264 IEEE TRANSACTIONS ON CIRCUIT THEORY, VOL. CT-14, NO.

3, SEI’TEMBER 1967
nomials in ladder networks,” IRE Trans. Circuit Theory (Corre- 191 M. N. S. Swamy, “Continuants and ladder networks,” Proc.
spondence), vol. CT-8, pp. 167-168, June 1961. IEEE (Letters), vol. 54, pp. 1110-1111, August 1966.
L61F. F. Kuo and G. H. Leichner, “An iterative method for [loI V. 0. Mowery, “On hypergeometric functions in iterated
determining ladder network functions,” Proc. IRE (Correspondence), networks,” IEEE Trans. Circuit Theory,-. vol. CT-11, pp. 232-247,
vol. 47, pp. 1783-1784, October 1959. June 1964.
[61T. R. Bashkow, “A note on ladder network analysis,” IRE ~1 A. M. Morgan-Voyce, “Eadder network analysis using
Trans. Circuit Th,eory (Correspondence), vol. CT-8, pp. 168-169, Fibonacci numbers,” IRE Trans. Circuit Theory (Correspondence),
June 1961. vol. CT-6, pp. 321-322, September 1959.
~1 S. C. Dutta Rov. “Formulas for the terminal imnedances and [I21M. N. S. Swamy, “Properties of the polynomials defined by
transfer functions of” general multimesh ladder networks,” Proc. Morgan-Voyce,” Fibonacci Quart., vol. 4, pp. 73-81, February 1966.
IEEE, vol. 111, pp. 1653-1658, October 1964. 1131 M. N. S. Swamy, “On the zeros of a set of polynomials,
f81G. H. Burchill, “A signal flow graph method for determining SIAM Rev., vol. 8, pp. 388-391, July 1966.
ladder network functions,” Proc. IRE (Correspondence), vol. 48, II41M. N. S. Swamy, ‘Some properties of the polynomial C,,(Z)”
p. 1175, June 1960. (to be published).

On the Realization of a Constant-Argument

0
Immittance or Fractional Operator
SUHASH C. DUTTA ROY, SENIOR MEMBER, IEEE

Abslract-Methods for realization of an immittance whose implies that the constant phase angle &, must lie within
argument is nearly constant at X=/2, 1x1 < 1, over an extended fre- the range -7r/2 to +~/2. If & = &r/2, 1x1 5 1, then
quency range, are discussed. In terms of the generalized complex
frequency variable s, these immittances are proportional to si, Bode’s integral relationship[” between the magnitude
and as such they are approximations of Riemann-Louville fractional and phase of a driving-point function implies that the
operators. First, we present a method which is applicable only for magnitude shall vary as wx. In terms of the complex
the special case 111 = $. This is based on the continued fraction frequency variable s, such immittances are of the form s’,
expansion (CFE) of - the irrational driving-point function of a uniform and as such they approximate the Riemann-Louville
distributed RC (URC) network; the results are compared with those
of earlier workers using lattice networks and rational function
fractional operators.“” The limiting cases X = f 1 are
approximations. Next we discuss two methods applicable for any realized by conventional inductors and capacitors while
value of X between -1 and +l. One is based on the CFE of a resistance represents the case X = 0. The problem of
(1 + Sir)*h ; the two signs result in two different circuits which realizing s’, - 1 5 X 5 +li may therefore be looked
approximate s-h at low and high frequencies, respectively. The upon as one of filling out the gaps between X = -1 to
other method uses elliptic functions and results in an equiripple
approximation of the constant-argument characteristic.
X = 0 and X = 0 to X = +l., These immittances are ir-
In each method, the- extent of approximation obtained by using rational and as such may be used to extend the domain of
a certain number of elements is determined by use of a digital conventional network synthesis to include irrational net-
computer. The results are given in the form of curves of ~/or work functions.
versus the number of elements, where o2 and or denote the upper ‘Besides being of academic interest, constant-argument
and lower ends, respectively, of the frequency band over which
the argument is constant to within a certain tolerance.
immittances are also useful in analog simulation of dif-
From the lumped element networks, we derive some i?C networks fusion problems such as heat flow or neutron diffusion
which can approximate s” more effectively than the lumped net- in a nuclear reactor, compensation of ‘servo-systems,‘31
works. The distributed structures can be fabricated in micromin- termination of a high-frequency cables;“” provision of
iature form using thin-lihn techniques, and should be more attractive a proper asymptotic roll-off for wideband high-gain am-
from considerations of cost, size, and reliability.
plifiers, ““and provision of broadband bridge balances
for certain electrolytic impedances (like microelectrodes
I. STATEMENT OFTHEPROBLEMAND REVIEW inserted in a living system) in which the magnitude of
OFPREVIOUSWORK the impedance varies as a-‘, 0 2 X < 1. [15’
HIS PAPER investigates the problem of realizing The problem has been previously studied by Carlson
a passive immittance whose argument is nearly and Halijak,L3’-161 who viewed the problem as one of
T approximation by a regular Newton process; Morrison,[‘6’
constant over a wide frequency band. Passivity
who used a parallel combination of a number of series
RC circuits; Lerner, ‘15’ who suggested the series connec-
Manuscript received December 1, 1967; revised March 31, 1967.
The work reported in this paper was supported in part by the tion of a number of parallel RC networks; and Steiglitz’211
National Science Foundation under Grant GK-1111. who suggested a rational function approximation. Hes-
The author is with the Department of Electrical Engineering,
University of Minnesota, Minneapolis, Minn. 55455 selberth”” has investigated the distributed RC version

.
DUTTAROY: REALIZATIONOFCONSTANT-ARGUMENTIMMITTANCE 265

of Morrison’s circuit. This network is, however, of the- frequency ratio q = u2/u 1 = w2/wl over which 4 is con-
oretical interest only, because it would be very difficult, stant to within a specified tolerance, with the number of
if not impossible, to fabricate the structure with present- elements, N.
day materials technology. Dutta Roy and Shenoi[” show In the Appendix, an alternative derivation of Steiglitz-
that a uniform Rc (m) network with suitable lumped Pullen formulas is presented by using the CFE of sl”
impedance terminations can realize an impedance s-l/’ given by Khovanskii; RC realizability of the rational
over a band of frequency whose upper end extends to functions is also demonstrated.
itinity. They also suggest two lumped impedance ap-
proximations to the distributed structures. III. REALIZATION OF s-r” USING THE DRIVING-POINT
II. SCOPE OFTHE PAPER PROPERTY OF URC NETWORKS

The first part of the paper is concerned with a method A. Driving-Point Property of m? Networks and Known
of realizing s1’2 immittance based on the continued frac- Lumped Network Approximations
tion expansion (CFE) of the irrational driving-point func-
The URC network shown in Fig. 1 has the following
tion of an m network. A variation of this expansion
open-circuit impedance matrix:“21
results in a series-R, shunt-c ladder from which is derived

I? (1)
a nonuniform z network which would be more effective coth 6 csch %“&
in approximating s”’ at low frequencies than the URC [z] = -+
network. The results are compared with those obtained ST Fcsch 6 coth 6
by following the approach of earlier workers, viz., property where
of lattice networks, as suggested by Carlson and Halijak,Ial
and rational function approximations suggested by R = total resistance of the resistive film,
Steiglitz[‘l’ and Pullen.[‘D’ The second part of the paper C = total capacitance between the upper resistive film
gives two methods for realizing sx for any value of X and the lower conductive film, and
between - 1 and +l. One of them utilizes the CFE of T = CR.
(1 + s*‘)“; the two signs result in two different circuits
The driving-point impedance function with the other
which approximate s-’ for high or low values of s. The
port open-circuited is, therefore,
low-frequency circuit has the form of a series-R, - shunt-c
ladder. Starting from this, a nonuniform RC network
is derived for realizing s-‘, and unlike Hesselberth’s ‘(2)
structure, can be easily fabricated.’ The second method
for sx uses elliptic function approximations of the constant which tends to R/V‘- ST as s tends to infinity. In fact,
argument characteristic and is found to be the best of it has been shown[” that with s = jw, u = wCR, Z(jw) %
all the methods. l/ fi to within an error of 1” in phase for 10 < u 5 m
The mathematical basis of the CFE methods is obtained and Z(jw) = l/G for 20 5 u 5 QJ. This is true for
from the classic work of Wall’zz1 and a monograph on the short-circuited driving-point impedance also. A lumped
continued fractions by Khovanskii.‘131 The elliptic func- network whose driving-point impedance approximates (2)
tion method starts from the approximating function used will therefore be an approximate realization of s-l”.
by Orchard”” in connection with his investigations of an Two such networks were suggested by Dutta Roy and
allied problem, namely, network functions with a constant Shenoi.[” One network uses the partial fraction expansion
imaginary part. of (2), viz.,
In all but the last method, we have discussed the RC
realization of impedances proportional to s’, - 1 5 X 5 0; qs> = AL& + -g 2Rh2n2)
(3)
the circuits for positive values of X can be easily derived n=’ 1 + s $$ 2
from these and will require RL elements. In the elliptic ( )
function method, we have obtained an RL impedance and results in the network shown in Fig. 2. The other
proportional to s’, 0 5 X < 1; an RC impedance propor- is obtained by using a finite n-section lumped RC ladder
tional to s’, - 1 < X 5 0, can be easily derived. of the form shown in Fig. 3, with R, = R/2n, R, =
In each method discussed, the extent of approximation R, = . . . = R, = R/n, C, = Cz = . . . = C,, = C/n,
obtained with a certain number of elements is determined R V&+1= RF% Cn+l = 0. The input impedance of this
by use of a digital computer. The results are displayed network as a function of the radian frequency is given by
in the form of variation of 4 versus some normalized
frequency u, and also in the form of variation of the
z,,,(+) = R(i + $)1’2 coth [n cash-1 (I+ s)] (4)
1It may not be out of place here to mention that the RC ladders
suggested in Morrison’s paperfi61 approximate ~-1’2 only and not
s-x for any other value of A, whatever the value of K and m may be and this approaches Z(jw) given by (2) as n tends to
(K and m are Morrison’s notation). The network in the lumped form, infinity. The extent of approximation obtained in the
as well as its distributed version, was exhaustively studied by this
author and the results are available from him. two methods is defined by the parameter q = u2/u1 =
266 IEEE TRANSACTIONS ON CIRCUIT THEORY, SEPTEMBER 1967
wZ/wI, where w2 and w1 are the upper and lower ends of
the frequency band over which the phase of the impedance
is -7r/4 f lo. The variation of 9 with the number of
elements N is shown in Fig. 4 (curves 1 and 2, respectively).
B. Two More Lumped Network Approximations
The hyperbolic tangent of an angle can be expanded
(a) (b)
into the following CFE’ (Wa11,*22’p. 349):
Fig. 1. (a) Thin-film realization. (b)lSymbol for an Ui% network.
tanh ~=~/[1+82/[3+82/[5+~~~+82/[(2r-l)+~~~. (5)
Combining this with (2), we get

z(s) =$+ I/[;+ I/[%+ l/[;+ -a-. (6)

The network representation of (6) is of the form shown

h---
ZR I (rw)
in Fig. 5 with C, = C/(4r - 3) and R, = R/(4r - l),
r = 1, 2, 3, . . . . If this ladder is terminated in the nth Fig. 2. Lumped network representation of partial
fraction expansion (3).
section, then the resulting network with N = 2n num-
ber of elements represents an approximation to the URC
driving-point function (2) and hence to S-I/~. The varia- R2 R3 R4 Rr
---- - -mm
tion of the phase of Z,,(jw)[= CFE (6) with s = jw
Z(S) *
terminated at the 5th constant term] with u = wCR xc, rzc2 ::c3 xc4 +-T
Cr
is shown in Fig. 6 (curve 1). The plot of 7 versus N is 0 , m-e
shown in Fig. 4 (curve 3). Fig. 3. A series--R, shunt-C ladder.
The CFE (5) can be utilized in another manner to
derive a network for realizing s-l”. Consider the impedance

Z(s) = R (7)
drta%k
As s tends to zero, Z(s) approaches d(R/Cs) so that
unlike (2), Z(s) given by (7) will be an ideal s-l’2 im-
pedance at very low frequencies. Combining (5) and (7),
we can write

Z(s) = l/[sC + 1/[3R + 1/[5sC + 1/[7R + . . . . (8)

The network representation of (8) is of the form shown

in Fig. 3 with R,=O, R,=(4r-5)R, and C,=(4r-3)C,
r = 1, 2, ..a . Terminating this network at the nth
capacitor will result in an approximation of s-l/’ with
N = 2n - 1 elements. The variation of phase with fre-
quency of Z,(jw) [ = CFE (5) with s = jw terminated at
the 5th capacitor] is shown in Fig. 6 (curve 2) and the
variation of 7 with N is shown in Fig. 4 (curve 4). This
does not differ much from curve 3, but the circuit has the
distinctive feature of realizing s-l” at low frequencies.
* For saving space,the CFE
al
bo +
a2
b, +
b, + -. N = NUMBER OF ELEMENTS

Fig. 4. Variation of 7 ( - 45” f 1”) with N = number of elements:

curve 1: Network of Fig. 2; curve 2: RC ladder approximation of
the Unc driving-point function; curve 3: CFE of URC driving-
point function; curve 4: CFE of (R/&T) tanh l/G; curve 5:
same as 4 with the first two terms missing; curve 6: CFE of sl/z;
will be written as curve 7: cascade of symmetrical lattices; curve 8: (1 + s)+ approx-
imation to s-A; curve 9: (1 f l/~)~ approximation to s-i; curve 10:
b. + al/PI + a2/[b2 + * -- + a,/[b, + -- - . elliptic function approximation of .+.
DUTTA ROY: REALIZATION OF CONSTANT-ARGUMENT IMMITTANCE 267

z(s)’ y~~~~~;--J$--;

Fig. 5. A series-C, shunt-R ladder.

-------_

I I I1111111 I I I I III1 I I I111111 I I1111111 I II

IO' IO"' lOr+2 lo’+3 IO'*4
K = UCR-

Fig. 6. Typical phase-frequency plots obtained from the various

approximation methods: curve 1: angle of Z(ju) given by CFE (6)
with s = ju terminated at the 5th constant term (N = 10,
T = -1); curve 2: angle of Z(jw) given by CFE (8) with s = jw
terminated at the 5th capacitor (9, -4); curve 3: angle of Z(jw)
given by CFE (14) with s = j, terminated at the 4th capacitor
(7, -4); curve 4: angle of Pro(j~)/Qro(j~) given.by (17) (11, -2);
curve 5: angle of Z.,(jw) given by (16) with n = 2 (8, -2); curves
6, 7, 8: angle of the right hand side of (19) with s = j, terminated
at the 6th resistor (11, -1) for X = 5, 3, and $, respectively;
curves 9, 10, 11: angle of the right-hand side of (22) with s = jw
terminated at the 7th capacitor (14, - 4) for X = 5, 3 and $,
respectively.

I
C. A Distributed RC Network Realizing s-l” at Low
Frequencies c(x) dx = (41 - 3)C, I = 1,2,3, ..a . (10)
s I-1
The network realizing CFE (8) is a series-R, shunt-C Summing the first I integrals, we get
ladder and will now be used to derive a distributed RC 1
structure which would be more attractive than the lumped r(x) dx = RZ(21 + 1) (11)
element network for approximating s-l” at low fre- s cl
2
quencies. This network can be thought to be the lumped
circuit equivalent of an RC network of infinite length, c(x) dx = CZ(2Z - 1). 02)
s0
with each resistance and capacitance in the former
from which it can be easily shown that
representing the integrated effect of a unit length in the
latter. That is, if r(z) and c(z) represent the resistance 44 = rob + l/4),
and capacitance per unit length at the point x of the (13)
c(x) = c,(x - l/4)
RC network, then
1 with r,, = 4R and c,, = 4C. Obviously, the distributed
r(x) ak = (41 - l)R, (9) network cannot start from x = 0, for this would require
s 1-l a negative capacitance per unit length for x < a. Start-
and ing at x > 2 would amount to some loss in the frequency
268 IEEETRANSACTIONSON CIRCUITTHEORY,SEPTEMBER 1967
band over which the phase is constant. Let us choose
our starting point to be x = 1; then for a length (I - l),
we shall miss the first resistance (=3R) and the first
capacitance (= C) in the equivalent lumped circuit, and
the CFE for the input impedance will be -+---t--2+
Fig. 7. Chain of identical, symmetrical RC lattices.
Z,(s) = 1/[5sC + 1/[7R + 1/[9sC + --a . (14)
by modified Halley and Newton processes, respectively.
The extent of effectiveness of the line from x1 = 1 to An alternative derivation, starting from the CFE of sl”,
xz = Z can be judged3 from the performance of its lumped has been given in the Appendix, where it is shown that
equivalent whose input impedance is given by (14). A both the formulas are contained in the following:
typical phase curve for (14) with s = jw terminated at
the 4th capacitor (thus resulting in N = 7) is shown in
Fig. 6 (curve 3), and the variation of q with N is shown
in Fig. 4 (curve 5).

D. Comparison with Other Approaches

It would be interesting to compare the above results where
with the performance obtained by using other approaches p = q = n/2 for n even, and
like lattice networks[31 and rational function approxima- p = (n + 1)/2, q = (n - 1)/2 for n odd,
tions [181.[211
1) Lattice Networks: Consider the tandem connection of and that the rational function (17) is RC ad.mittance
an infinite number of identical, symmetrical networks realizable with N = n + 1 number of elements. The fre-
shown in Fig. 7. The characteristic impedance of each quency variation of phase of P,,/Q,, (hence N = 11)
network is Z,, = dm so that if the number of net- is shown in Fig. 6 (curve 4) and the variation of q with
works cascaded is infinite, the input impedance Z(s) will N is shown in Fig. 4 (curve 6).
be an ideal s-l” impedance. To obtain the extent of A comparative discussion of the various approaches
approximation obtained by connecting a finite number n described here is given in Section VI.
of such lattices, observe that the propagation function
is given by IV. REALIZATION OF s-',X < 1 :METHOD I
BASED ON THE CFE OF (1 + s*~)~’
T = CR The function (1 + x)” may be expanded into the
following CFE (Wa11,‘221p. 343):
so that the transmission parameters of n lattices in cascade
are (1 + 2)” = I/[1 - ,,/[I + +/[I
cash (n-J 2, sinh (nr)
05) +1-X
1 + ~-
2(2 + x) x/[l + ..f . 08)
cash (ny) 2.3x A 3.4
Replacing x by ST, X by --X, and multiplying the whole
Hence, the open- and short-circuited driving-point im- function by R, we get
pedances are
Zoc = A,/C, = 2, coth (n-d, R(l+ ST)-’
= 1/[; + I/[& + l/[cl Fx)R
Z,, = &/A, = 2, tanh (ny).

The plot of the angle of Z,,(jw) versus u = ~7 for n = 2

is shown in Fig. 6 (curve 5) ; the plot of Z,,(jw) will be The network representation of (19) is of the form shown
similar with 4 starting at -a/2 at u = 0 and tending to in Fig. 5 with C, = ~0,
zero as u approaches infinity. The plot of r] as a function
of the number of elements is shown in Fig. 4 (curve 7), R =n(l-X)(2--)...(r-l---)
v 2h (1 + X) (2 + X) . * . (r - 2 + X) ’
where N should be taken as 4 or integral multiples of 4.
The plot of 7 for Z,,(jw) is approximately the same as this. r = 1,2, f.f wo
2) Rational Function Approximations: HalijakIg’ has
shown that Steiglitz and Pullen formulas can be derived and
(l+X)(2+X)...(r-2+f)
3 It should be noted that an m line with T(Z) = TO(Z+ a) and
c(z) = c&z - $) does not appear to belong to any of the categories ‘* = Xc (2r - 3)(1 - x)(2 - X) - * - (r - 2 - X) ’
of lines for which the Riccati equation for line voltage is exactly
solvable. See the references111,161
~I’] for such categories. r = 2,3, --. . (21)
DUTTA ROY: REALIZATION OF CONSTANT-ARGUMENT IMMITTANCE 269

1000
c, = ; , c, = (2r-1)(1-X)(2-A). * .+-l--x) c,
7 X(1+X)(2+X)*. *(r--1+x)
5
r = 2,3, **- . (24)
3
If ST<< 1, then the left-hand side of (22) becomes RT-‘s-’
so that the above network will be an approximation
to sex impedance for low frequencies. The phase-frequency
100 plot for the driving-point impedance of this network
7 terminated at the 7th capacitor (resulting in N = 14)
5 is shown in Fig. 6 (curves 9, 10, and 11 corresponding
t
to x = Q, +, and 3, respectively). The corresponding 7’s
-7 3 are shown in Fig. 4 (curve 9) and Fig. 8 (curves 3 and 4,
respectively).
The network of Fig. 3 with elements given by (23)
and (24) is a series-R, shunt-C ladder and may be used
IO
to derive an %? structure in the same manner as was
7
done in the previous section for s-l/’ impedance. In this
5 case,
I
3 2x(1 + A>
s0 (1 - X)(2 - A) + . **

I
I 5 7 IO 5 7 100
1
+ 2X(1 + X)(2 + X) . . . (I - 2 + X)
(1 - X)(2 - X) * *. (1 - 1 - X) 1 (25)

N, TOTAL NUMBER OF ELEMENTS c(x) dx = i + ;I: ; ;; + ;I; 5 ;;;; ; ;; + - - .

s 0
Fig. 8. Variation of v with N: curve 1: -60” f1.5”, for (1 + s)-A
approximation to .@; curve 2: -30” Z!Zl”, for (1 + s)+ approxi-
mation to smx;curve 3: -60” f 1.5”, for (1 + l/~)~ approximation
to s+; curve 4: -30” & l”, for (1 + l/~)~ approximation to
svA; curve 5: -60’ zk 1.5q for elliptic function approximation;
+ (21 - l)(l - X)(2 - X) . . * (I - 1 - X) .
X(1 + X)(2 + A) * * * (I - 1 + X) 1 (26)

curve 6: -30” f l”, for elhptic function approximation. Let r(x) = Rf(x) and C(X) = Cg(z); then (25) can be
written in the form
When ST >> 1, the left-hand side of (19) becomes RT-‘s-’
so that if X < 1, the above network will represent a frac-
rp-2+x 1
tional capacitor. The phase-frequency plot of the driving-
point impedance of this network terminated in the 6th
resistance (resulting in iV = 11) is shown in Fig. 6 for I P-1 J
X = 5, 3, and $ (curves 6, 7, and 8, respectively). The
variations of q (-60” f 1.5’ and -30” f 1”) with N Using the relation’23’
are shown in Fig. 8, (curves 1 and 2, respectively), while 2 r(z + 1)
the variation of 71(-45” f 1”) with N is shown in Fig. 4 0W = Iyw + 1)Iyz - w + 1) ’
(curve 8).
A second network can be derived from (18) by replacing where I? stands for the gamma function, the right-hand
x by l/(s~) X by --X, multiplying both sides by l/R and side of (27) becomes
taking the inverse of it. The result is
ryl - X) 1 r(p - 1 + xl
c (2%
m) D=1 uk - A) ’
Using the relation between gamma and beta (0) functions,
viz.,[231

The network representation of (22) is of the form shown

in Fig. 3 with we finally get
R = R R = 2x(1 fx)c2 + x> ‘*’ (r - 2 + “>R
1 9 7
(1 - h)(2 - X) * * * (r - 1 - X) ’
r = 2,3, *. * (23) Similarly,
 270 IEEETRANSACTIONSON CIRCUITTHEORY,SEPTEMBER 1967
Unfortunately, (31) and (32) cannot be reduced further
into closed algebraic forms. A graphical method was there-
fore adopted; for a particular X, the quantity to be
differentiated is plotted against x and the slopes are
evaluated at various points. The results for x = +, 3, $
are shown in Figs. 9 and 10. It may be seen that for
X = 3, we get the II?$! line. For x = Q, we require a line
in which the resistance decreases and the capacitance
increases with distance while the reverse is the case for
X = $. It may also be noted that these lines do not belong
to the class f(x)g(x) = constant

V. METHOD II FOR s-' USING ELLIPTIC

FUNCTION APPROXIMATION
I I1Io.2 We start from the specification that the tangent of

ok0
0 I 2 3 4 5 6-
the phase angle 4 is desired to remain constant at tan $,, = (Y,
where & = XT/~, (XI < 1, within an error of &E over
the frequency band w1 to w2. As discussed earlier, we can
X-
assume, without any loss of generality, 40 and a! to be
Fig. 9. Plot of f(z) given by (31).
positive; this amounts to assuming the desired immittance
to be proportional to s’, 0 < X 5 1. Thus, our phase
characteristic will be of the form shown in Fig. 11(a)
or (b), where we have assumed an equiripple or Chebyshev-
type variation within the desired band, in order that el-
liptic functions can be used to our advantage. Suppose
the desired immittance is F(s) and let

F(s) = p(s) = Ml + Nl (33)

Q(s) M, + N, ’
M, and M, denoting even and N, and N, the odd parts
of P(s) and Q(s), respectively. Then

F(s) =

= (MIM, - N,N,) + (N,M, - M,NJ

M;-N; -pa (34)

Let
E(s) = M,M, - NlNz
and (35)
0 I 2 3 4 5 6- O(s) = N,M, - M,N,.
x-
Then
Fig. 10. Plot of g(z) given by (32).
(36)
I
is the tangent function and the phase 4(o) of the im-
s0 s(x) dx = )=I
2 (2~ - 1) ,,:‘“;$,, :’ x>
mittance F(jw) is given by
(30)
r Wh z - A + 1)
tan4(w) = Loo. (37)
(1 + lM(2 - 1, z + A)’ j Ww)
Hence,
The odd rational function of degree n which gives a real
frequency phase behavior depicted in Fig. 11 can be
f(x)=2g[ppp(y;p] (31) expressed asus
and
(3%
(32) s = wlsc(u; k), (39)
DUTTA ROY: REALIZATION OF CONSTANT-ARGUMENT IMMI’M’ANCE 271

tan Q To derive the p.r. immittance function F(s) starting

from T(s), we first form
X(s) = O(s) + E(s) = fYs>Q(--4. (44)
Now SC = sn/cn so that from (36), (38), and (44), we get
P(s)Q(-s) = (constant).

First, consider the en function. Its zeros are all on the

real axis, symmetrically disposed with respect to the
jw axis. P(s) should be assigned the left-half plane zeros
and Q(-s) the right-half plane zeros. Forming Q(s)
from Q(-s) would give the game zeros to Q(s) as those
of P(s). Thus in F(s), the contribution due to the zeros
of the cn function will cancel, and it suffices to consider
the roots of the equation

1+ &se& ; k,) = 0.
The range of variation of (Y is from 0 to ~0 so that the
factor 2&;/(1 + 1~:) behaves exactly like the gain con-
stant of a closed-loop feedback system, and the SC func-
tion behaves as the open-loop gain. The positions of the
roots can therefore be visualized by drawing the root
loci, as shown in Figs. 12 and 13. In Figs. 12(a) and 13(a)
the loci on the left-half plane represent those of the zeros
(b) of P(s) while the loci on the right-half plane are those
Fig. 11. Equiripple approximation of tan +o = 01to within an error of the zeros of Q(-s). The locations of the zeros of P(s)
of Z!ZE over the band w1 to 02. and Q( -s) for a particular value of a! are depicted in
Figs. 12(b) and 13(b). The mirror images of the zeros
of Q(-s) are shown by triangles; these represent the
where SC is tLe Jacobian elliptic tangent function, u is a
zeros of Q(s), and hence the poles of P(s)/&(s). It can
parametric variable, and the modulus k is defined by
be easily seen that the zeros and poles of P(s)/&(s)
k’ = (1 - ky = UJWz. (40) alternate for all values of a. For n even = 2N, P(s)/&(s)
has a pole as its first critical frequency and has N poles
The modulus k, is so chosen that and N zeros, all nonzero and finite. Thus, if at all realizable,
P(s)/&(s) should be an RL admittance or an RC im-
nK’/K = K:/K,. (41) pedance. But the requirement that the phase angle be
K, K’, K,, and K: are the real and imaginary quarter- positive leads to the conclusion that for n even, P(s)/&(s)
periods of elliptic functions with moduli k and k,, respec- is not a realizable imniittance. For n odd = 2N + 1,
tively; k: is related to E and a by the first critical frequency of P(s)/&(s) is a zero with
N + 1 zeros and N poles and hence a pole at infinity.
1 - 1c: Combining these with the fact that the angle is positive,
j-q-g = f* (42) we see that P(s)/&(s) is realizable as the admittance of
an RC or the impedance of an RL network. In either case,
T(s), as given by (38), has simple zeros and poles which a canonical realization would require n + 1 number of
are interlaced along the real s axis. The locations are elements.
given by the equation The relation between E/LY expressed as a percentage
s = fw,sc(rK/~; k). and the frequency ratio wZ/w, of the approximation range
(43)
is exactly the same as that given by OrchardL’8’ for
Odd values of T locate the poles while even values, in- networks with a constant imaginary part, in the form
cluding zero, gives the zeros. When n is even, T(s) has of curves with n as a parameter. The values of 7
n number of finite poles and (n - 1) number of finite (-45O f lo, -30’ f lo, and -60’ f 1.5’) as a func-
zeros (hen&, a zero at infinity). When n is odd, T(s) tion of N (= number of elements) can be obtained from
has (n - 1) number of finite poles and n number of these and have been plotted in Fig. 4 (curve 10) and
finite zeros (hence, a pole at infinity). Fig. 8 (curves 5 and 6, respectively).
272 IEEETRANSACTIONSON CIRCUITTHEORY,SEPTEMBER 1967
” EVEN=ZN
procedure given by Orchard.[‘71 Once the poles and zeros
+J have been obtained, further calculations are necessary
t to obtain the element values in an actual network realiza-
(a)
tion of this immittance. Unless one starts with a large
number of digits, errors accumulate at each step of nu-
merical computation and there is no guarantee that the
final network will meet the desired specification.
On the other hand, the other methods start from the
117versus N curve and the element values are simply
obtained from the quotients of the particular CFE used.
The second best method for s-l” is based on the rational
function approximation of sl” provided the number of
elements that can be used is greater than 6. For N < 6,
the method based on the CFE of tanh 0 gives a greater
frequency ratio. The low- and high-frequency approxima-
tions of tanh 0 do not differ much in the q versus N
curve. The third best method for s-l” is that of employing
lattices where the number of elements should be an
integral multiple of 4. The other methods for s-“~ require
a rather excessive number of elements for even a decade
Fig. 12. Root loci of (46) with n even.
of frequency ratio. Of the low- and high-frequency ap-
proximations based on the CFE of (1 + srl)*‘, the latter
n ODD= 2N+I appears to be slightly better with regard to frequency
iw ratio; but the former has the distinctive feature of having
-T (al a series-R, shunt-C ladder, and as such, can be used to
derive a distributed structure. As was pointed out earlier,
the distributed structures derived in this paper can be
easily fabricated and will be more attractive from con-
siderations of cost, size, and reliability.

ACKNOWLFDGMENT
The author wishes to thank Prof. B. A. Shenoi and
R. Srinivasagopalan for helpful discussions on the subject
matter of this paper.

APPENDIX
AN ALTERNATIVE DERIVATION OF STEIGLITZ-PULLEN
FORMULAS AND DEMONSTRATION OF RC REALIZABILITY
Khovanskii”” (p. 108) gives the following CFE for x1”:

Fig. 13. Root loci of (46) with n odd.

X II2 = a + (G)/[l + (G)/[l

+ (*)/[l + .-. (47a)

VI. CONCLUDING DISCUSSIONS L
Of all the methods discussed so far, Figs. 4 and 8
=a+1 la-i+1/[20+1/[3
show that the elliptic function approximation gives the /L x-a
best result, in the sense that for the same number of
+ 1/[2a + **. (47b)
elements, it gives the highest frequency ratio r], or for
the same f, elliptic function approximation requires the Ku01141 used (47b) to synthesize a resistive ladder net-
least number of elements. The difficulties in employing work whose input impedance is the square root of an
this approximation will be primarily computational. The integer n; for realizability, n > a”. If in (47b) x is re-
determination of pole-zero positions from (46) will itself placed by the complex frequency variable s, then the
be a difficult experience in numerical solution of a tran- realization would require a negative resistance. Thus,
scendental equation. Further the values of elliptic SC the CFE’s of (47) do not seem to be useful for realization
functions required in this exercise lie outside the range of a fractional inductor or capacitor. Let us, however,
of existing tables and the first thing to do would be to form the first few convergents of (47a) by using the
construct a table of such elliptic functions following the formula
DUTTA ROY: REALIZATION OF CONSTANT-ARGUMENT IMMITTANCE 273

P
2 _ boP,-l + a,P,-, Formula (55) contains both Steiglitz’s1211 odd formula and
n 2 2, (43) Pullen’s’1e1 even formula. The former can be derived by
Qn - bnQn-1+ anQn-*’
carrying the following steps: put n = 2N, multiply both
where P,/Q, is the nth convergent of the CFE
numerator and denominator by seN; put N - r = P,
b. + alAb, + a2/[b2 + . -- + aJ[b, + . . . . (49) change the limits accordingly and use the identities

Rewrite (47a) with z replaced by s in the following form:

sl/2 = a + (s - a”)/[2a + (s - a2)/[2a + - - . . (50) K:i$ =r23

Then comparing (49) and (50) and using (48), we find

b, = a, al = a, = . . . = a, = . . . = s - a’, b, = b, = . . . = b, = . . . = 2a,
(51)
PO/Q0= a, PI/Q1 = (s + a2)l(2a), P,/Q, = (a3 + 3as)/(3a2 + 4,
... ... ... -*- .

By continued fraction development starting from the and

lowest powers, each of these convergents is found to be
RC admittance or RL impedance realizable. For example,
The result is the inverse of Steiglitz’s formula, as expected
2 = l/[i + l/[$s + l/[$$ because his formula approximates s-l”, while (55) ap-
proximates Y. The steps to be carried out for deriving
+l/[~~~~/[~+~/[~~111111
(52) Pullen’s formula are as follows: put n = 2N - 1, multiply
both numerator and denominator by smN,put N - r = p,
which obviously represents a series-C, shunt-R ladder. change the limits accordingly, use the identities
The input impedance of this network is then an RC im-
pedance approximation of sml”.
Let us now investigate whether the nth convergent (2N2: a,) = f;) and (2N “; + 1) = (2p2: 1> 7
is also RC admittance realizable or not. From the first divide both sides by s, put p - 1 = p’ in the denominator
few convergents, we guess the following formula for the and change the limits accordingly. The result is Pullen’s
nth convergent: formula.
To demonstrate the RC admittance realizability of (55),
= g (” i 1)an-2r+1sr p = q = n/2 for n even assume all the convergents up to the nth to be RC admit-
P
and (53) tance realizable. Let n be even; then (n + 1) is odd and
if $ (z. z ~)ane2’sr
P = (n + 1)/2,
P n+l (‘g’ (” : 2)sr =-. N,+,(s)
q = (n - 1)/2 forn odd. (56)
Q = 2 (; 1 f)$
-2 a+lts)
Following Halijak,“’ the usual method of induction may
now be employed and it can be easily shown that if (53)
is correct for convergents up to the nth, where n is an Let P,/Q, = N,(s)/D,(s); then comparing (55) and (56),
arbitrary even or odd number, it is also true for (n + 1)st it can be easily shown that
convergent. Since it is true for n = 0, 1, 2, . . . 5, it is R+l(s> = N,(s) + Q,(s)
therefore true for any value of n. (57)
Formula (53) can be written in the form N,+,(s) = N,(s) + SD&).
Thus
P
-=n+l
Q?L+, 1 + D,;s),N,(s) + 1 + i&D.(s). @)
N,(s)/D,(s) is, by hypothesis, an RC admittance; hence,
Obviously, a is a frequency and magnitude scaling param- D,(s)/N,(s) is an RC impedance and so also 1 + D,,(s)/
eter, and it suffices to consider the expression for a = 1, N,,(s). Thus the first term in (58) is an RC admittance.
VlZ., Similarly, it is easy to see that the second term in (58) is
s times an RC impedance; i.e., it is also an RC admittance.
2 (lci,l)srp = q =n/2forneven Thus the (n + 1)st convergent is an RC admittance if
P
-?A= the nth convergent is so. Similar proof may be given for
Qn $ (2”, I :)sr p = (iY$ 1)/2, p = (n - 1)/2 (55) n odd, and all the convergents are RC admittance re-
for n odd. alizable.
 274 IEEE TRANSACTIONS ON CIRCUIT THEORY, SEPTEMBER 1967
REFERENCES miniaturization,” Royal Radar Establishment, Ministry of Supply,
Malvern, Worcs., England, Tech. Note 654, August 1959.
I11H. Berger, “Generalized nonuniform transmission lines,” ml A. N. Khovanskii, The Application of Continued Fractions and
IEEE Trans. Czrcuit Theory, vol. CT-13, pp. 92-93, March 1966. Their Generalizations to Problems in Approximation Theory, P. Wynn,
121T-I. W. Bode, Network S?ynthesis and FeedbaclcAmplifier Design. Transl. Groningen, The Netherlands: Noordhoff, 1963.
New York: Van Nostrand, 1945, ch. 14. ml F. F. Kuo, “The use of convergents in the synthesis of mth
131G. E. Carlson and C. A. Halijak, ‘Simulation of the fractional root ladder networks,” IEEE Trans. Circuit Theory (Correspondence)
derivative operator + and the fractronal integral operator l/d,” vol. CT-12, pp. 442-444, September 1965.
Ka;m?te U. Bull+, vol. 45, pp. l-22, July 1961. 1161
R. M. Lerner, “The design of constant-angle or power-law
“Approxrmatlon of fixed Impedances,” IEEE Trans. magnitude impedance,” IEEE Trans. Circuit Theory, vol. CT-14
f’%rirt ThLory (Correspondence), vol. CT-g, pp. 302303, September pp. 98-107, March 1963.
1161R. Morrison, “RC constant-argument driving point admit-
161- “Approximation of fractional capacitors (l/s)*‘, by a tances,” IRE Trans. Circuit Theory, vol. CT-6, pp. 310317, Septem-
regular Ne’wton process,” IEEE Trans. Circuit Theory, vol. CT-11, ber 1959.
pp. 210-213, June 1964. 11’1H. J. Orchard, “The computation of elliptic functions of
161S. C. Dutta Roy, ‘Some exact steady-state sinewave solutions rational fractions of a quarterperiod,” IRE Trans. Circuit Theory,
of nonuniform RC lines,” Brit. J. Appl. Phys., vol. 14, pp. 378-380, vol. CT-5, pp. 352-355, December 1958.
June 1963. 1181-, “Network functions with a constant imaginary part,”
171-, “Some exactly solvable nonuniform’ RC lines,” IEEE IRE Trans. Circuit Theory, vol. CT-6, pp. 370-374, December 1959.
Trans. Circuit Theory (Correspondence), vol. CT-12, pp. 141-142, 1191K. A. Pollen, “On the use of the Pascal triangle with network
March 1965. functions,” Proc. First Annual Allerton Conf. on Circuit and System
I*1S. C. Dutta Roy and B. A. Shenoi, “Distributed and lumped Theory (University of Illinois, November 1963)) p. 698.
RC realizations of a constant-argument impedance,” J. Franklin lZolW. J. D. Steenaart, “The simulation of coaxial cable by a
Inst., vol. 282, pp. 318-329, November 1966. distributed RC homogeneous transmission line and approximate
lo1C. A. Halijak, “An RC impedance approximant to (l/s),‘2,” realization of its 20,” IEEE Trans. Circuit Theory (Correspondence),
IEEE Trans. Circuit Theory (Correspondence), vol. CT-11, p. 495, vol. CT-11, pp. 502-504, December 1964.
December 1964. ml K. Steiglitz, “An RC impedance approximant to s-“~,” IEEE
1101 C. A. Hesselberth, “Synthesis of some distributed RC net- Trans. Circuit Theory (Correspondence), vol. CT-11, pp. 160-161,
works,” Univ. of Illinois, Urbana, Ill., Coordinated Science Lab. March 1964.
Rept. R-164, ch. 4, August 1963. lZ21H. S. Wall, Analytic Theory of Continued Fractions. New York:
IllI E. Hille, Functional Analysis and Semi-Groups. New York: Van Nostrand, 1948.
Am. Math. Sot., 1948, sec. 21.12, p. 439. 1231E. T. Whittaker and G. N. Watson, A Course on Modern
llel C. Helm, “Distributed resistor capacitor networks for micro- Analysis. London: Cambridge University Press, 1927, ch. 12.

Parallel Interconnection of &Port Networks

ABRAHAM LEMPEL, MEMBER, IEEE, AND ISRAEL CEDERBAUM, FELLOW, IEEE

Absfract-In this paper, parallel interconnection of n-port net- tion for n-ports with n > 2. As far as the sufficiency
works without transformers is considered, and a necessary and part of this condition is concerned, the generalization is
sufiicient condition is given, under which the short-circuit admit-
tance matrix (Y matrix) of the resultant n-port is equal to the sum quite simple. However, as shown in the sequel, when the
of Y matrices of the component n-ports. The given condition applies necessity part is considered, an essential difference turns
to nonsingular n-port networks containing a finite number of linear, out to exist between the cases n = 2 and n > 2. Con-
lumped, and reciprocal R, L, and C elements. A possible application sequently, the reasoning on which the proof of the original
of the results to the RLC n-port synthesis problem is indicated. test for 2-ports is based, does not yield itself to a
straightforward generalization. The techniques applied
I. INTRODUCTION in this paper use matrix algebra rather than physical
ARALLEL interconnection of n-port. networks arguments involving current-voltage relationships. It is
without transformers is considered and a simple believed that the results obtained here for general n-ports
P algebraic condition is given in this paper under will also contribute to a better understanding of the n-
which the short-circuit admittance matrix (Y matrix) port synthesis problem.
of the resultant network is equal to the sum of Y matrices The n-ports considered throughout, are supposed to be
of the component networks. nonsingular, with a finite number of linear, lumped, and
To the best knowledge of the authors, the only published reciprocal R, L, and C elements which may assume any
result on this subject is the well-known Brune’s test”’ finite real value. The vertices at which the ports are
for interconnection of 2-ports. The results obtained here connected are called terminals. The number of terminals
are, to a certain extent, a generalization of Brune’s condi- and distinct terminal pairs are denoted by t and m =
+t(t - 1)) respectively. The number of terminals is at
Manuscript received January 3, 1967; revised June 6, 1967. least n + 1 and at most 2n. An n-port network is called
The authors are with Technion-Israel Institute of Technology,
Haifa, Israel. reduced if all of its vertices are terminals. The reduction