Professional Documents
Culture Documents
MARCH 1966
T. CHAN,
MEMBER,
IEEE,
AND
Abstract-In
this paper we generalize Youlas theory of broadband matching to include both passive and active l-port load impedances. The design philosophy of lossless coupling networks for an
active load is different from that of a passive load; however, the
design theory is quite similar. The theory is illustrated in a detailed
example with a tunnel diode as the active load. Both series resistance
and inductance are used in the equivalent circuit representation.
New results in terms of theoretical limitation on gain-bandwidth
and a sufhcient condition for potential stability are obtained.
E. s. KUH, FELLOW,
IEEE
I. INTR~D~TIoN
HE CLASSICAL problem of broadband matching
was first introduced by Bode in the study of
coupling network design for vacuum tubes [l],
Fano later extended Bodes result to arbitrary passive
load impedances and derived a set of integral constraints
on the reflection coelhcient [a]. Fanos method depends
on, as a crucial step, the Darlington equivalent representation of a passive load. The design procedures for
the optimum matching network was worked out, but
only for a few classes of simple impedances. Recently,
Youla developed an alternative theory of broadband
matching which bypassed the step of finding the Darlington equivalent, and consequently, was able to handle,
more complex impedances with relative ease [3]. The
present paper is a generalization of Youlas work to
include both passive and active load impedances.
The design philosophy for obtaining a coupling network for an active load is different from that for a passive
load; however, the theory involved and the design procedure are quite similar. These concepts are brought out
in Section II along with some preliminary notions of
importance. The necessary constraints on the complex
reflection coefficient and the sufficient conditions for
the realization of the coupling network are introduced
in Section III. The presentation is maintained in a simplified form deliberately so that readers can follow the
physical significance of each step without being bogged
down by the details which are necessary for the general
theory. As a matter of fact, the succeeding four sections
are devoted to a special, but important, example using
the tunnel diode to illustrate the main ideas, while the
general matching theory is postponed until Section VIII.
For those who are specially interested in the tunnel
diode amplifier problem, this paper gives a fairly com-
[71-POI.
CONCEPTS
Z(s) - z(7s)
p(s) = Z(s) + z(s) *
0)
on p(s) imposed
by the
1966
MATCHING
SIMPLIFIED
CONSTRAINTS
ON THE
REFLECTION
COEFFICIENT
The crux of the whole matching problem is to determine various necessary constraints on p(s), which
are imposed by z. From these necessary constraints a
set of sufficient conditions is obtained which guarantees
that Z(s) of (l), expressed in the following form, is a
positive real function
Z(s) = 44 + d-4
1 -
1) In
DIODE AMPLIFIERS
_ z(s).
PC4
R + jX -RL + 3,
R + jX + RL -I- jXL
(44
(R - RJ2 + (X + X,)
= (R + R,) + (X + X,)
(4b)
and
(jw)
if
z(juO)
Ip(jb.hJI I
if
z(jmO) is passive.
1,
is active,
(5)
IL .GJI 5 11 sol.
(f-3
IL 4---so)1
5 IL sol.
Re [Z(ju)] = Ev Z(S)~.,~,
Using the type 2 constraints, i.e., (5), we obtain immediately Re [Z(jw)] 2 0, for all w.
Next, let us investigate the open RHP behavior of
Z(s). From (2) we see that poles of Z(s) are contributed
by zeros of 1 - p(s), poles of Z(S) and poles of 2(--s).
In the type 1 constraints, we rule out zeros of 1 - p(s)
in the open RHP except at a transmission zero, where the
zero is cancelled out by the transmission zero. In the
type 3 constraints we have restricted p(s) to be a zero
at a pole of z(s), where x(-s) is regular. Thus from (2),
Z(s) = x(-s) is regular at that frequency. Similarly,
we have restricted p(s) to have a pole at a pole of x(-s),
where z(s) is regular. Thus the poles of p(s) and 2(--s)
cancel, hence, Z(s) is also regular. Therefore, Z(s) is
analytic in the open RHP.
Finally, we must check the jw-axis poles of Z(s) and
the residues. Since we have not expressed explicit conditions on the constraints ,in terms of Km,, K, etc., we
cannot give a proof at this moment. However, it is
reasonable to believe that since constraints on p(ju) are
determined from the conditions that the jw-axis poles
and zeros of Z must be simple and K-, and K, must be
real and positive, we only need to trace it backward to
complete the proof of positive reality.
We will assume, therefore, that the constraints of
types 1, 2, and 3 are necessary and sufficient for a positive
real Z(s). The complete statement is given in Section
VIII as a theorem with its proof. The purpose of this
rather crude presentation in this section is to give the
reader various concepts of the general matching theory,
which are needed for the tunnel diode problem to be
treated in the next four sections.
IV.
THE TUNNEL
(8)
DIODE
AMPLIFIER
PROBLEM
*
x(s) = r + Is + s+
(7)
On the jw-axis the angle condition for passivity is equivalent to the non-negativeness of the real part.
The above four types of constraints in p are used in
the next four sections with the tunnel diode as the active
load.It turns out that the first three types of constraints
on p(s) represent sufficient conditions for realizability
of Z(s). We will present a reasonable justification of our
claim as follows: let us consider first the real part of Z(jm) :
MARCH
2(-s)
= r - Ls - -sfl
r<l
(94
(9b)
= r + &
= rs2s~~ 1 r -
(10)
Let us go through the four types of constraints imposed on p(s) by z. For the type 1 constraints, we first
calculate the transmission zero. From (lo), the transmission zeros are found to be at
S,
= hjw,
= *j
1
;-
1 = *j
l-r
-.
01)
1966
At =tj~,,
obtain
MATCHING
= 1.
Substituting
(18) and (20) in (16), we obtain the behavior of p(s) at ~0:
(12)
PM
From
!++1+(2!2&$+...
!++1+(g-;);+
2r
(10)
IPGJ.4l >
(14)
0 5 a < 0,.
1,
(15)
w 2 0,.
1,
hence, x
(16)
(17)
1,-r+--&
=
Is++&
2r
=1-z+***
0%
+ K, + - - * ,
(19)
a4
-=
4s)
Taylors
... ,
p(s) = 1 ++
(21)
... .
(22)
(23)
and
UlJ <r-
+
With the constraints of
sider the design. The first
magnitude square-function
obvious constraints of type
IPb.4l
>
PW
IPWI
...
-- 4-4
4s)
1+
=I-K-,+ls
DIODE AMPLIFIERS
for
wo=O,
for
w,#O.
1,
0 5 0 < WV,
w = w,,
= 1,
5
1,
w, <w
<
= 1,
w=
00.
expansion of
PW
(24)
f=Q)
A typical plot is shown in Fig. 4, where K2 > 1 represents the gain in the pass band and w, is the bandwidth.
A typical function is given below:
Km,s + K, + . . .
zs+r+*
K2 - 1
A++
(-p)f+
.. . .
IP(%412 =
(20)
1 +
EWW /WC,
(25)
IP(
lll?L
10
IEEE TRANSACTIONS
ON CIRCUIT THEORY
MARCH
V. INTEGRAL
CONSTRAINTS
GAIN-BANDWIDTH
K2
AND
LIMITATIONS
-----
OJ-
(32)
WC
Fig. 4.
wr
IPw12.
(26)
The factorization
is not unique, we, therefore, express
p(s) in the following form:
P(S) = PO(S)PI(S)
P2W,
In
[~~(s)~2(s)l
9 ReX;>O,
i=l,2,..+,
= 1
In
p2(s>
+fs,mhIPO(&)
s2
+&4
w2[cEw
(33)
Erj
(27)
Pzb)
(30)
and
Jn
~~(4
In
Pi
+~~$J$b.
(34)
I*+0= F;
O(ls - jw,I).
+ o(+)
(36)
To employ (34), we must extend its validity to the jwaxis. Youla has established the following: at s = 03,
since In p(s) has a zero of order one, the first-order coefficients of Taylors expansions at infinity for all terms
in (34) must be equal. Similarly at s = jw,, since ln p(s)
has a zero of order one, the zeroth-order coefficients of
the Taylors expansions at jw, for all terms in (34) must
be equal. The expressions for In pi(s), ln p*(s), and
[2s/r(s + w)] are given below [the forms of pi(s) and
p2(s) are given in (28) and (29)]. At s = a
ln p,(s) = -;
In p2(s) = -c
2s
&2
2
+
21
w)
+ *** .
--;
(374
+ * *. .
+
**-
(37c)
1966
CHAN
AND
KUH:
MATCHING
THEORY
At s = jw,,
FOR TUNNEL
DIODE
11
AMPLIFIERS
3.000
1+
In pi(s) = j2 tan- wI
2.667
... .
(384
2.333
2.000
(38b)
.2 -. w,
2s
7r(s2+ w) = 3 ; w2 - w;
Substituting
-2r
-.=
1
(38~)
y 1.667
4 1.333
1.000
0.667
0.333
0.000
-2
- 2
Ai
0.
+ ! S,- h Ip(jo>l~WF
or
In \p(jw)l dw = 1 + C Xi - i*
i
Substituting
(39)
0 = j2 tan- w,
I+
j Ctanm1wz~~{-12
or
% m1,b(jw)[
dw
n-s0 w,- w2
= tan- 1 + 4 C tan- w; ~Y;,,z*
*
I
W,
(40)
on gain
h-~Ip(j4l = H,
= 0,
o<w<w,
w > w.
(41)
1
tanh- (wJw,) = wr tan-1 W,
1 -r/l
%hJ,
W,H =1-i+
7r
CXi,
i
r < 1,
I<
and
(43)
Eliminating
H from the two equations, we obtain a
formula which gives a relation for w, in terms of the
tunnel diode parameters and the Xis to be adjusted.
L + C * tan- w4Try{. 2
tanh- (0,/w?) = or tan- w,
i
1
* (44)
WC/W,
1 -;+
pi
The left-hand side of the equation is monotonically increasing in w,/w, and is larger than unity. It gives the
1 These equations were used by Smilen
. and._.Youla
- -_in the derivation of necessary conditions for potential stability 1131.
(45)
1
1
1 - w, tan- (J-Q
(464
(46b)
(innepers)
= x
(47)
It should be emphasized that by increasing 2 (using degenerate matching for example) the optimum bandwidth is reduced (see Fig. 5). However, from (47) the
gain-bandwidth measure can be increased.
12
EZ;$
; y$twqn.
is monotonic
= [K(l
- $)
6 - 49 = -1 tan- 1 )
WV
2/
= --tan-
(50)
2Kw,w, - K 2w,2
- Kw,
= -cos -1 w,
, (59)
WV- Kw,
W,
I-
& = -tan-l
The frequency w, is a normalized cutoff frequency, however, the behavior of IpI2 at jw, deperids on the gain and
thgratio w,/w,. In the stop band
Idj412
< 1,
IPW12 = 1,
0, <w
< 00)
(51)
w = w,, 03.
(52)
P(S)
I- =
= k.
(53)
The factorization
of (48) is straightforward.
Let us
denote the LHP poles and zeros by s, and so and let us
choose p2(s) = 1 in (29). We thus have, from (27),
(8 - som - soy s - 1
P(S) = (s - sJ(s - 3,) .-.s + 1
- 0: f
j$d2wrw,
+ WC,
>
or
2Kw,w, - K2w: W,
2w,w, - wf
W,
or
sin 6 -sine,=-&
1-i.
(554
$2 -1
-
WI
+ Kw,~.
(61)
-** *
so = u. f jw,
= -$
p(s)
s, = a, =t jw,
= -3d2w7wc
of (31). Expanding
(60)
Thus
(54)
-. - WC
= -co,cj -1 WI
W,
2r 1
= l -z+
Y~w,wc - 4
WV - WC
(5%
where
(49)
+ @I.
of a pair of
8,
b(j412
(56)
(4%
and is given by
= K2,
MARCH
(55b)
- cash,
and
2
= 1 - cos 0,.
(64)
1966
- 1 tan- b 5 0 (65a)
n
< 0. (65b)
As n approaches infinity,
(66)
WC
-=IWV
1
w, tan- WV
(67)
Thus, if the given tunnel diode has its parameters satisfying the necessary condition for potential stability, i.e.,
(46b) ; a finite w, can be obtained. In the limit,
1
?I
w,
tan-
VII.
EXAMPLES
1
WV
ILLUSTRATIVE
r = 0.6
and
1 = 1.8.
1
- - 1 = 0.815,
d- r
tan-
-!- = 0.888,
W,
and
1 - ; = q.667.
First, we will use the approximating
function of the
previous section as given by (48). We need to solve for
K
and w, from the two constraint equations of (58) and
(62). They are
i (4, -
e,)
= tan- I = 0.888,
W,
and
= 92.7,
and
w, = 0.00184.
MARCH
II = 1.8
I249
I = 0.6
.I
(a)
dB
60
I
0
I
0.2
I1
0.4
0.6
II
0.8
W--c
1.0
limits on gain-bandwidth.
With a brick-wall type of
response the maximum bandwidth is given by (45),
We = 0.46 w, = 0.375. The maximum gain for the optimum bandwidth is given by (47), gain = 48.6 dB.
Obviously a more efficient approximation can be used.
Let us refer to (25) and choose f to be a Chebyshev
polynomial of fifth order. Thus
K2 - 1
~2C%JJ/Wc)
I&412 = 1 +
l +
1 -
(b)
Fig. 8.
(6%
Next, (70) is substituted in (68) to obtain a set of simultaneous equations. These equations together with (71),
(73), and (74) are solved with a computer. We obtain
WC =
PM
(74)
0.238,
(75)
1309s5+1922s4+1504s3+742.8s2+231.4s+37.45
1309s5+176.7s4+104.6s3+9.592s2+1.709s+0.06264
s-l
'&g-P
(76)
and
K = 599
,,+~,(~)+~2(~)2+ba(~)3+b4(;)4+b5(~)s
or
55dB.
(77)
First
0.815,
(70)
from
(30)
(71)
Next
Z(s) =
2r 1
PCS> Is-
and
ao+a,s+a2s2+a,s3+a,s4+a,sb
p(jw,) = 1,
03 0.4
The constraints
0.2
W-
(w/wJ2
0.1
033)
where
()
I I I,!,
0.05
0.01
=l-Ts+
(72)
(73)
(78)
A GENERAL
MATCHING
THEORY
1966
MATCHING
OW
15
DIODE AMPLIFIERS
(87)
. .
di = 0,
i = 1,2, **. ,r - 1.
4s) + d-s)
-
PC4
=
as>
1+- d--s)
Ci?i=Ci,
4s) .
1+z(s)
44
Z(S) = z(s)
-- + x(-s)
1
4s)
@W
4s)
(90)
(814
P(S>
i=r+WZ,~**,r+2?YZ-l.
P(S)
&(s
jd
(91)
* . a ,
RelN&b)l , o
(81b)
-d,
and
+ a-(,-,)(S
+
2(-S)
= b-*(S -
a0
- Si)-(-
+ Ul(s -
SJ- + b-(a-l)(S
si>
+ **+ *
Si)-(n-l)
+ bo + f&b -
SJ +
CL = 0,
i = 1,2,
.-a
,r
(94
19
(83)
- Si)-(k-l) + * *
+ co+ Cl(S
- Si>
+ ***
p(s) = d-,js - si): + d-,,-,,(s
K-,(s
- jwi)-'
i=l,2,a**,r+m.
1,
w-9
(85)
(83)
r+m+l,**.,r+2m-1,
d r+2m-1 = C F+P?lS--l
- K-,C,+,/a-,
where K-, is real and positive.
Condition 5: On the jw-axis, Ip(
ing :
IP(G>l
Theorem
Let z(s) be a given rational impedance which may be
either active or passive but non-Foster. Then
z(S) =
4s) + d-4
1 - p(s)
i = r + m,
(96)
and
+ K, + - . . ,
di = Ci,
(84)
- s,)-(-~) + . . .
+ do + d,(s - si> + - - .
=
(93)
and
-z( -s)
~
44
Z(s)
do = co = 1,
(82)
* a
--*
(92)
1,
if
1, if
(97)
x(jw) is active,
(984
z(j,) is passive,
i.e., Re [x(jw)] 2 0.
_ z(s)
(if degenerate),
(98b)
i<m-n-l,
(99)
16
IEEE TRANSACTIONS
ON CIRCUIT THEORY
function
and
di = Cc, i = m - n,
= 1 and
-1 2,-,-1
n)
(if m < n)
and
= Cm-l + K-,(1
- C,)/u-,
(if degenerate),
(102a)
P(s)]~+~.
and if
Co = 1 and
Ci = 0,
di = Ci,
[-C,+Js
* 1 [
i = 1, 2, . . . , q.
i = 0, 1, 0.. , m + q,
(102b)
= landCo
= l,d,
d, =
(s -
jw,>+
(103a)
(degenerate),
= Co = land
a-ICI
a-, + K-1
(103b)
jm,)m-l
d r+2m--1= C 7+2n--l -
and if m = 1,
andifm
...
1
,
(107)
C,+mK-1
fLf8
(138)
-1
..
which gives
(106)
WQ
d,-I
zero, and
= 2 Re$m+O)l (s - jwo) + . . . ,
I cmmn+i = 0, i = 1, 2 * * * ) qI
m - n (if m 2.
4jwo) + 4--ho)
1 - P(S)Is=i,,= K-,(s - jwo)- + . . .
m - n + 1, ***
if m 2 n, C,-,
an-n+,
1 -
MARCH
+ *- * .
(109)
A. Necessity
PC3
PCS> I-
[-Cr+JS
- spy-
+-**]+[e (s- s,)
+*--1
=[c&s
+...j
1+[F (s_s,)m
(110)
sJr+m
--*I
* 1 - E@J (s - s,) + * * *
a-,
[
1
,
(105)
If m # 0, we
- spy- + **-1
I**lip= [Cm-,(s
+ [(s
I *-a.
- spy+ - *-]}{l - 2
1966
[Co
CI(S
jw,)
(112)
*/loo
= 4jwo) + 4--jwo)
- 4jw0),
-4s
- jwo)
= 2
(&+2m--l
-
Cr+2m-1)(~ -
Re W41 cs _ jwo)-' + . . . .
-4
jwX1 + * *- .
(116)
z(s)
=[a-,(~
- jw,Jmm
+..-1(dm-1
,?GftT.Fm-l
= fi
(117)
Co
= a-, l--d,
J sme*e'
(s - jw,>-' + - - - .
(1 W
- C,,-J(s - sJ2- + . - . .
- [CL,(s - SJ+ - * *]
- jw,)-m +
.w Is+.= [a-,(~
=*
Substituting
(93) and (95)-(97)
for the degenerate case
-**I
&n--n
17
CONCLUSION
p1
Sl2l
Ls21 s221
by the scattering matrix of a lossless 2-port with respect
to passive reference impedances 1 and -~(jw,). Thus the
incident and reflected waves are related by
IEEE
TRANS.4CTIONS
ON
CIRCUIT
VOL.
THEORY
CT-13,
NO.
hfARCH
1966
REFERENCES
Design.
But by definition,
Therefore,
G. YOULA,
Abstract-It
is iirst shown that the power gain of a filter which
has been partitioned into two component parts may be expressed in
terms of a formula involving only the two impedances seen looking
to the left and the right of the common junction. By imposing the
constraints of symmetry and antimetry this formula leads quite
naturally to two global equations for positive-real (pr) functions.
Theorems 1 and 2 present necessary and suf6cient conditions for
the existence of solutions. Moreover, the construction of these pr
functions is made to depend on two algorithms of an extremely
simple character. The theory is fully illustrated by means of four
worked, nontrivial examples. Finally, it is pointed out that synthesis
by bisection is often wasteful of reactsnces (especially in the symmetric case), and a careful count of elements is presented for antimetric filters.
FELLO\v,
IEEE
I. INTRODUCTION
HE PROBLEM
of synthesizing, by physical bisection, the transducer power gains of electrically
symmetric
and antimetric filters has been treated
by several authors [l]-[3]. Recently, Navot and Zeheb
[2] and Navot [4] pointed out that the method of polynomial identification advocated by Guillemin [l] is valid
only for a restricted class of power gains. In [4] Navot
described algorithms (his generalized even and odd theorems) which are applicable in the general case.
In this paper an entirely different approach is taken.
It is first shown that the power gain of a filter which
has been partitioned into two component parts may be
expressed in terms of a formula involving only the two
impedances seen looking to the left and the right of the
common junction. By imposing the constraints of symmetry and antimetry, this formula leads quite naturally