Zhang, “A robust
algorithm for automatic development of neural network models for
microwave applications,” IEEE
Trans. Microwave Theory Tech., vol. 49, pp.2282– 2291, Dec.2001.
[4] V. K. Devabhaktuni, C. Xi, F. Wang, and Q. J.Zhang, “Robust
training of microwave neural models,” Int. J.RF Microwave
Computer-Aided Eng., vol.12, pp.
109–124, 2002.
[5] J.Lakshmi Narayana, Dr.K.Sri Rama Krishna and Dr.L.Pratap
Reddy, “ Design of Microstrip Antennas using Artificial Neural
Networks”, International Conference on Computational Intelligence
and Multimedia Applications, SivaKasi,pp.332-334,2007
[6] J.Lakshmi Narayana, Dr.K.Sri Rama Krishna and Dr.L.Pratap
Reddy, “ ANN Modles for coplanar strip line analysis and synthesis” ,
International Journal of Computer Science and Network security,
Korea,Vol.08,No:10, pp:200-204.Oct 2008.
[7] P. M. Watson and K. C. Gupta, “EM-ANN models for microstrip
vias and interconnects in dataset circuits,” IEEE Trans. Microwave
Theory Tech., vol. 44, pp. 2495–2503, Dec. 1996.
[8] J. W. Bandler, M. A. Ismail, J. E. Rayas-Sanchez, and Q. J. Zhang,
“Neuromodeling of microwave circuits exploiting space-mapping
Figure 10: ANN Training Graph Results for Phase of S-Parameters technology, ”IEEE
Trans. Microwave Theory Tech., vol. 47, pp. 2417– 2427,Dec. 1999.
[9] P. M. Watson and K. C. Gupta, “Design and optimization of CPW
Figure 9 and 10 represents the minimization of error as circuits using EM-ANN models for CPW components,” IEEE Trans.
much as possible, so that accurate results are obtained after Microwave Theory Tech., vol. 45, pp. 2515–2523,Dec. 1997.
ANN training .Table 1 and 2 shows the comparison between [10] G. L. Creech, B. J. Paul, C. D. Lesniak, T. J.Jenkins, and M. C.
Calcatera, “Artificial neural networks for fast and accurate EM-CAD
the data obtained from the EM simulation and ANN trained of microwave circuits,” IEEE Trans. Microwave Theory Tech., vol.
data for the filter. Results are very good . Comparison 45, pp. 794–802, May1997
between the ANN model’s output with that of the equivalent [11] A. H. Zaabab, Q. J. Zhang, and M. S. Nakhla, “A neural network
circuit model, can be defined in terms of root mean square modeling approach to circuit optimization and statistical design,”
IEEE Trans Microwave Theory Tech., vol. 43, pp. 1349–1358, June
error : 1995.
[12] J.Lakshmi Narayana, Dr.K.Sri Rama Krishna and Dr.L.Pratap
Reddy “Performance Evaluation of Neural Models for the Analysis of
GaAs MESFET”, GITAM Journal of Information Communication
Technology, Vizag, Vol.01, pp.31-41, 2008
where N is the sampled point number, SijEM and SijMol are [13] Y. Fang, M. Yagoub, F.Wang, and Q. J. Zhang, “A new macro
modeling approach for nonlinear microwave circuits based on
the EM simulated S parameters and the modeled S recurrent neural networks,”IEEE Trans. Microwave Theory Tech., vol.
parameters, respectively. 48, pp. 2335–2344,Dec. 2000.
[14] J. Xu, M. Yagoub, R. Ding, and Q. J. Zhang,“Neural-based
dynamic modeling of nonlinear microwave circuits,”IEEE Trans.
V.CONCLUSION Microwave Theory Tech., vol. 50, pp. 2769–2780, Dec. 2002.
[15] Q.J. Zhang , K. C. Gupta and Vijay K Devabhaktuni, “Artificial
Neural Networks for RF and Microwave Design-from theory to
This paper presents the application of artificial neural Practice” ,IEEE Trans. Microwave Theory Tech., vol. 51, pp.1339-
networks to the design of a bandpass filter. It has shown that 1350, April 2003.
[16] Haykin, S., Neural Networks: A Comprehensive Foundation,
the developed ANN model for the considered filter can be Macmillan College Publishing Comp.,New York, USA, 1994.
as accurate as an EM simulator and also computational [17] M.Makimoto & S. Yamoshita, “Microwave Resonators and
efficient when conducting a massive and repetitive design Filters for Wireless Communication” (1/e), Springer Publications,
analysis. Accurate and simple neural models are presented 2001.
to compute the S-parameters of Microstrip Band Pass filter
for the required design specifications and trained by using
different learning algorithms to obtain better performance
and faster convergence with a simpler structure. It was
shown that the accuracy of the neural models trained by
Feed forward Back Propagation algorithm are found to be

40