IbEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL.

MTT-26, NO, 3, MARCH 197% 175’

Expefin~ental Study ofllymmetric Microstfi~g

Bends and Their Compensation
RENE J. P. DOUVILLE, MEMBER. IEEE, AND DAVID S. JAMES, MEMBER, IEEE

Abstract–-An experimental study of microstrip bends is described.

For right-angle corners, equivalent circuit parameters covering a
wide range of frequencies, linewidths, and permittivities are pre-
~r
sented and compared with the limited theoretical values available.
w
‘he influence of mitering or rcnunding of the corner was investigated, *
H
and useful empirical expressions are given for the optimum miter. (0) (b) (c) (d)

I. INTRODUCTION

LTSCHULER AND OLINER [1], [2] have in-

A vestigated right-angle
ically, using a Babinet
bends in stripline both analyt-
equivalence to waveguide, and n

4
experimentally. Campbell [3] extended the analysis to the
special case of a 45’,50 percent miter. Silvester and Benedek
Flg 1. (a) Mlcrostrlp geometry. (b) Right-angle bend. (c) Mitered bend.
[4] have numerically evaluated the excess capacitance, and (d) Round-reference loop. (e) Square loop w,th movable probe. (f) Bend
Gopinath and Easter [5], Horton [6], and Thomson and equivalent circuit. (g) Alternative lumped-element equivalent circuit.
Gopinath [7] the excess inductance associated with the
microstrip right-angle bend. In all cases, the values obtained
fields. Therefore, it is possible for the equivalent circuits to
are quasistatic only. Wolff and Menzel have used mode-
contain negative elements. Originally, it was planned to
matching techniques to compute frequency dependent S-
include allowance for radiation from the bends in the form of
parameters for an unsymmetrical right-angle bend [8]. No
a resistive element. However, preliminary measurements
theoretical results are available for designing matched
indicated the radiation contribution to our measurements to
bends.
be essentially negli@ble. Again this should not be construed
To date. only limited experimental results have been
to mean that under all circumstances the radiation asso-
reported, most notable being those of Easter et al. [9]-[1 1],
ciated with the bend will remain insignificant. 1
and Kelly et al. [12], all for the special case of lines on
substrates having ~, = 10.
11. EXPERIMENTAL METHODS
For these reasons, measurements were undertaken to
more adequately characterize the symmetric microstrip A. hsonoill J!.00P

right-angle bend over a wide range of useful impedances, The method used in the majority of the measurements was
substrate permittivities and frequencies. The possible im- first described by Douville and James [15], and later used by
provements provided by geometrical modification to the Hoefer and Chattopadhyay to evaluate shunt lposts in
outer portion of the right-angle bend. including a 45” miter, microstrip [16]. It may briefly be described as follows. A
were investigated. The miter geometry which minimizes the standing wave pattern is excited on a square microsl.rip loop
bend VSW17 was determined over a useful range of practical resonator (Fig. 1(e)). Very light coupling (P < 0.05) is used,
parameters. thereby minimizing any influence of the coupling strip cm
Fig. 1(a)-(c) define those parameters of the microstrip and the loop-resonant frequency. With the launcher at midpoint
bend which will be discussed. Although the lumped-element and for all even harmonic resonances. voltage nodes or
equivalent circuit of Fig. 1(g) gives more physical insight antinodes are located simultaneously at all four corners. If
into the nature of the discontinuity y, that of Fig. 1(f) is of voltage nodes exist at the corners, the voltage across B’ is
more practical convenience in circuit design. The reference zero, and thus 1?’ has no effect on the total electrical length
planes T1 and T2, taken to be at the inside edges of the and resonant frequency of the loop. If the coupling point is
microstrip lines, do not completely enclose the discontinuity moved off center about one quarter wave such that a voltage

Manmcrlpt reccwed May 9, 1977; rewsed September 30, 1977 1 Some Idea of the magmtude of the radiated power J$’,d for a 90’ bend
R. J, P, Douwllc IS with the Commumcatlons Research Centrc, Depart- may be obtained from Lcwin [13] ah modified by Hammerstad [14];
ment of Communications, Ottawa. Ont., K2H 8S2. Canada. Pr.d/P,n. - 0,07[fi,mm) .f&Hz)]2/Z0 Ceff. where Pin. IS the power mcldent on
D S James was with the Commumcatlons Research Centre. Ottawa. the corner and the other symbols have their usual meanmg. For the fairly
Ont . Canada He M now with Ferranti Sohd-State Microwave, Man- extreme example of a 25-Q hne on a 0.635-mm substrate with 8, = 2.5, the
chester, England. loss would be expected to be about 0.35 dB at 12 GHz.

0018 -9480/78 /0300-0175 $00.75 @ 1978 IEEE

176 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL MTT-26> sO. 3, MARCH 1978

antinode exists at the corner, B’ has a field maximum across ments with a network analyzer. Most of these tests were
it causing a reduction in the loop-resonant frequency. undertaken on 0.635-mm thick substrates.
By measuring these two frequencies, the value of the
111. RESONANT-LOOP EXPERIMENTS
normalized discontinuity susceptance B’ may be computed
as All experiments were performed on oversize 30 x 30 x
B 0.5-cm substrates. The errors encountered by Stephenson
B’=— (1)
Yo(j.) “tan(:”w) and Easter in their tests and attributed to dimensional
where inaccuracies aggravated by the very small dimensions of
conventional microstrip were minimized by scaling up in
L frequency with a node at the corner, this fashion. As a consequence, the measurements were
.ta frequency with an antinode at the corner, performed over the frequency range 0.2-3 GHz, so that, by
k harmonic number (only even values used), suitable resealing. the data may be applied to the much
Yo(fn) the characteristic admittance at fn, higher frequencies generally encountered in microwave
B unnormalized discontinuity susceptance. integrated circuits. For example, for a typical alumina
To obtain the equivalent electrical length of the corner, it substrate of thickness 0.635 mm, the scale factor is 8,
is necessary to measure the resonant frequency j, of a effectively extending the results to 24 GHz.
circular loop fabricated on the same substrate and having a The substrates were mounted in a large metal box. The
mean circumference equal to twice the length of one side of type N launcher, mounted on a slider, was positioned along
the square loop (Fig. 1(d)). The value of A/ may be obtained one side of the square resonator such that either one or the
as other of the two resonances was suppressed. To avoid
the possibility of a changing physical environment

where
/j/=[c

()
ti

dfn
(2)
influencing
ments reported
the measurements, most B’ and AF measure-
here were obtained with the box completely
closed. The inevitable box and surface modes were sup-
A/ half the total bend equivalent length between planes. pressed by judicious use of absorber material.
T1 and Tz, Preliminary measurements of the round loop-resonance
j round loop-resonant frequency, frequency revealed that Stycast2 having an s, of about 10
/, mean circumference of round loop. possesses a temperature coefficient of capacitance of about
500 ppm~C, so that a change of only a few degrees Celsius is
The element values appropriate to the equivalent circuit of
accompanied by a measurable change in frequency. For
Fig. 1(f) and those for the lumped-element equivalent circuit
purposes of calculation of B’, this change was inconsequen-
of Fig. 1(g) maybe compared by equating the corresponding
ABCD matrices, thereby leading to the following tial. However, since the frequencies used in obtaining AZ are
taken after a fabrication cycle, changes of only a few degrees
expressions:
Celsius between measurements on the square-loop and
B’ _ (B’l – sin 26) s’ f3J+0 ~,_ ~A8CW round-loop resonators may have a substantial influence. In
(3)
27rf – 2rr,j — order to minimize this effect, all measurements were per-
formed at a controlled temperature of 19 t 2°C.
LC = 2A fL,T, (4]
where To Justify the use of the arithmetic mean circumference of
the round-loop resonator in the determination of A/, the ,s.Ff
e 2nA{/2g = (2A[/H)(7Lllj&/c],
determined using a 50-Q linear resonator constructed on the
Cm,Ln microstrip capacitance and inductance per
same Stycast substrate as a round one was measured and is
meter,
shown in Fig. 2.3 The effect of the round-loop curvature is
B; Coedz~ (j’ ).
obviously negligible. It was also verified both experi-
For this discontinuity, the second term in (3) typically is not mentally and analytically that for the coupling gap used, the
small in comparison to the first. The designer is therefore influence of the coupling capacitance on the B’ and A/values
cautioned against simply estimating the expected mismatch was negligible.
at frequency f from a normalized susceptance equal to This method, similar to that described by Easter and
27cfcdZo(f ). Stephenson [9] minimizes the problems encountered in
measuring the reflection and transmission properties of
B. Meander Line discontinuities through transitions or launchers. The char-
Some results were additionally obtained by use of swept acteristics of such transitions are generally poor and
frequency return loss measurements on circuits consisting of
a number of bends arranged in meander-line fashion. With
2 Emerson and Cummg, Inc.
half-wavelength spacing between bends it is possible to 3 The 20 values mdlcated throughout were obtained using ZO(j ) =
obtain reasonably accurate data for the shunt susceptance of ZO(&eff ~/c,rf) where Seff ~ was obtained using Hammerstad [17] and E,ff IS
the discontinuity, but not for the corresponding reference that measured and appropriate to X-band and 0.635-mm substrates. Note
that although these values of Zo(j’) do not have any substantial effect on
plane shift A~. For small bend VSWR’S, the sign of the shunt the results, care should be exercmed in using them in an absolute sense as
susceptance was verified by means of impedance measure- suggested by Schmitt and Sarges [22],
DOUVILLE AND JAMES: STUDY OF SYMMETRIC MICROSTRIP BENDS
177

FREQuENCY (GHzl S,=108. WfH =09

0 5– MITRE =
0
~~

0 4— .
730
4 - 7“” . ROUNDED OuTER, RADIuS W

720 0 3–
r

710 0 2–
i
4, .
.
— –2,00 ; 01—
M/4
N

, //
~.-– -u
/’
,
(,) REXOLITE (6,=2 601 ROUND-COVER OFF
(lo ,’ !,
(?11)STYCAST (6,-10 2) ROUND
,, ,, ON
$$ OFF
.O,j ’4’
,/’;l. ) ($”) $$ $$ $, $$ ON
6.60 (v) $ LINEAR- $, ON
i / f,

\ 2A~ = 200/0 OF kg/4

\
6.50 1 ! 1 1 1 I I [ , , I I
I I ! 1 I , [ I \
0 I 6,= 2.6, W/ H= 275
0.5 1.0 1.5 2.0
FREQUENCY (GHz)
1,6
MITRE =
Fig. 2. Measured frequency dependence of C,ff for round and linear 50-f2
(nom.) resonators on Rexolite and Stycast, H = 5,08 mm, [,4

: 1.2
:

1.0
unrepeatable. However, the resonant loop method used here
has the advantage over that used by Stephenson and Easter 0,8
I \
in that it eliminates the necessity of interpolating between
~-
alternate harmonics in order to calculate B’. This interpola- 30
f (GHZ)
tion procedure is rendered inaccurate by virtue of the
strongly dispersive behavior of microstrip. In the method
Fig. 3. Frequency dependence of normalized bend electrical length for
used here, both ~n and ~. are measured at every harmonic,
various miters. Lines are 50 Q (nom. ) on 5,08-mm thick Stycast (a), and
thus eliminating the need for any interpolation. In fact, by Rexolite (b),
interpolating between even harmonics, it is possible with
this method to obtain B’ values at the odd harmonics as well.
An attempt to do so, however, demonstrated that the measured dispersive Eeff curves indicate a definite frequency
interpolation process is sufficiently inaccurate to introduce dependence of the lid proximity effect, the correction has not
significant scatter in the results. been incorporated into the curves of Fig. 3(a).
Since the computed effects of the proximity of the lid and
IV. RESULTS sidewall on C.ff are quite small, it is considered that the
discontinuity fields and therefore the B’ and A/ values will
A. Bend Equivalent Electrical Length
not be substantially influenced.
Fig. 3(a) and (b) illustrate curves of normalized A# It maybe noted that the bend electrical length can become
measurements as functions of frequency and percentage a significant fraction of a quarter wavelength, as graphically
miter for nominally 50-Q lines on Stycast of 8, = 10.8 and illustrated by the dashed curves in Fig. 3(a) and (b). Fig. 3(a)
Rexolite of c, = 2.6. Similar curves were obtained for 33-Q and (b) also reveals that a commonly used practice, that of
and 76-Q lines on 8, = 10.8 as well as for43-Q lines on t, = 5. rounding the outer corner of the bend, has essentially no
Using a finite-difference computer program as well as a effect on Al, as expected. Additionally, it is apparent that
program based on Smith’s analysis [19], the effect of the substantial mitering may take place before the bend electri-
proximity of the sidewalls and lid on the quasistatic value of cal length is greatly influenced.
C,ff was calculated. In Fig. 2 the measured effect of the lid The measured frequency dependence of Al appears to be
(5.5H above the substrate) is apparent and, in the low more pronounced for c. = 10.8 than for the lower 6,’s,
frequency limit, in good agreement with that predicted. The although at least some of this dependence maybe attributed
computations also indicated that in most cases, even though to the proximity effect described above, Although some
the sidewall proximity is much greater for the square loop possible reasons for the cyclical nature of the data of Figs. 3
than for the round, there k no substantial effect on the A/ and 4 have been investigated, the cause is not understood,
values if the lid is installed. However, in the sole case of 50-Q and therefore, to facilitate ready comparison with published
lines on c, = 10.8, a change in Ecff of approximately 0.7 results, the dotted line approximations to the data have also
percent from that with the sidewalls far removed was been drawn.
computed. Although this has no effect on the B’ values, it
suggests an upward adjustment of the 2A//H curves in Fig. B. Bend Susceptance
3(a) by + 0.15. Since the frequency dependence of this Fig. 4(a) and (b) shows the measured values of the
correction could not be readily ascertained, and since the normalized bend susceptance for nominally 50-Q lines cm
178 IEEE TRANSACTIONS ON MICROWAVE THEORY A?+D TECHNIQUES, VOL. Mrr-26, IWO. 3, MARCH 1978

330
,, =10.8

G, = 10.8, W/H = 0.9

/
.05– 2.5–
—MITRE :0%

———wOLFF [21] ,
G, =9.7. w/ H= O.9Z (5011 NOM.)

.04-
XOUO cc
3 2,0—
$

03– “~4~ER
x
‘\ . 50%
. l.5—
\
x
02- .

E \ ., =10.8
z . \\ , x
60 “/. ROUNDED, RADIUS=4U, 48 S)
xx 6,=10.8
: 1.0 .
x I , , 1 ! , , 1 i i I 1 I 1 I ( 1 1 I , 1 1 I
.: 0.5 Lo 15 2.5
x -. ~vSWR=l.1 0 2.0
m 01—
f (G HZ)
x
Fig. 5. Variation of VSWR with frequency for unmitered 9P bends
. . . (~= 1.5 GHz).
.
0
-----7= I .0 1.5 2.0
f (GHz) 6, = 10.8 and 2.6 as functions of both frequency and percen-

.
tage miter. Again, similar curves were obtained for other
x 80 “A
-.01 –
x
geometries and 8,’s. As expected, as the miter percentage is
x
x increased the bend susceptance eventually becomes induc-
x ,/
~ tive; the point at which this transition takes place is the
/
/ optimum miter. In these and all other cases studied, substan-
-.02– x

tial mitering (- 50 percent) is required before any

appreciable reduction in the bend susceptance occurs. Also
from Fig. 4(a) it can be seen that rounding the outer
08
MITRE = perimeter of the bend yields little improvement over that of
0 %
the unmitered bend. In all cases examined, the frequency
er = 2.6, W/H= 2.75
.07
dependence of B/j27 is pronounced only for the unmitered
case and is almost totally eliminated as the optimum miter is
.06
approached. Curves of constant VSWR = 1.1 are super-
vS~R=l.1
. imposed on Fig. 4(a) and (b) and illustrate that even for
\
.05 .
frequencies as high as 1.5 GHz (12 GHz on 0.635 mm), the
\
11 \
tolerance of the optimum miter is quite substantial.
.04 \ The importance of mitering is graphically illustrated in
\
Fig. 5 where the VSWR of the unmitered bends is illustrated
i \
for several E, and Z. cases studied. In many practical cases,
the mismatch can be seen to become severe at higher
frequencies. Although not always convenient, a 90° arc can
\ . sometimes be used satisfactorily as an alternative to an
I
‘___ abrupt bend, and Fig. 5 includes performance obtained with

1
.01
-—_
.
50% a relatively large arc of mean radius 4H in a 48-Q line on
.
8. = 10.8.

C. Comparison with Published Data

/ For the electrical length component, Fig. 6 shows that

-.02 /
/
1 / reasonably good agreement is observed between measured

L--———————
-.03
“ 70%
data extrapolated to zero-frequency and the theoretical
.
// quasi-static data of Thomson and Gopinath [7]. The data
-.04 points for the dotted curve represent similarly extrapolated
Fig.4. Frequency dependence ofnormalized bend susceptanceforvar- values associated with the optimum miter; the optimum
ious miters. Lines are 50 Q (nom. ) on 5.08-mm thick Stycast (a), and
miter will be shown in Section V to be dependent on WJH
Rexolite (b).
only. For the 50-percent miter case the A~ results derived
from Campbell’s curves [3] along with the Babinet equiv-
lJOUVILLE AND JAMES: STUDY OF SYMMETRIC MI CROSTRIP BENDS
179

TABLE I
COMPARISON OF MEASURED AND CALCULATED VALUES FOR B

f-1.5 GHz3$6

~r W/H z.(f) tiPwMl B&S2 Exp’t

(0)
———-—=.= -——
1
2.6 2.75 50 5.4 5.1 4.6 49 5.5 5.Z 5.3
5.0 2.15 44 5.7 5.2 5.3 43 6.5 6.0 6.5
10.1 0.88 52 5.3 4.7 4.2 50 5.3 4.7 4.6
10.8 0.9 50 5.5 5.0 3.9 48 5.6 5.1 4.9
10.8 1.75 35 7.6 7.1 6.3 33 8.0 7.5 B.3
10.8 0.28 78 3.3 2.8 1.9 76 3.4 2.’3 2.3
15” 0.48 55.5 4.8 L.2 -. 52.5 4.8 4.2 3.?
255 0.33 49.5 5.2 .— 46.5 5.4 -- 2.7

1 x 10IJ
I [B’/f(G~z)H(m)
.J
Notes: 1. Ref. I-llj
2. “ [ 41
3. From (3) k (4) Al obtained from [7 1. Resealed for H as per notes 4 or 5 for f + g.
4. H = 0.635 mm
5. H = 1.52 mm
6. Z.(f) & .,ff (f) computed from Hammerstad [ 17] and Getsinger [ 18].

1,4—
component B’ which are in approximate agreement with the
measurements, as shown in Table I. Note that there is
A greater discrepancy between theoretical and measured data
I 2- OPTIMALLY .“
MITRED .“
for narrow lines than for wide lines. This may be a con-
sequence of the greater dependence of B’ values on At for
small W/H and therefore a higher reliance on the accuracy of
1.o- .’
.“ /uNMITRED the theoretical AI? values. Easter’s measured data (Fig. 6)
.“ suggest that the magnitude of the theoretical AL’ values may
//”
in fact be unrealistically high for narrow lines (W/H < 1).
0.8– “A
.“ For the 50-percent miter, susceptance values derived from
!“ /’

A ..” Campbell’s results together with the Babinet equivalence

.“ /’
between parallel-plate guide and microstrip again show
0.6— .“
.“
// poor agreement with the measurements.
.“
‘/
n The frequency variation of the equivalent circuit par-
.“ ameters is measurable, although the magnitude of the
o.4– .“ “/ a
1 .“ / variation re~orted here is typically somewhat greater than
s .“

: .“ that reported by Easter [i ~]. The only theo~etical data

.“
0.2- .“ available for the dynamic case is that due to Wolff [21] for
.“
.“ the 50-Q bend on alumina, where agreement with measure-
i ment is poor (Fig. 5).
..” /
.“ x /
0 I
1 1 I I I
I
I I I I 1 V. THE OPTIMUM PERCENTAGE MITER M
0.5.’ 0 1.5 2.0 2.5 3.0
....” . w/H
...
A. Right-Angle Bend, 45° Miter
-0.2 ... THEORY, THOMSON fk GOPINATH [71
{/- Fig. 7 indicates the variation in the optimum percentage
A,/ 0 A ~xj;,~ y&pRyl; f+ (j

/ x
} miter M as a function of linewidth for e, values in the range

-0.4 J /0 2.5 to 25. It is remarkable

independently
how closely the various points
ofs, appear to lie on a single smooth curve
Fig. 6. Variation of normalized bend electrical length with linewidth,
described by the following empirical expression:

alence between parallel-plate guide and microstrip [20] are M = 52 + 65 exp (– 1.35 W/H),
quite misleading.
W/H >0.25, E, <25. (5)
Table I presents a comparison of the measured suscep-
tance results for the O-percent cases with theoretical data For wide lines, M appears to reach an asymptotic value of
available. The simple expression Cd = Cm W, particularly approximately 50 percent, although there does not appear to
convenient for practical circuit design, may be obtained by be any obvious physical explanation for this. Over the whole
use of the simple uniform plane wave model(UPWM)[11]. range of parameters investigated the commonly used 50-
Such computed results, when used in (3) along with the percent miter never results in over-compensation, invariably
2A//H values of Fig. 6, yield values of the shunt susceptance providing a better match than for the unmitered case.
180 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. MTT-26, NO. 3, MARCH 1978

1080
\, WI H (q

25 .
2.75 (2.6)
\
\ ,\ *
●
.

‘, \ 1.2 / t *

,5 0 RESONANT LOOP
\ ‘\ MEASUREMENTS
8 MEANDER VSWR }
., =,3,.
‘–-– LIMITs ON M FOR
2.5 } VSWRC\l, f<2GHz, H= 508 mm
, ‘\

/ /

\
,;.8,

“y, M.52
‘---
+65 ,XP (-135 W/H]
\ A’.,
\ \,
\ ?5 101
0,6; ,.75 ‘ (,0s)
\ \
. I
2A1/H j

..~o 05 10 15 20
W/n
2.5 30 3.5 40
;

0
.o.g/f(GHz) 1.0 20 3.0

Fig. 7. Optimum miter percentage as a function of normalized linewidth ●

●
for various 8, values.

However, for narrow lines (e.g., 50 Q on 0.635-mm garnet)

Fig, 8. Frequency dependence of normalized bend electrical length for
the mismatch associated with the 50-percent mitered case is
several optimally mitered configurations.
still substantial (Fig. 9(b)).
For lines of nominally 50-!2 characteristic impedance the
measured optimum miter for a given 8, is alternatively from the general expression (5), or for the specific case of
available from the following very simple empirical 50-!2 lines from (6). These expressions permit the design of
expression bends having maximum VSWRS of 1.1 up to 16 GHz on
0.635-mm substrates.
M = 46 + 2.56., g, s 16, 20 N 50 Q. (6)
An estimate of the low frequency value of the electrical
For design purposes, (5) and (6) may be considered to be length 2AE for the optimally mitered case maybe read from
accurate to t 4, resulting in a maximum VSWR of 1.1 up to 2 the dotted curve shown in Fig. 6. The frequency dependence
GHz (H = 5 mm, W/H > 0.25). These tolerance limits of this component may be estimated by reference to Fig. 8.
account for curve-fitting and measurement errors as well as
C. 45 and 6P Bends
the apparent slight sensitivity of the optimum miter to
frequency. For narrow lines some care is necessary in Often bends of other than a 9W angle are desirable. Fig.
fabrication to avoid accidental loss of continuity or over 9(a) presents measured 12-GHz data for B’ as a function of
compensation due to poor control of etching. percentage miter for the case of a 45° bend in a nominally
The frequency variation of the measured normalized bend 50-Q line on alumina, for which the optimum miter is seen to
electrical length, 2A#/H, is shown in Fig. 8 for several be approximately 30 percent. It maybe useful to note that
optimally mitered configurations of practical significance. two unmitered 45° bends used in place of a single unmitered
90° bend always provide a lower mismatch, similar to the
VI. DESIGN INFORMATION
experience with stripline [1]. For the corresponding case
This section presents a summary of information useful to involving unmagnetized garnet substrates, 60 and 45° bends
the circuit designer who is concerned with accounting for or may be designed by use of the results plotted in Fig. 9(b); the
minimizing the effects of symmetric bends in microstrip. 60° case is particularly useful for circulator designs.

A. Unmiteved Bends D. Comparison with Published Data

First, an estimate of the low frequency value of the Very few results are available for comparison. The opti-
equivalent electrical length 2At’ is obtained from Thomson mum miter percentage given by Kelley et al. [12] for a 90°
and Gopinath’s curve of Fig. 6 [7] for any given linewidth. bend in a 50-Q line on alumina is 57 percent and does not
With the Cd value obtained from the uniform plane wave agree well with our observation. However, the measured
model (Cd = Cm W) and by use of(3), this then yields a low value of 72 percent reported by Easter et al. [10] is in
frequency estimate for B’. An indication of the frequency excellent agreement. For 8, = 2.6 none of the results and
dependence of the circuit elements may be obtained by trends reported by Hammerstad [14] agree well with our
reference to Table I, Figs. 4 and 5, and [1 1]. data.

B. Mitered Bends VI. CONCLUSIONS

Over a wide range of 6,’s and normalized linewidths, the We have described microstrip measurements undertaken
optimum miter percentage, M, can be readily calculated using both a resonant-loop technique and a simple return
DOUVILLE AND JAMES: STUDY OF SYMMETRIC MICROSTRIP BENDS 181

047
d:
The optimum miter was also determined for 45° bends in

~=$ $*’S SO-Q line on alumina,

on garnet.
There is generally
and for 45 and 6~ bends in 50-Q line

significant frequency dependence of the

03
elements of the discontinuity equivalent circuit. However,
for the unmitered case the experimental data, when extrap-
olated to zero frequency, agree reasonably well with theor-
etical results available from quasistatic analyses. There is no
02
theoretical data presently available for mitered
configurations.
ACKNOWLEDGMENT
E’
\ The authors are grateful to R. Marchand, E. Minkus, and
01
45” N. Morin for performing most of the measurements, and to
G, R. Painchaud, F. Bouchard. and M. Stubbs for some of
h \ the computed results,

I
REFERENCES
o I 1 ! t 1 I 1
I
100 [1] H. M. Altschrrler and A A. Oliner, “Discontinuitles in the center
“A MITRE _ conductor of svmmetric strip transmission frne.” IRE Trans. A4tcro-
wave Theory Tech., vol. MT’T-8, pp. 328-339, May 1960.
[2] A, A. Oliner, “Equivalent circuits for discontinuities in balanced strip
transmission hne,” IEEE Trans. kficrowaue Theory Tech.i vol.
-ol — MTT-3, pp. 134-143. Mar. 1955,
[3] J J. Campbell, “Application of the solutions of certain boundary
value problems to the symmetrical four-port junction and specially
truncated bends in parallel-plate waveguides and balanced strlp-
transmlssion hnes,” IEEE Trans. Microw,aue Theory Tech., vol.
MTT-16, pp. 165-176. Mar. 1968.
[4] P, Benedek and P Silvester, “Mlcrostrlp discontinuity uapacltances
(a) for right-angle bends, T junctions, and crossings,” IEEE Truns
03
1 Mtcrowau. Theory Tech., vol. MTT-21, pp. 341-346, May 1973.
[5] A. Gopinath and B. Easter, “Moment method of calcedatmg discon-
tinuity inductance of microstnp right-angled bends,” IEEE Trans.
Microwave Theory Tech,, vol. MTT-22, pp. 880-883, Oct. 1974.
[6] R. Horton, “Electrical characterization of a right-angled bend m
mlcrostrip hne,” IEEE Trans. Microwauc Theory Tech., vol. MTT-21,
o 2–
pp. 427-429, June 1973.
[7] A. Thomson and A Gopinath, “Calculation of microstrip discontin-
uity inductances,” IEEE Trans. Mlcrowaue Theory Tech., vol.
MTT-23, pp. 648-655, Aug 1975.
B’ [8] L Wolff and W. Menzel, “A universal method to calculate the dyna-
mical properties of microstrip dlscontinruhes,” presented at Proc. 5th
ol– European Microwave Corf, Hamburg, Germany, Sept. 1975.
/ [9] B. Easter and 1, M, Stephenson, “Resonance techmques for estabhsh-

R\ !
mg the equivalent circmts of small discontinuities in microstrlp,”
Electron. I.ett., vol. 7, p. 582, Sept. 1971.
[10] B. Easter, J. G. Richings, and I. M. Stephenson, “Resonant
techniques for the accurate measurement of microstrip properhes
o \ \ and equivalent circtuts,” in Proc. 1973 Earop.an Microwaue Con$
I I I I I I
10 20 30 40 5 100 (Brussels, Belgium), Paper B.7.5,
% MITRE _ [11] B. Easter, “The equivalent clrceut of some mlcrostrlp dlscontmu~tles,”
IEEE Trans. Microwave Theory Tech., vol. MTT-23, pp. 655-660,
Aug. 1975
[12] D. Kelley, A G. Kramer, and F, C. Wdlwerth, “Mlcrostrlp filters and
couplers,” IEEE Trans. Microwau. Theory Tech., vol. MTT-16, pp.
%i-562, Aug. 1968.
[13] L. Lewin, “Radlatlon from discontmultles in striplines.” Inst. Elect.
Fig, 9, Variation of normalized susceptance with miter for nominally Erg. Mormgr., no 358E, pp. 163-170, Feb. 1960.
[14] E. O. Hammerstad, “Microstrlp handbook,” ELAB Rept., no.
50-Q lines at 12 GHz on 0.635-mm substrates. (a) 90 and 45” bends on
STF44-A74169, Unw. of Trondhelm. Norway. Feb. 1975.
alumina. (b) 90, 60, and 45° bends on unmagnetized garnet.
[15] R J. P. Douwlle and D. S. James, “Experimental characterization of
mlcrostnp bends and their trequency dependent behawor,” IEEC’E
1973 Conf Dtgest, Toronto, Canada, pp. 24-25, Oct. 1973.
loss method. The 90° bend electrical length and shunt [16] W. J. R. Hoefer and A. Chattopadhyay, “Evaluation of the equivalent
susceptance values have been obtained for a useful range of circuit parameters of mlcrostrtp dlscontmulhes through perturbation
of a resonant ring,” IEEE Tram. Mtcrowao. Theory Tech.. vol.
line widths, dielectric constants and frequencies. MTT-23, pp. 1067-1071
We have investigated the reduction in discontinuity capa- [17] E. O. Hammerstad, “Equations for mlcrostrip circuit design,” m
Proc. 5th Europeun Microwau. Conj. (Hamburg, Germany), pp.
citance obtainable by rounding or 45° mitering of the bend
268-272, Sept. 1975.
and useful empirical expressions have been presented for the [18] W, J. Getsinger, “Microstrlp dispersion model,” IEEE Trans. M’lcro-
optimum miter percentage as functions of linewidth and E,. waue Theory Tech, vol. MTT-21. pp. 34–39, Jan. 1973.
182 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. MTT-26. NO. 3, MARCH 1978

[19] J L Smith, “The even- and odd-mode capacitance parameters for Trans. kf~crowaue Theory Tech., vol. MTT-19, pp. 818-824, Oct.
coupled lines in suspended substrates,” IEEE Trans. Mlcrowaw 1971.
Theory Tech., vol. MTT-19, pp. 424-431, May 1971. [21] 1, Wolff, Sept 1975, private communication.

[20] W. H, Leighton and & G. Milnes, “Junctlorr reactance and dimen- [22] H. J. Schm,tt and K. H. Sarges, “Wave propagation m microstrip,”

sional tolerance effects on X-band 3-dB directional couplers,” IEEE Nachrtchtentech Z., vol. 24, no. 5, pp. 26(-264, 1971.

Some Considerations About the Frequency

Dependence of’the Ch aracteristic I mpedance .

of Uniform Microstrips
BRUNO BIANCO, LUIGI PANINI, MAURO PARODI,
AND SANDRO RIDELLA, MEMBER, IEEE

Abstract—Various possible definitions of characteristic im- of different lengths [11]. This may explain the availability y of
pedance are derived from two different microstrip models obtaining computations of ;} which (with some adjustment of par-
similar results. It is shown that slightly different definitions yield ameters) compare well with measurements [1], [2].
strongly different behavior versus frequency.
Insofar as the characteristic impedance is concerned, the
case is far more involved. Strictly speaking, we could only
I. INTRODUCTION
i define the characteristic impedance pertinent to each mode;
HE FREQUENCY DEPENDENCE of the character- then when one speaks of the characteristic impedance of the
T istic impedance ZO(0) of uniform microstrips has been microstrip, it is intended, either explicitly or not, that some
discussed in various papers [4]-[10] with completely differ- hypothesis is made about the mode coupling imposed by
ent results. In Denlinger’s approach [4] Zo(m) is a decreasing the terminations.
function, while the other authors have obtained an increas- The various 20(u) definable are different. In the present
ing behavior. work we shall consider a number of different possible
In this paper it is shown how it is possible to obtain such definitions of Z., on the basis of two microstrip models, due
different results, starting from the definitions of voltage, to W. J. Getsinger and H. J. Carlin, respectively.
current, and power in mi,crostrip.
II. GETSINGER’S MODEL
Microstrips are transmission lines difficult to analyze. A
microstrip, due to the presence of two distinct dielectric In this model [1] the microstrip fundamental mode is
media, cannot carry a pure TEM mode. In most practical approxirnated by a longitudinal-section-electric mode.
cases it can be assumed that only one mode (referred to as However, as the actual structure precludes a direct analysis,
the fundamental mode) does propagate; however, the per- the microstrip (Fig. 1(a)) is substituted by the structure of
tinent propagation constant y does not depend linearly on Fig. l(b); here the three unknowns (a’, b’, and u) are found by
the frequency; this is the dispersion phenomenon. This supposing that the new structure exhibits the same per-unit-
situation ofsingle-mode propagation is assumed, in general, length capacitance and inductance as the original one. A
in the available theoretical calculations. Regarding measur- third equation arises when we suppose that the calculated y
ements, y or c, = – y2/co2e0 PO can be determined by means fits exactly with a measured value at a given frequency, e.g.,
of techniques which eliminate the effects of the strip termina- at 10 GHz. With this adjustment Getsinger was able to give a
tions, such as the use of sliding probes or by comparing lines good analytical expression for y. The analysis of the struc-
ture of Fig. l(b) yields [1]:
Manuscnpt received July 26, 1976: revised July 28, 1977. H,,, = – A,~~,Sh(~,*,) (1)
B. Blanco and S. Ridella are with the CNR Laboratory for Electronic
Clrcults, Vla alt’Opera Pla, 9/b 16145 Genoa, Italy. H*,, = ‘Ai~2ch(~,#~) (2)
L. Panini is with the Marconi Itahana S.p.A., Geneva-Cornlgliano,
Italy En,i = S/io~/4iCh(yi+i) (3)
M. Parodi is with the Department of Electrical Engineering, University
of Genoa, Viale F. Causa. 13 16145 Geneva, Italy. yz + y: = s2&o&*#o (4)

0018 -9480/78 /0300-0182 $00.75 @ 1978 IEEE