TH3F-F4

Adaptive Frequency Sampling Algorithm for Fast and Accurate

S-parameter Modeling of General Planar Structures

2 2
Tom Dhaene I, Jan Ureel , Niels Fache I and Daniel Dc Zuttcr

I HP-Belgium, Lammerstraat 20. 9000 Ghent, Belgium

2
rNTEc. University of Ghent, Sint-Pietersnieuwstraat 41,9000 Ghent Begiwn

ABSTRACT frequency range is oversampled some important features

could still be missed. So, traditionally some a priori
A new adaptive technique is proposed to represent the knowledge of the dynamiCs of the Circuit parameters is
spectul response of general planar structures over some required in order to select an appropriate sampling
freqUl ncy range of mterest With a minllnal number of distribution for the spectral response of a general planar
frequency samples. Rational fitting nmctions are used to structure.
model and mterpolate the S-parameters obtained
throufh electromagnctic simulation. In this paper, a new sampling scheme IS presented, The
'Adaptive Frequcney Sampling' algorithm (AFS) selects
Thc adaptivc algonthm doesn't reqUire any a pnori the frequency samples automatically in consecutive
knowledge of the dynamics of the S-parameters in order iterations and interpolates all S-parameter data using
to select an appropriate sampling distribution. This rational fitting models, This algorithm allows important
great!\' Improves the transparent lise of any details to be modeled by sampling the response of the
e1ectn .magnetic simuiat0f. structure more densely where thc S parameters are
changing more rapidly. It tries to minimize the total
Ke)'\'I\)rds: adaptive frcqm.:llcy s,unpling, fitting number of samples needed and to maximize the
technique, electromagnetic simulation. information provided by each new sample,

This tool will help the novice user to discover the

1. INTRODUCTION fascinating world of electromagnetic simulators as
accurate full-wave simulations can be obtained much
Thc combination of ovcrsampling and straight line faster without any a priori knowledge of the structure,
interpolation is a standard (but inefficient) approach
used III circuit simulators and electromagnetic
simulators to simulate and represent the circuit 2. THEORY
parameters of any deVice o\'er a frequency range of
interest However, oversampling imphes a waste of All S-parameters of a general planar stmcture are
resources 111-12]. represented by rational fittmg models based on a limited
number of samples, The 'Model Based Parameter
Thc elcctromagnctlc snnulatlOn of complex structures Estimation' technique (MBPE) [3 [ is used to fit a
takes so long that the user often reduces the number of physics based model (pole-zero behavior) to the S­
sample points in order to get the results in a moderate parameters, The MBPE technique can be seen as a
time. If the sampling rate is reduced, many important 'Frequency Domain Prony Method' (FDPM) The S­
features such as coupling effe(ts and n:'sonances may be parameter fitting model is given by
missed (undersampling) Fvell if most I)f the desired

1427

CHH77-4/95/0000-1427$01.00 © 1995 IEEE 1995 IEEE MIT-S Digest

 3. EXAMPLES
(1)
3.1. Semi-circular patch radiator

d The reflection coefficient S 11 of a field-coupled semi­

S(f) (1 + LDi fi) (2) circular patch radiator has been analyzed between 2
i = 1 GHz and 4 GHz using the new adaptive frequency
sampling algorithm and the commercial electromagnetic
where n is the ordcr of the numerator, and d is the order planar simulator HP Momentqm [4]. The cross section
of the denominator. A least square fit is used to and the top view (mesh) are shov.n in figure 2. The two
calculate the coefficients of the rational transfer semi-circular patches of different diameters are
functions. These simple and compact rational fitting indirectly excited by the covered microstrip line. There
models are used for fast frequency interpolation. are two resonance frequencies due to the different radius
of the patch radiators. The AFS technique is used to
Ihe S-parameters S(f) of the planar structures are locate the antenna resonances very fast as opposed to
calculated with the ekctromagnetic simulator HP­ simulating the structure with a densely spaced set of
Momentum [4]. This full wave simulator can handle frequencies.
strips, slots as well as Via's. It is based on the MPIE
formulation [51. The AFS algorithm automatically selects II sample
points (in 10 consecutive iterations) and interpolates all
The flow chart of the adaptive frequency sampling data using rational fitting models. The fitting error is
algorithm is shov.n in figure I . The algorithm starts less than -70 dB. The simulation results correspond very
with two samples at the end frequencies. New samples well with earlier calculated results (2.75 GHz - 3.35
are selected automatically. If the fitting error between GHz) [6] as can be seen in figure 3. The automatically
the S-parameter models and the real data cxceeds a selected sample points are marked by small triangles.
certain limit, then extra sample points are added and Note that the sampling rate is somewhat denser near the
new updated fitting models are calculated This is resonance frequencies of the structure. In figure 4 the
repeated until convergence is reached. The location of simple straight line interpolation (40 samples) is
new frequency samples is found by minimizing the compared to the new adaptive sampling simulation (11
maximum fitting crrors of all S-parameters with respect samples). There is an important speed improvement
to the frequency (factor 4 to 5) due to the fact that far less sample points
are required.
MIN - MAX( fitting error) (3)
3.2. Microstrip lowpass filter
A reliable way to estimate the fitting errors is necessary.
Different criterions can be used and combined. The The cross section and the top view of the lowpass filter
difference between under study are shov.n in figure 5. An edge mesh is used
1 two d the real data and the rational fitting model, for increased accuracy [4]. Note that the microstrip
2 two different fitting models which have a different structure is quite long (as a function of the wavelength).
numerator and denominator order, and/or
two models which use different data samples and/or The AFS algorithm automatically selects 22 samples to
use different frequency ranges model the reflection and the transmission coefficients of
can indicate where the fitting error exceeds an the planar structure in the frequency range: 0.1 Ghz - 20
acceptable threshold and consequently where one or GHz. The AFS simulations correspond very well with
more samples might be needed. The phYSical nature of measurements (0.7 GHz - 20.05 GHz) as can be seen in
the fitting models for the passive planar structures is figure 6. The AFS sample points are marked by small
also checked. triangles.

1428
 4. REFERENCES

1. G. 1. B urke, E. K. Miller, S. C hakr abart i and K.

Demarest, " Usin
g ,I;fodel-Based Parameter Estimation
til Increasethe Efficiency of Computing Electro­
magnetic Transfer Functions", IEEE Transactions on
1\,1agnetics. Vol. 25. No.4, pp. 2807-2809, July 1989. E_, = 1.61
H-15S0 . . feedlin.
2. J Ureel, N. Fache. D. De Zutter and P Lagasse,
E_, = 1.61
"Adaptive Frequency Sampling of Scattering H· 1590.. ground
I"arameters obtained by Electromagnetic Simulation",
[EEE Antennas and Propagation Society - International 17 S nn
Symposium. \/oL 2. pp. 1162-1 1 65, June 1994.
:---
./
3. J N. Brittlgham. E. K. M iller and J. L W il lows. "Pole
Fxtraction from Real-Frequencv in/ormation", I:;'t.
l'\
Proceedings of the IEEE. Vol. 68. No. 2, pp. 263-273,
february 1980. ,
4.313 .
f'. R1
I 1 mm

4. flP Mome ntu m 85158A. HI··-Eesof. Santa Rosa. Ca. k1

II
5. J Se rc u, N. Fache and P. Lagasse, "Mixed Poten tial �
Integral Equation Technique for Hybrid Microstrip­
Slotline Multilayered Circuits using a M ixed 16.5 mm

Rectangular-Triangular Mesh", IEEE Transactions on

Microwave Theory and Techn iq ue s , a cc epte d for
publication. figure 2: Cross section and top view (mesh) of patch
antenna.
6. r'. Alonso-Monferrer, A. Kishk, A Gl i ss on. "Green's
Functions Analysis of Pl anar Circuits in a Two_Layer
G r ou nde d Medium". IEEE Transactions on APS, Vol.
40, No 6, pp. (i90-696, June 1992

I
IA--
0

.. � I
r....
5. FIGURES T
�

-
""
-0

"
I D esign planar structure and s p ec ify frequency range
"
0
-'
-V'
2. P erform electromagnetic analYSIS in new
,-
c

"
+'
"
sample point(sl De:

oJ,.
:< Build fitting m o del and estimate fitting error
'"
oJ,. '"
O�

<1 If ( fitt i ng enor > maximum enor ) I

2. OE +09 FREQ 4.0E'09

Detennine frequency with maximum error AFS : 11 samples 6.
and Repeat 2 &3
figure 3: Reflection coefficient S II of patch antenna.
�. S t op
-&- : AFS simul atio n 12 - 4 GHz]
-
: reference paper 12.75 - :1.35 Ghz]
figure I: Flow chart of Adaptive Frequency Sampling
(AFS) algorithm

1429
 100.0E+06 FRFQ 20.0E+09

"
"
"
...J
c
'- N
"
+'
(n
III
'"

2 . 0E+09 f req 4. OE + 09 100.0[+06 FRF:O <'0.OE+09

L-____________________________

figure 4: Reflection coefficient S L L of patch antenna. figure 6: Reflection and transmission coefficient of
a) Adaptive Frequency Sampling: 11 samples lowpass filter.
b) Straight Line interpolation: 40 samples � : AFS simulation 101 - 20 GHzl
(AFS) algorithm. : measurements [0.7 - 20.05 Ghz]

E_f "" '1.3

H =- 63_, _'"__::.
9 r_o_
. un_d

12 mIl 67 mi 1
63 mi I

124 mi

?4 mi 1

figure 5: Cross section and top vkw (mesh) of lowpass

filler.

1430