ADDITIONAL TABLES FOR DESIGN OF OPTIMUM

LADDER NETWORKS *
BY
LOUIS W E I N B E R G I

P a r t H**
IV. MAXIMALLY FLAT TIME DELAY: BESSEL POLYNOMIALS (7, 8, 9)i

The preceding two sets of polynomials were used in the approxima-

tion of a desired magnitude characteristic. However, the nonlinear
phase characteristic of both approximations and the resulting variation
of the time delay preclude their use where a constant time delay is a
paramount requirement. For such time-delay filters an excellent ap-
proximation is given by the use of B e s s d polynomiads. This approxi-
mation yields a maximally flat time delay along with a low-pass magni-
tude characteristic.
The Bessel polynomials in the variable 1/s are defined by
- (n + k)!
y.(1/s) = ~oX = (n -- k) w.k! (2s) k" (7)

The polynomials of interest are derived from the above as

h.(s) = s"y,,(1/s)

-~- ~ aks k. (8)

k~O

The transfer function is given by

H
(9)
- h.(s)"

The constant H is equal to a0 for a ladder terminated in a one-ohm

resistance, that is, for this configuration at s = 0, Z~I(0) is unity. This
transfer function has a maximally flat time delay. By this is meant
that the time delay ta is given by a function of the form
dO to(bo -[- b~o~~ + b2o# -at - . . . 2r- b,,_1¢o 2'~-~) (10)
ta = - d--~ = bo + bio~ ~ "at- b2o~4 + ' " + b._lo~ ~ - ~ + b.o~2~'
* This paper is based on the a u t h o r ' s report with the same title, Technical Memorandum
No. 434, Hughes Research Laboratories, Culver City, Calif.
1 Research Laboratories, Hughes Aircraft Co., Culver City, Calif.
** P a r t I appeared in this JOURNAL for July, 1957.
2 The boldface numbers in parentheses refer to the references appended to P a r t II of this
paper.
r27
128 Louis WEINBERG [J. F. I.

where 0 is the phase of Z21(joa), and to is the zero-frequency time delay.

It is noted that the first (n - 1) coefficients of the denominator are
equal to the corresponding coefficients of the numerator. Therefore the
Maclaurin series for td obtained by dividing the numerator by the de-
nominator, namely,

-d-:
dO = to
( 1 b'co2~blb~co~n+2
+
.... ) (11)

will have the first (n - 1) derivatives of td (considered as a function of

c0*) at o~ = 0 equal to zero. Thus the time delay is as flat as possible
in the vicinity of co = 0; hence the term maximally flat time delay.
The delay is very closely equal to to, the zero-frequency value, up to a
certain frequency (which is an increasing function of n), and then de-
clines smoothly for values greater than this frequency.
To determine the value of n to use for satisfying a specific require-
ment, it is necessary to have expressions for the magnitude and time
delay. We give here the exact expressions in terms of Bessel functions
of half an odd integer ;3 but most often the values of the magnitude and
time delay for varying n given in Table V I I I suffice so that the need
TABLE VIII.--Significant Values of u for Time Delay and Loss Characteristic
of a Maximally Flat Time-Delay Network.
a) Time-DelayTable: Giving Frequencies (u) b) Loss (L = - - 2 0 log I Z~(ju) I, in db) Table: Giving
at which Time Delay Deviates a Specified Frequencies (u) at which Loss is a Specified Number
Value from its Zero-Frequency Value of db Down from its Zero-Frequency Value
u f o r l % uforl0% ufor20% ufor50% uforl/50uforl/20uforl/10uforl/5 uforl/2 uforl ufor3
n deviation deviation deviation deviation n db db db db db db db
1 0.1O 0.34 0.50 1,00 1 0.07 0.11 0.14 0.21 0.35 0.51 1.00
2 0.56 1.09 1.39 2.20 2 0.11 0.18 0.26 0.36 0.57 0.80 1.36
3 1.21 1.94- 2.29 3.40 3 0.14 0.23 0.34 0.48 0.75 1.05 1.75
4 1.93 2.84 3.31 4.60 4 0.17 0.28 0.40 0.56 0.89 1.25 2.13
5 2.71 3.76 4.20 5.78 5 0.20 0.32 0.45 0.64 1.01 1.43 2.42
6 3.52 4.69 5.95 6.97 6 0.22 0.36 0.50 0.71 1.12 1.58 2.70
7 4.36 5.64 6.30 8.15 7 0.24 0.39 0.54 0.77 1.22 132 2.95
8 5.22 6.59 7.30 9.33 8 0.26 0.41 0.59 0.83 1.31 1.85 3.17
9 6.08 7.55 8.31 10,50 9 0.28 0.44 0.62 0.88 1.40 1.97 5.39
I0 6.96 8.52 9.33 IL67 10 0.30 0.47 0.66 0.93 1.48 2.08 3.58
11 7.85 9.49 10.34 12.84 II 0.31 0.49 0.69 0.98 1.55 2.19 3.77

for using the exact analytical form is eliminated. The time delay is
given by

us ~ {J2-~-du) + J'~+t(u)}

3 Tables for these functions are given in "Tables of Spherical Bessel Functions," 2 vols.,
NBS, Math. Tables Project, Columbia University Press, 1947. The particular combination
o f t h e s p h e r i c a l B e s s e l f u n c t i o n s t h a t o c c u r s i n t h e m a g n i t u d e a n d p h a s e f u n c t i o n s is t a b u l a t e d
i n T a b l e 13 o f " S c a t t e r i n g a n d R a d i a t i o n f r o m C i r c u l a r C y l i n d e r s a n d S p h e r e s , T a b l e s o f
A m p l i t u d e a n d P h a s e A n g l e s , " Office o f R e s e a r c h a n d I n v e n t i o n s , U . S. N a v y D e p a r t m e n t ,
July, 1946.
Aug., 1957.] OPTIMUM LADDER NETWORKS 12 9

and the m a g n i t u d e is
H (13)
JZ.(ju)] =
un+l { ~u EJ~-~-~(u)+ J2,+~(u)-]}
T h e loss in db, L = - 20 log tZ21(ju)J, tends to the Gaussian form with
increasing n,
10 u 2
L = (2n - 1)In 10" (14)

In the a b o v e formulas u is the normalized frequency variable w/o~0 and

J is a Bessel function.
Use of Eq. 14 gives the 3-db b a n d w i d t h as

U3db ~----~ / ( 2 n -- 1) In 2, (15)

which approximation is good for n >_ 3.

In T a b l e V I I I are given values of u for four significant points on the
time-delay curves and seven significant points on the loss curves. T h e
element values corresponding to the values of n of 1 through 11 are
given in T a b l e IX.
Example 4.1. Design a ladder n e t w o r k with a d e l a y of 0.1 ~ sec and
a c o n s t a n t loss (not greater than 1 db) up to 2.7 m c / s . T h e n e t w o r k
is to be t e r m i n a t e d in a load resistance of 2000 ohms and is to be driven
b y a current source.
Since to = l/w0 = 0.1 sec., then w0 = 107. F o r f = 2.7 m c / s ,
w = 5.47r X 106, and w/o~0 = 0.547r, which is a p p r o x i m a t e l y 1.7.
Using T a b l e V I I I (b) for u = 1.7, we see t h a t for n = 7 the loss is
less than 1 db. N o w b y using T a b l e V I I I ( a ) , it is shown t h a t the time
delay for n = 7 is c o n s t a n t at this frequency.
Consulting T a b l e I X ( a ) we find the element values for n = 7; the
unprimed values are used since n is odd and the input is a current source.
W e remove the normalization b y multiplying C's b y 1/Rwo = 0.5 X
10 -1°, and L ' s b y R/wo = 2 X 10 -4, and thus obtain the final n e t w o r k
given in Fig. 9.

ZI 6.97,,10-5 4.5~4i0-5 2.11xiO-s

~ !IO
-|x It|o-It xlo-I~' x10'-42 2000
I
E2

FIG. 9. Time-delay ladder obtained in Example 4.1.

I30 L o u I s WEINBERG [J. F. I.

TABLE IX.--Element Values (in ohms, henrys, farads) of a Normalized

Maximally Flat Time-Delay Network.

Value of n CI or L~' l.~ or C2' C3 or L3' L~ or C4' Cs or Ls' L~ or C6' C7 or L7' Ls or C8' C9 or Lg' Li0 or Ci0' Cn or Ln'

~r=O
1 1.0000
2 0.3333 1.0000
3 0.1667 0.4800 0.8333
4 0.1000 0.2899 0.4627 0.7101
5 0.0667 0.1948 0.3103 0.4215 0.6231
6 0.0476 0.1400 0.2246 0.3005 0.3821 0.5595
7 0.0357 0.1055 0.1704 0.2288 0.2827 0.3487 0.5111
8 0.0278 0.0823 0.1338 0.1806 0.2227 0.2639 0.3212 0.4732
9 0.0222 0.0660 0.1077 0.1463 0.1811 0.2129 0.2465 0.2986 0.4424
10 0.0182 0.0541 0.0886 0.1209 0.1549 0.1880 0.2057 0.2209 0.2712 0.4161
11 0.0152 0.0451 0.0741 0.1016 0.1269 0.1499 0.1708 0.1916 0.2175 0.2639 0.3955

b) r = 1/8
1 9.0000
2 8.6533 0.0433
3 7.1426 0.0615 1.3652
4 6.0700 0.0589 2.3569 0.0127
5 5.3229 0.0535 2.5118 0.0246 0.5401
6 4.7803 0.0484 2.4267 0.0283 1.1309 0.00601
7 4.3691 0.0442 2.2790 0.0288 1.3738 0.0133 0.2881
8 4.0462 0.0407 2.1256 0.0280 1.4536 0.0168 0.6627 0.00350
9 3.7848 0.0378 1.9841 0.0267 1.4558 0.0184 0.8666 0.00830 0.1788
I0 3.5682 0.0354 1.8591 0.0254 1.4215 0.0189 0.9718 0.0111 0.4348 0.00228
II 3.3850 0.0334 1.7502 0.0240 1.3710 0.0188 1.0191 0.0128 0.6014 0.00589 0.1159

¢) r = 1/4
1 5.0000
2 4.6409 0.0898
3 3.7994 0.1258 0.6973
4 3.2221 0.1198 1.1956 0.0258
5 2.8247 0.1084 1.2690 0.0498 0.2731
6 2.5375 0.0980 1.2231 0.0571 0.5703 0.0121
7 2.3202 0.0893 1.1470 0.0580 0.6915 0.0268 0.1451
8 2.1496 0.0823 1.0689 0.0563 0.7306 0.0338 0.3333 0.00704
9 2.0114 0.0764 0;9973 0.0537 0.7310 0.0369 0.4354 0.0167 0.0899
lO 1.8967 0.0716 0.9342 0.0509 0.7132 0.0379 0.4878 0.0224 0.2184 0.00459
11 1.7999 0.0676 0.8794 0.0482 0.6875 0.0377 0.5112 0.0256 0.2998 0.0115 0.0603

d) t = 1/3
I 4,0000
2 3.6330 0.1223
3 2.9601 0.1700 0.5298
4 2.5075 0.1613 0.9046 0.0347
5 2.1981 0.1457 0.9577 0.0669 0.2063
6 1.9750 0.1316 0.9217 0.0765 0.4300 0.0163
7 1.8064 0.1199 0.8636 0.0776 0.5207 0.0358 0.1093
8 1.6740 0.1104 0.8044 0.0753 0.5497 0.0453 0.2509 0.00942
9 1.5667 0.1026 0.7503 0.0718 0.5496 0.0494 0.3275 0.0223 0.0676
I0 1.4777 0.0962 0.7027 0.0680 0.5360 0.0506 0.3668 0.0299 0.1642 0.00614
ii 1.4024 0.0907 0.6615 0.0644 0.5165 0.0504 0.3842 0.0342 0.2252 0.0153 0.0455

e) r = 1/2
I 3.0000
2 2.6180 0.1910
3 2.1156 0.2613 0.3618
4 q.7893 0.2461 0.6127 0.0530
5 1.5686 0.2217 0.6456 0,I015 0.1393
6 1.4102 0.1999 0.6196 0.i158 0.2894 0.0246
7 1.2904 0.1821 0.5797 0.1171 0.3497 0.0542 0.0735
8 1.1964 0.1676 0.5395 0.1135 0.3685 0.0683 0.1684 0.0142
9 1.1202 0.1558 0.5030 0.1081 0.3680 0.0744 0,2195 0.0336 0.0453
10 1.0569 0.1460 0.4710 0.1024 0.3586 0.0763 0.2456 0.0450 0.1100 0.00925
11 1.0033 0.1377 0.4433 0.0970 0.3454. 0.0758 0.2570 0.0515 0.1503 0.0228 0.0309

p 'r = 1
2.O0OO
2 1.5774 0.4226
3 1.2550 0,5528 0.1922
4 1.0598 0.5116 0.3181 0.1104
5 0.9303 0.4577 0.3312 0.2090 0.0718
6 0.8377 0.4116 0.3158 0.2364 0.1480 0.0505
7 0.7677 0.3744 0.2944 0.2378 0.1778" 0.1104 0.0375
8 0.7125 0.3446 0.2735 0.2297 0.1867 0.1387 0.0855 0.0289
9 0.6678 0.3203 0.2547 0.2184 0.1859 0.1506 0.1111 0.0682 0.0230
10 0.6305 0.8002 0.2384 0.2066 0.1808 0,1539 0.1240 0.0911 0.0557 0.0187
11 0.5989 0.2834 0.2243 0.1954 0.1739 0.1528 0.1296 0.1039 0.0761 0.0465 0.0154
Aug., 1957.] O P T I M U M LADDER NETWORKS 131
V. NORMALIZATION,DUALITY,RECIPROCITY T ~ O R g M ; FREQUBNCY TRANSFORMATIONSAND
TRANSFORMATION OF SYMMETRICALNE'rWORKS
Normalization
The element values in the tables are normalized with respect to the
load resistance R1 and the radian frequency. In other words, the value
of R1 is considered as one ohm and that of the cutoff frequency (or w0 =
1~to for the time-delay networks) is one radian per second. These fre-
quency and impedance normalizations m a y be removed simply.
Since the impedance of the three different kinds of elements appear-
ing in a network is given respectively by R, Ls, and 1/Cs, we note t h a t
if the frequency is multiplied by a constant the resistance is unaffected,
but t h a t to maintain the impedance of the inductance and capacitance
invariant, it is necessary to divide L and C by the same constant. This
provides the simple rule for removal of the frequency normalization : to
raise the radian frequency oa = 1 to oa = ~0~,divide all L's and C's in the
network by we. On the other hand, to raise the impedance level by a
factor H we m u s t multiply the impedance of each type of element by
this factor, t h a t is, multiply every R and L in the network by H, and
divide every C by H. T h u s we see only simple multiplications are
involved.
The two rules m a y be combined into one operation: to raise the
radian frequency to w, and the impedance level by H, we multiply every
resistance by H, every inductance by H/w,, and every capacitance by

Duality
T h e dual of a ladder network m a y always be realized simply. T h e
impedance of every series arm is replaced by the admittance of a shunt
arm, and vice versa. In simpler terms, this means t h a t every capaci-
tance of C farads is replaced by the dual element which is an inductance
of C henrys, every inductance of L henrys is replaced by a capacitance of
L farads, and every resistance of R ohms becomes a conductance of R
mhos; if the original element is a series arm then the dual element be-
comes a shunt arm, whereas if the original element is a shunt arm then
the dual element is a series arm. For example, the dual of the network
in (a) of Fig. 10 is given by the one in (b).

4 3
I• I 2

(o) (b)
FIG. 10. Ladder network and its dual (values in ohms, henrys, and farads).
132 LouIs WEINBERG [J. F. I.

W h a t are the characteristics of the dual network with respect to

that of a given network? The impedances (admittances) of one net-
work (both transfer and driving-point) become admittances (imped-
ances) of the other. Thus in Fig. 10(a) the input is a voltage source
and the output a current so that the transfer function is the admittance
Y~I = I2/EI. In the dual given by Fig. 10(b) the transfer impedance
Z'21 = E2'/I1' is the same rational function as Y21 of (a).
It is therefore clear that the primed and unprimed values lead to dual
networks.
Reciprocity Theorem
Often a network designed by the use of the tables does not have the
configuration demanded in a particular problem. For example, a shunt
capacitance may be desired at the output and a resistance at the input,
but the network obtained has the form shown in Fig. 11 (a). By the

Ii2~ L3 L! II Lt L3

(ol (bl
FIG. 11.--Ladder network and one obtained from it by use
of reciprocity theorem.

use of the reciprocity theorem the network of Fig. 11 (b) with the de-
sired configuration may be obtained.
The reciprocity theorem states that the transfer impedance (or
transfer admittance) remains unchanged if the excitation and measur-
ing instrument change places. Thus in Fig. 11 (a) we have the transfer
impedance
Z21 = E~ = p(s_)) (16)
I1 q(s)'
where the excitation is a current source I1 flowing into the input termi-
nals and the output is a voltage (measured by a voltmeter across R).
Now if the current source is placed across R and the voltmeter placed
across C4, then the conditions of the reciprocity theorem have been satis-
fied. Thus the transfer impedance of Fig. 11 (b) is also equal to p/q.
It is therefore clear that by use of reciprocity a whole set of new net-
work configurations may be obtained.
Frequency Transformations (10)
The tables give the element values for low-pass filters. However,
for the Butterworth and Tschebyscheff cases corresponding character-
istics may be obtained for the high-pass, band-pass, and band-elimination
Aug., 1957.] OPTIMUM LADDER NETWORKS 133

filters by the use of transformations of the frequency variable. These

transformations do not work for the Bessel-polynomial case because the
distortion of the frequency-variable scale makes the phase characteristic
nonlinear as a function of the new frequency variable.

High-Pass Filters
A normalized low-pass filter characteristic is shown in Fig. 12(a);
the corresponding high-pass characteristic is given in Fig. 12(b). T h e

I v,,.al
I ~. "I~ I~!7 I"....
1 O~ t O~
(a) (b)
Fm. 12. Low-pass characteristic and the corresponding high-pass
one obtained by a frequency transformation.

latter characteristic m a y be obtained from the former by the use of the

transformation s' = 1/s. Since by use of this transformation the im-
pedance of an inductance Ls becomes the impedance L/s', the impedance
of a capacitance 1/Cs becomes s'/C, and the value of a resistance remains
unchanged, a simple rule for converting a low-pass ladder network to a
high-pass one m a y be formulated. The rule is: replace every induc-
tance of L henrys by a capacitance of 1/L farads ; replace every capaci-
tance of C farads by an inductance of 1/C henrys; and leave the resis-
tances unchanged. T h u s if the network in Fig. 13(a) has a low-pass

2 112

c
(a) 0
(b)
FIG. 13. Low-pass network and its corresponding high-pass network.

characteristic, then the corresponding high-pass network is given in

Fig. 13 (b).

Band-Pass Filters
A low-pass filter of bandwidth coc m a y be converted to a band-pass
filter of bandwidth coc = cob - coa by use of the frequency transformation

s = + co0 (17)
St
134 Louis WEINBERG [J. F. I.

T h u s the right-hand side of Eq. 17 is substituted for every s in the

transfer function. Here ~0b is the upper frequency limit and, ~, is the
lower frequency limit of the band, while ~0 is the center frequency of the
band. The band limits have geometric s y m m e t r y about the center
frequency, t h a t is, ~0~b = ~02.
However, it is not necessary to actually carry out the functional
transformation, since there is a simple rule for converting the low-pass
network to a band-pass one: for each inductance in the network of L
henrys add a capacitance in series with it of value 1/(~00~L) farads; for
each capacitance in the network of C farads add an inductance in par-
allel with it of 1/(u0~C) henrys (that is, the added element always reso-
nates with the original element at the center frequency ~0); leave the
resistances unchanged.
T h e complete process for converting a normalized low-pass filter to a
desired band-pass one m a y be given as the following:
1. Determine the desired bandwidth o~c = ub - ua and the desired
center frequency u02 = ~ac0bfrom the given data.
2. Change the b a n d w i d t h of the low-pass filter to uo.
3. Perform the low-pass to band-pass transformation on the network.
4. Remove the level normalization from the resulting band-pass
filter.
Example 5.1. Design an equal-ripple band-pass filter with the fol-
lowing characteristics :
(a) T h e ripple in the pass band is 1 db.
(b) T h e center frequency is f0 = 1000 cps.
(c) T h e bandwidth f~ measured at 1-db points is 100 cps.
(d) At the frequencies corresponding to three times f, the response
is to be down approximately 50 db.
(e) The network is driven by a current source and should have a
load resistance of 1000 ohms.

L3 C3 LI Ci,

FIG. 14. Band-pass filter for Example 5.1.

In order to design this filter it is not necessary to find the actual fre-
quencies at which the response is down 1 db and 50 db, b u t if we wished
to find t h e m we could use the formulas f , fb = f , ( f , + 100) = 106 and
fs0(fs0 -t- 300) = 108, where f , is the lower 1-db frequency and f60 is the
lower 50-db frequency.
Aug., I957.] OPTIMUM LADDER NETWORKS 135

F r o m Table V we find t h a t the 1-db ripple corresponds t o , = 0.5088.

We now calculate n and find t h a t n = 4 yields approximately 49-db
a t t e n u a t i o n at o~ = 3. Therefore using n = 4 and the primed values of
Table V(a), we find the element values:

L I ' = 1.0495 L a ' = 1.9093

C ~ ' = 1.4126 C 4 ' = 1.2817

T h e bandwidth is now changed to o~o = 27r )< 100 by dividing the

above values by o~. T h e network is then converted to the band-pass
form and the impedance level raised to 1000 ohms. T h e final network
given in Fig. 14 has the element values (in ohms, henrys, and farads) :
R = 1000 L3 = 3.04
L1 = 1.67 C3 = 8.33 × 10 -9
C1 = 1.52 )< 10 -8 L4 = 1.15 )< 10 -3
L~ = 1.41 X 10 -3 Ca = 2.20 × 10 -e
C2 = 2.25 × 10 -6

Band-Elimination Filters
T h e transformation from a low-pass to a band-elimination character-
istic is given by
$I
s = + (18)

As for the band-pass filter the transformation can be achieved by direct

operation on the low-pass network. T h e rule follows:
(a) Add a capacitance in parallel with each inductance in the low-
pass network; the value of the capacitance is 1/(o~0~L), where L is the
value of the original inductance.
(b) Add an inductance in series with each capacitance of the net-
work; the value of the inductance is 1/(,002C), where C is the value of
the original capacitance.
(c) Since the resistances are unaffected by the transformation, their
values are not changed.

Transformation of Symmetrical Networks

It has been pointed out t h a t the B u t t e r w o r t h and Tschebyscheff
networks obtained for r = 1 and n odd are symmetrical. This sym-
m e t r y allows a n y specified resistance ratio to be obtained simply; the
m e t h o d used transforms the symmetrical network to an unsymmetrical
one with the desired resistance ratio.
If the symmetrical network is divided as shown in Fig. 15, then the
I36 Louis WEINBERG [J. F. I.

over-all transfer impedance is given in terms of the impedances of the

c o m p o n e n t networks by (11)
Z21aZ21b
Z21 - Z~ + Zb" (19)

The subscripts a and b have been used to designate the networks on the
left and right, respectively. But because of the symmetry, the com-
ponent networks are the same and consequently Z21b = Z21a and Zb =
Za. Now suppose it is desired to increase the resistance ratio by r. If
the impedance level of No is multiplied by r, the desired effect will have
been accomplished. But this change also increases Z21, and Z, by r.
Because Zb = Z~, however, the Z2~ of the whole network is not changed

NETWORK NETWORK I
N N

Zo Zo

FIG. 15, Decomposition of a symmetrical network into two network halves.

except by a constant multiplier. For example, if r = 10 then the

transfer impedance before the level change is

= (z,,o), (20)
2Za '
whereas after the change it is
, 10(Z~lo) ~
Z 21 ~
11zo ' (21)

which differs from Eq. 20 only by a constant multiplier.

An analogous situation of course holds for transfer admittances.
T h u s it is possible to obtain two different networks with the same
value of r; one is derived from the table for the desired value of r, and
the second, as indicated above, by means of a transformation of a sym-
metrical network, the symmetrical network being obtained from the
table for r -- 1. For example, for the transfer impedance of a Tscheby-
scheff network with n = 3, r = 1/2, and a 1/10-db ripple, the network
shown in Fig. 16 is obtained from Table I I (e). However, if Table I I (f)
is used the symmetrical network in Fig. 17 (a) results; multiplying the
impedance level of the left half of this network by 1/2 yields the network
Aug., I957.] OPTIMUM LADDER NETWORKS I37

of Fig. 17(b), for which r is now 1/2. Inspection of the networks in

Figs. 16 and 17(b) shows that they differ, even though their transfer

I, 0.60
o

o c

OHMS, HENRYS, FARADS

FI6. 16. Normalized Tschebyscheff network with r = 1/2, n = 3,
and 1/10-db ripple.

I, 1.15
~ o I
o,,

0
,i +,.o, +,.o, l,,. -D
(o)
ZI ~ 0.86
o

1/24 -2.06 1.03 I Ez

(b)

OHMS, HIENRYS, FARADS

FI6. 17. Normalized Tschebyscheff network with n = 3, and 1/10-db ripple; (a) r = 1; (b)
r --- 1/2 achieved by an impedance level change on half the network.

impedances are identical. The reason for this is that the tables are
derived for a network reflection coefficient all of whose zeros lie in only
one half-plane, whereas the network obtained by transformation of the
symmetrical network has the zeros of its reflection coefficient alternating
in the left and right half-planes. This phenomenon has important
implications and is discussed elsewhere (12).
CONCLUSION

The design of three classes of practical networks with resistance

terminations at both ends becomes simple by use of the tables presented
in this paper. The tables give the element values for the normalized
low-pass network with a Butterworth, Tschebyscheff, or Bessel-poly-
nomial characteristic. The low-pass networks that are realized in the
Butterworth and Tschebyscheff cases can also be easily transformed
to serve high-pass, band-pass, or band-elimination functions.
~38 Louis WEINBERG [J. F. I.

In the future tables will be presented for networks with uniform dis-
sipation and for networks whose reflection coefficients possess zeros that
alternate in the left and right half-planes.

Acknowledgment
The author expresses his thanks to Tadayoshi Suzuki, Mathematics
Section, Systems Analysis Department, Hughes Aircraft Company, for
his intelligent programming of the computations.
REFERENCES FOR PART n
(7) W. E. THOMSON,"Delay Networks Having Maximally-Flat Frequency Characteristics,"
Proc. I.E.E., Vol. 96, Part 3, p. 487 (1949).
(8) W. E. T~OMSON, "Networks with Maximally-Flat Delay," Wireless Engineer, Vol. 29,
p. 256 (1952).
(9) L. STORCH, "Synthesis of Constant-Time-Delay Ladder Networks Using Bessel Poly-
nomials," Proc. I.R.E., Vol. 42, p. 1666 (1954).
(10) J. G. TRUXAL, "Automatic Feedback Control System Synthesis," New York, McGraw-
Hill Book Co., 1955.
(11) L. WEINBERG, "Synthesis of Transfer Functions with Poles Restricted to the Negative
Real Axis," J. Appl. Phys., Vol. 24, pp. 207-216 (1953).
(12) L. WEINBERO,"Explicit Formulas for Tschebyscheff and Butterworth Ladder Networks,"
Technical Memorandum No. 459, October 1956, Hughes Research Laboratories, Culver
City, Calif.