IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES,VOL. 42, NO.

7, JULY 1994 1429

[8] EEsof s Touchstone and Libra Reference Manual.

0 DSlSl21 + DB[Sl2I o DB[S12I [9] G. Matthaei, L. Young, and E. M. T. Jones, Microwave Filters,
FILTER IDEAL MEAS Impedance-Matching Nehvorks, and Coupling Structures. Cambridge,
0.000 Mass: Artech, 1980, pp. 83-103.

dB(S21)

Computer-Aided Analysis and Design of

-1 2.00
a Planar Multilayer Marchand Balun

Randal Schwindt and Cam Nguyen

Abstmct-Analytical expressions for the scattering ( S ) parameters for

-21.00 a planar multilayer Marchand balun, constructed of broadside-coupled
microstriplines, are derived for the first time, based on the coupled-
Fig. 7. Measured and predicted responses for the 9th-order filter. ‘‘FILTER” line quasi-TEM normal mode parameters. New design equations and a
depicts predicted response of the fabricated filter. The data marked “IDEAL” new design procedure for the balun based on the derived S-parameters
was computed using ideal inductors having no parasitics. and a circuit synthesis technique are presented. Calculated results for
a GaAs microwave monolithic integrated circuit balun designed by this
procedure are compared to a general full-wave planar analysis, and good
coupling between elements, most likely the open stubs, which would performance and agreement are obtained.
not be predicted without full wave analysis.
I. INTRODUCTION
Iv. SUMMARY AND CONCLUSIONS The balun is an important component for many microwave ap-
In summary, we have demonstrated an accurate and practically plications. It is needed to transform an unbalanced transmission
useful means of deriving models from miniature chip and wirewound line to a balanced transmission line as well as to achieve the
inductors. First, the test fixture is characterized from a set of open- and balanced characteristics in various balanced components such as
short-circuit measurements. The model for the fixture is used to de- mixers, amplifiers, multipliers, oscillators, detectors, etc. In 1944,
embed the effects from the inductor measurements. CAD optimization Marchand presented a design for a balun constructed of coaxial
methods are used to derive accurate parasitic inductor models the transmission lines, which support true-TEM modes [l]. This was
show good correlation to the in-circuit behavior of the devices. The perhaps the first reported balun. More recently, interest in the
inductor models have been shown to be usable as the basis for development of wide-band, compact planar baluns for microwave
accurate filter designs exhibiting first-pass success. These procedures integrated circuit (MIC) and microwave monolithic integrated circuit
can produce microwave-valid hybrid component models necessary for (MMIC) applications has steadily increased. Pavio and Kikel adapted
today’s rapid design cycles. Once a test fixture has been characterized, Marchand’s design for use in MIC and MMIC using planar broadside-
many similar devices can be characterized completely with relative coupled transmission lines but did not present a quantitative analysis
ease. The models generated can be used with confidence. or design technique [2]. Tsai and Gupta [3] presented an analysis
of and design procedure for Pavio’s version of the Marchand balun
ACKNOWLEDGMENT using an equivalent circuit model for the two-coupled-line sections,
based on Tripathi’s quasi-TEM analysis [4], and design curves for a
The authors would like to thank Mike Wyatt, Honeywell Inc., and fourth-order version of the original Marchand balun [5].
Rudy Henning, University of South Florida, for their inputs to the In this short paper, we present a new and efficient analysis and
work, and to thank HP-EEsof for donating software used in this work. design procedure for the multilayer planar Marchand balun. Our
approach consists of two steps. First, we derive analytic expressions
REFERENCES for the generalized scattering (S) parameters of the balun in terms
of the coupled-line quasi-TEM mode parameters, considering the
[l] G.F. Engen and C. A. Hoer, “Tbru-reflect-line:An improved technique
for calibrating the dual six-port automatic network analyzer,” IEEE different c- and 7r-mode phase velocities. Second, we derive design
Trans. Microwave Theory Technol., Dec. 1979, pp. 987-993. equations for the balun based on a circuit synthesis technique and the
[2] “Network analysis: Applying the HF’ 8510 TRL calibration for noncom- obtained S-parameters with the assumption of equal phase velocities.
ial measurements,” HP hoduct 8510-8, Oct. 1987. A systematic design procedure is then suggested in which the balun
[3] A. J. Baden Fuller, “Equivalent circuit of microstrip spiral inductor
generation by computer,” Electron. Lett., Mar. 28, 1985. parameters are obtained through a best numerical fit to derived
[4] D. Cabana, “A new transmission line approach for designing spiral
microstrip inductors for microwave integrated circuits,” IEEE MTT-S Manuscript received July 27, 1993; revised January 21, 1994. This work
1983 Digest, p. 245, 1983. was supported in part by TRW, Inc., and the NASA Center for Space Power.
[5] C. S. Atcbinson, et. al., “Lumped-circuit elements at microwave fre- C. Nguyen is with the Department of Electrical Engineering, Texas A&M
quencies,” IEEE Trans. Microwave Theory Technol., p. 929, Dec. 1971. University, College Station, TX 77843 USA.
[6] CoilCraft Corporation, Cary, Ill. R. Schwindt was with the Department of Electrical Engineering, Texas
[7] W. Gaiewski, L. Dunleavy, and L. &is, “Hybrid inductor modeling at A&M University, College Station, TX 77843-3218 USA. He is now with
L-band using 1-port fixtured S-parameter measurements,” presented at Texas Instruments, Dallas, TX 75265 USA.
the 41st ARFTG Conf. 1993. JEEE Log Number 9402422.

0018-9480/94$04.00 0 1994 IEEE

1430 IEEE TRANSACITONS ON MICROWAVE THEORY AND TECHNIQUES, VOL.42, NO. 7, JULY 1994

design equations. The design procedure, which assumes equal phase

velocities, is exact for the balun realized in a homogeneous medium.
Currently, the procedure typically used in the design of this planar
Marchand balun involves a great deal of trial-and-error using a
full-wave analysis program with relatively little information for a
good initial design; different balun dimensions are tried until a
desired performance is found. This brute-force process can be quite
inefficient given a poor initial design. Our new analysis and design
procedure for the planar Marchand balun will help to greatly increase
the accuracy and efficiency of the design process by yielding an
initial design more accurately and more quickly. Because the theory
neglects some discontinuities and couplings introduced in the physical
implementation and layout of circuits, optimization with a full-wave
analysis may still be required, but the design process and design time
should be significantly improved.
The new design procedure has been used to design a balun for
implementation in MMIC technology, and the predicted performance
compares well to that calculated by a full-wave analysis. This
agreement confirms the validity and usefulness of the theory and
design procedure.

II. BALUNSCATITXI" PARAMFXERS

Fig. l(a) shows the port number convention for the two-coupled
line structure and the interconnections and port number convention
for the planar Marchand balun addressed here. Line 1 is assumed to
be the upper strip in the broadside-coupled microstripline structure
shown in Fig. l(b). The balun S-parameters are derived as

+ cos@,A sin@,a(l - R c ~ / R , ~ )
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 42, NO. 7, JULY 1994 1431

giving an acceptable balun performance. Toward this purpose, we

04 I Line 1B J (open)
will impose certain conditions on the balun S-parameters and attempt
A-Section B-Section
to find realizable mode parameters for the coupled lines which
satisfy them. Furthermore, we will assume equal phase velocities
for the coupled lines in deriving the conditions imposed on the mode
parameters. This assumption results in an approximate design for
the planar Marchand balun immersed in an inhomogeneous medium.
The approximateness, however, is minimized if one uses broadside-
coupled lines so configured that only a small difference in mode phase
velocity exists, as discussed in the design example to be presented
later.
By analogy with standard passive network synthesis techniques,
we will specify the frequency response of the squared magnitude
of the transmission coefficient. The standard synthesis techniques
(b)
for distributed element two-ports use the Richards variable as a
Fig. 1. Interconnections and port number convention for planar Marchand frequency variable which, for lossless networks, has the form S =
balun (a) and the employed broadside-coupled microstripline structure (b).
j R = j tanwllv where I is chosen to be a quarter-wavelength at
the center frequency of the passband. Besides being necessary for
simplifying the derivation, the assumption of equal phase velocities
is required because we will specify the frequency response in S or
R rather than w . The lines for the balun will have physical length
- s i n 6 , ~sin0,A
[Z c ~ Z , ~ ( -R R,A)’
c~
such that the average of the mode phase velocities corresponds to

+ 201202 (2+ 3
1 quarter-wavelength at the center frequency of the passband. We can
obtain expressions for ,521 and s31 from (2) and (3) assuming equal
phase velocities and 2 0 2 = 2 0 3 (typically balun output ports see the
same terminating impedances).
We find that for both Szl and $ 4 1 , the order in n of the numerator
equals the order in R of the denominator which suggests a high-
pass characteristic (in R,implying a bandpass behavior in w ) which
- sinOcB sin6,B
(-ZTB
ZcB
R:~ + agrees with our expectation based on the original Marchand balun
performance. Moreover, S3 1 has a fourth-order high-pass response
and S21 has a sixth-order high-pass response. Thus, the highest-order
- R , B / R ~ B2)0-3 response possible is limited by 1,531 1.’ Because of this, and because
ZcB
we require 5’21 = -S31 for balun performance, we will specify
+ C O S ~ ~sinO,B(l-
B the frequency response of I S ~ and ~ require SZI= - & I .
I ~ then
Specifying the behavior of IS31 1’ will generate a set of equations to
be satisfied by the mode impedances and mode voltage ratios of the
A Z B= 2203 - COS 6 ~ O,BZO~
, COS coupled line sections. Requiring Szl = -Ssl will generate another
set of equations to be satisfied by the mode impedances and voltage
- sin6,B sin0,BZo3 (%+E) ratios, and these constitute the conditions for the design procedure.
For this work, we specify the frequency response of ls311~as
quasi-Butterworth

In the expressions above, all mode characteristic impedances have

been expressed in terms of the line 1 mode impedances, 2, and Z,, K is the parameter to be adjusted for achieving certain bandwidths.
of the two normal modes, referred to as the c-mode and the n-mode. The form of S31 requires that F3(x) be no higher than third-
R , ( R , ) denotes the ratio of the c-mode (n-mode) voltage amplitude order. A higher-order approximating function can be shown to imply
on line 2 to the c-mode (n-mode) voltage on line 1. 6, and 8, are the R,A = R,A, which is physically impossible. It is not known a
electrical lengths of the line for the c-mode and n-mode. Formulas priori if this approximating function is optimal or even completely
for the mode characteristic impedances, voltage ratios, and phase realizable, but it is not obviously nonrealizable as is, say, a true third-
velocities of the c-mode and the n-mode are given in [6] and [7]. order Butterworth response, which is not possible because it implies
Zo;(i= 1, 2 , 3) are terminating impedances at ports i. Subscripts A R,A = R,A. The five design equations are obtained by equating the
and B denote the A- and B-sections, respectively. coefficients of like powers of R
Ai - 2Kkfkg = 0 (13)
III. BALUNDESIGN
The best attainable S-matrix possible for the balun has the form

With the circuit’s S-parameters derived, we will suggest a sys-

tematic design procedure to find the coupled-line mode parameters
1432 IEEE "JSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 42, NO. 7, JULY 1994

RRZ~ -
- -ZcB
k3 +
= ( 1 - R , A / R ~ A ) Z ~ A R(1
: ~- R , A / R , A ) Z , A R ~ ~(20)
+ (~R~BR,B
Z,B
(24)

Obviously, kl # 0, or else the numerator of s 3 1 equals zero,

but the solution must not require kz = 0 , because this implies
that R,A = R,A which is impossible. Because the approximating
function Fn(x)goes as 2 7 , if n > 3, then equating like powers of
!2 yields the equation k i k; = 0 which cannot be physically satisfied.
Four additional design equations are now obtained from setting
S a l = - s 3 1 as follows.

+
ZoiZoz[R,~(l- z ~ A / z m A ) R , A ( ~- ZrA/ZcA)]
.
[
- ZoiZoz[R,~(l- ~ c
1
ZTB
+
(1 - R c ~ / R r ~ ) - (1 - R,B/RCB)-

+
A / ~ T A R
ZcB 'I
) , A ( ~- zrA/zcA)]
R,A - R,A
R c ~ R n ~ ( Z T- ~Z R
c ~T R~ c ~ )
IEEE TRANSACTIONS ON MCROWAVE THEORY AND TECHNIQUES, VOL. 42, NO. 7, JULY 1994 1433

I 2 3 4 5 6 7 8
FREQUENCY (GHz)

~ R , A R , A- & R ~ A
- Fig. 2. Calculated I < Sp1- < 1, IS21 I and IS11I of a designed MMIC
- = 0. (29) balun versus frequency.
~ ~B R Z -B
~ R c B R-
Thus, we have nine design equations and eight mode parameters
to be determined; some redundancy may exist, but this has not been 0 ; 2, = -1232 0; u, = 1.196 x 10’ d s ; v, = 9.29 x
explored. The approach taken in the balun design from this point is 10’ d s ; R, = 0.0105; R, = 1.011. Phase velocities differ by
motivated by concems of the realizability of the mode parameters and 25%.
the complexity of the design equations. Rather than attempt to solve Fig. 2 shows the frequency response of the absolute value of the
the design equations exactly, we have used a spectral domain routine phase difference between the outputs and the frequency responses
to compute mode parameters for ranges values of physical dimensions of IS211 and IS111 according to this theory and according to sim-
of the structure, such as strip widths and strip offset distances. These ulations from the full-wave analysis software program Sonnet [8].
data have been numerically tested in the design equations to obtain The magnitude of ,921 is very nearly equal to that of S B for ~ the
a “best fit.” We are therefore assured of the realizability of the mode whole range of frequencies and, thus, is not shown. The “ideal”
parameters. response is the response assuming the phase velocities are equal
It was found that the last four equations were relatively “easy” to their average, and “calculated” responses are computed from the
to satisfy; that is, the left-hand sides were relatively small in value. exact S-parameters derived here, which take into account the different
However, the five equations specifying the response of IS31I2 had c- and n-mode phase velocities. The difference in mode phase
much larger values. Therefore, in the numerical routine used to test velocities produces the difference between the ideal and calculated
for the best fit, the sum of the absolute values of the left-hand sides performances, as expected. We see good phase response in Fig. 2,
of (13)-(17) was tested to be below a user-specified value, and the as the phase difference varies by less than 2” from the desired
sum of the absolute values of the left-hand sides of (26)-(29) was 180”. The slight discrepancies between our calculated results and
tested to be below a different user-specified value. Sonnet’s can be attributed to the fact that Sonnet accounts for all
discontinuities and coupling mechanisms, such as the discontinuities
due to changes in line widths, fringing at the open-circuited end of
IV. DESIGNEXAMPLE the B coupled line section, and coupling between the ends of the
To verify this theory and show its usefulness in MMIC applica- lower strips at the output ports. In addition, we assume perfect shorts
tions, a balun has been designed for implementation on a 125-pm while Sonnet models the shorts as via holes. Nevertheless, we see
semiinsulating GaAs substrate (E,. = 12.9). The upper dielectric is that our theoretical results agree quite well with Sonnet’s, confirming
silicon dioxide (E,. = 3.9), a good insulator which can be deposited the usefulness of our design procedure for at least first-iteration
reliably in layers as thin as 0.2 pm. For this balun, the thinner the designs.
upper dielectric, the better, due to the tighter coupling between strips.
The upper dielectric layer has been chosen to have 0.75 pm thickness. V. CONCLUSION
The input and output feed lines all have 20 = 50 0. The S-parameters for a planar multilayer Marchand balun, employ-
In order to avoid difficulty in implementation, we require the top ing broadside-coupled microstriplines, have been derived for the first
conductor to be at least as wide as the lower conductor. We also time, based on the quasi-TEM normal mode parameters of the coupled
require, of course, the phase velocities to be close in value. We find lines. New design equations and a new procedure based on the derived
that the phase velocities are similar when the bottom strip is much S-parameters and a circuit synthesis technique have been presented,
wider than the top. Furthermore, the thinner the upper dielectric and and results show good agreement with a general full-wave analysis,
higher its dielectric constant, the more similar the phase velocities verifying the usefulness of the design procedure in the computer-aided
become, due to the tight coupling. This again confirms our intuition design of such baluns. The complete balun S-parameters as well as
that tighter coupling enhances the balun performance and the design design equations and procedure are expected to be very useful in the
procedure’s accuracy. analysis and design of the planar Marchand balun. The theory can
The design procedure results in the top strip of each section also be implemented for other microwave circuits as well.
being centered over the bottom strip, and the A- and B-sections
having identical line widths, presumably due to the output‘ lines’ ACKNOWLEDGMENT
characteristic impedances being equal to the input lines. The balun’s
physical and mode parameters are: top strip width = 1 mil; bot- The authors wish to thank the anonymous reviewers for many
tom strip width = 30 mils; length = 262.4 mils; 2, = 50.1 useful comments and suggestions.
1434 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 42, NO. 7,JULY 1994

REFERENCES geneous medium,” IEEE Truns. Microwave Theory Tech., vol. MlT-23,
pp. 734-739, Sept. 1975.
[ 11N. Marchand, “Transmission-line conversion transformers,” Electronics, [5] J. H. Cloete, “Graphs of circuit elements for the Marchand balun,”
vol. 17, no. 12, pp. 142-145, Dec. 1944. Microwave J., vol. 24, pp. 125-128, May. 1981.
[2] A. M. Pavio and A. Kikel, “A monolithic or hybrid broadband com- [6] C. Nguyen and K. Chang, “On the analysis and design of spurline
pensated balun,” in IEEE MTT-S Int. Microwave symp. Dig.,1990, pp. bandstop filters,” ZEEE Trans. Microwave Theory Tech., vol. MTT-33,
483-486. pp. 1416-1421, Dec. 1985.
[3] C. M.Tsai and K. C. Gupta, “A generalized model for coupled lines [7] V. K. Tripathi, “Equivalent circuits and characteristics of inhomo-
and its applicationsto two-layer planar circuits,” IEEE Truns. M i c m a v e geneous nonsymmetrical coupled-line two-port circuits,” ZEEE Trans.
Theory Tech., vol. M‘TT-40, pp. 2190-2199, Dec. 1992. Microwave Theory Tech., vol. M’TT-25, pp. 140-142, Feb. 1977.
[4] V. K. Tripathi, “Asymmetric coupled transmission lines in an inhomo- [8] Sonnet Release 2.3, Sonnet Software,Inc., Liverpool, NY,1991.