Esta é a versão que foi publicada, com pequenas alterações, preparando a versão para a

enciclopédia de RF. A versão que foi enviada para publicação em 1997 aparentemente foi

perdida no “crash” de um HD.

ACMQ, 16/1/2003
Chebyshev filters

Antônio Carlos M. de Queiroz

Electrical Engineering Program-COPPE / Electronic Engineering Department

Federal University of Rio de Janeiro

CP 68504

21945-970 Rio de Janeiro, RJ, Brazil

Any signal can be considered to be composed of several sinusoidal components with different

frequencies, amplitudes, and phases. Filtering is one of the fundamental methods for signal

processing, where the signal is processed by a linear system that changes the amplitudes and

phases of these components, but not their frequencies. In a simple form, this processing can

be used to let pass or to reject selected frequency bands, ideally with no attenuation at the

passbands and infinite attenuation at the stopbands. This article discusses a class of

approximations to this kind of ideal filter, known as Chebyshev filters. It starts with a

discussion on a technique for the derivation of optimal magnitude filters, then discusses the

direct and inverse Chebyshev approximations for the ideal filtering operator, ending with

comments on extensions of the technique. Tables with example filters are included.

The magnitude approximation problem in filter design consists essentially in finding a

convenient transfer function with the magnitude satisfying given attenuation specifications.

Other restrictions can exist, as structure for implementation, maximum order, maximum Q of

the poles, etc., but in most cases the problem can be reduced to the design of a normalized

continuous-time low-pass filter, that can be described by a transfer function in Laplace

transform. This filter must present a given maximum passband attenuation (Amax), between w

= 0 and w = wp = 1 rad/s, and a given minimum stopband attenuation (Amin) in frequencies

above a given limit wr rad/s. From this prototype filter, the final filter can be obtained by

frequency transformations, and by continuous-time to discrete-time transformations in the

case of a digital filter (1).

A convenient procedure for the derivation of optimal magnitude filters is to start with the

transducer function H(s) and the characteristic function K(s). H(s), than can also be called

attenuation function, is the inverse of the filter transfer function, scaled to min |H(jw) = 1|.

K(s) is related to H(s) by the equation:

H  j 1  K  j
2 2
(1)
This greatly simplifies the problem, because K(jw) can be a ratio of two real polynomials inw,

both with roots located symmetrically on both sides of the real axis, while H(jw) is a complex

function. K(s) is obtained by replacingw by s/j in K(jw), and ignoring possible ±j or –1

multiplying terms resulting from the operation. The complex frequencies where K(s) = 0 are

the attenuation zeros, and K(s) = ¥ corresponds to the transmission zeros. If K(s) is a ratio of

real polynomials in s, K(s)=F(s)/P(s), H(s) is also a ratio of real polynomials, with the same

denominator, H(s)=E(s)/P(s), and E(s) can be obtained by observing that for s = jw, Eq. (1) is

equivalent to

H s H  s  1  K s K  s  E s E  s   P s P  s   F s F  s  (2)

Because E(s) is the denominator of the filter transfer function, that must be stable, E(s) is

constructed from the roots of the polynomial P(s)P(–s) + F(s)F(–s) with negative real parts.

The desired transfer function is then T(s) = P(s)/E(s).

Chebyshev polynomials

Two important classes of approximations, the direct and inverse Chebyshev approximations,

can be derived from a class of polynomials known as Chebyshev polynomials. These

polynomials were first described by P. L. Chebyshev (2). The Chebyshev polynomial of order

n can be obtained from the expression:

 
C n x  cos n cos  1 x (3)

It is simple to verify that this expression correspond, for –1 £ x £ 1, to a polynomial in x.

Using the trigonometric identity cos (a + b) = cos a cos b – sin a sin b we obtain:

    
C n 1 x  cos n  1cos  1 x  xC n x   sin n cos  1 x sin cos  1 x  (4)
Applying now the identity sin a sin b = ½[cos(a – b) – cos(a + b)] and rearranging, a

recursion formula is obtained:

C n 1 x  2 xC n x   C n  1 x  (5)

For n = 0 and n = 1, we have C0(x) = 1 and C1(x) = x. Using Eq. (5), the series of Chebyshev

polynomials shown in Table 1 is obtained.

The values of these polynomials oscillate between –1 and +1 for x between –1 and +1, in a

pattern identical to a stationary Lissajous figure (3). For x out of this range, cos–1x = jcosh–
1x, an imaginary value, but Eq. (3) is still real, in the form:

  
C n x  cos nj cosh  1 x cosh n cosh  1 x  (6)

For high values of x, looking at the polynomials in Table 1, we see that Cn(x) » 2n–1xn,

growing monotonically. The plots of some Chebyshev polynomials for –1 £ x £ 1 are shown

in fig. 1.

The Chebyshev low-pass approximation

The normalized Chebyshev low-pass approximation is obtained by using:

K  j  C n  (7)

The result is a transducer function with the magnitude given by (from Eq. (1)):

H  j 1   C n 
2 2
(8)

The corresponding attenuation in decibels is:

 
A 10 log 1   C n 
2
 (9)

The parameter e controls the maximum passband attenuation, or the passband ripple.
Considering that when Cn(w) = ±1 the attenuation A(w) = Amax, Eq. (9) gives:


  10 0.1 Amax  1 1/ 2
(10)

Figure 2 shows examples of the magnitude function |T(jw)| in the passband and in the
stopband obtained for some normalized Chebyshev low-pass approximations, with Amax=1

dB. The magnitude of the Chebyshev approximations presents uniform ripples in the
passband, with the gain departing from 0 dB at  = 0 for odd orders and from –Amax dB for

even orders.

The stopband attenuation is the maximum possible among filters derived from polynomial
characteristic functions and with the same Amax and degree (4). This can be proved by

assuming that there exists a polynomial Pn(x) that is also bounded between –1 and 1 for –1 £

x £ 1, with Pn(x) = ±Pn(–x) and Pn(+¥) = +¥, but that exceeds the value of Cn(x) for some

value of x > 1. An approximation using this polynomial instead of Cn(x) in Eq. (7) would be

more selective. The curves of Pn(x) and Cn(x) will always cross n times for –1 £ x £ 1, due

to the maximum oscillations of Cn(x), but if Pn(x) grows faster, they will cross another two

times for x ³1 and x £ –1. This makes Pn(x) – Cn(x) a polynomial of degree n + 2, because it

presents n+2 roots, what is impossible since both are of degree n.

The required approximation degree for given Amax and Amin can be obtained by substituting

Eq. (6) in Eq. (9), with A(wr) = Amin and solving for n. The result, including a

denormalization for any wp is:

 cosh  1 
n (11)
cosh  1  r /  p 

where is convenient to define the constant:

1/ 2
 10 0.1 Amin  1 
  0.1 A  (12)
 10
max
 1 

The transfer functions for the normalized Chebyshev filters can be obtained by solving Eq.

(2). For a polynomial approximation, using P(s) = 1, from Eq. (7), it follows that

2
  s 
E ( s ) E ( s ) 1    C n    (13)
  j 

The roots of this polynomial are the solutions for s in

s  s j
C n   cos n cos  1    (14)
 j  j 

Identifying

s
n cos 1 a  jb (15)
j

j
it follows that:  cos( a  jb) cos a cos jb  sin a sin jb cos a cosh b  j sin a sinh b .

1
Equating real and imaginary parts, we have cos a cosh b 0 and sin a sinh b  .

Since cosh x ³ 1, the equation of the real parts gives


a 1  2k  k 0, 1,..., 2n  1 (16)
2

and as for these values of a, sin a = ±1, the equation of the imaginary parts gives
 1
b sinh  1 (17)


applying these results in Eq. (15), it follows that the roots of E(s)E(–s) are:

s k  k  j k k 0, 1,..., 2n  1
 1  2k 1 1
 k  sin sinh sinh  1 
2 n n  (18)
 1  2k 1 1
 k cos cosh sinh  1 
2 n n 

The roots sk with negative real parts (kn) are the roots of E(s). By the expressions in Eq.

(18), it is easy to see that the roots sk are located in an ellipse with vertical semi-axis
1 1 1 1
cosh sinh  1  , horizontal semi-axis sinh sinh  1  , and foci at ±j. The location of the
n  n 

roots can be best visualized with the diagram shown in fig. 3 (3).

Realization of Chebyshev filters

These approximations were originally developed for realization in passive form, and the best

realizations were obtained as LC doubly terminated structures designed for maximum power

transfer at the passband gain maxima. These structures are still important today in high-

frequency filters and as prototypes for active and digital realizations, due to the low

sensitivity to errors in element values. At each attenuation zero, and the Chebyshev

approximations have the maximum possible number of them distributed in the passband,

maximum power transfer occurs between the terminations. In this condition, errors in the

capacitors and inductors can only decrease the gain (5). This causes zeros in the derivative
 T  j /  L, C at all the attenuation zeros, and keeps low the error in all the passband. Table

2 lists polynomials, poles, frequency and Q of the poles, and LC doubly terminated ladder

structures, with the structure shown in fig. 4(a), for some normalized Chebyshev low-pass

filters. Note in the realizations that odd-order filters have identical terminations, but even-

order filters require different terminations, because there is no maximum power transfer at w

= 0, since the gain is not maximum there. With the impedance normalization shown, it is
clear that the even-order realizations have antimetrical structure (one side is the dual of the

other). The odd-order structures are symmetrical.

The Inverse Chebyshev low-pass approximation

This normalized inverse Chebyshev approximation is the most important member of the

inverse polynomial class of approximations. It is conveniently obtained by using the

characteristic function obtained from

F  j    n
K  j    n (19)
P j C n 1 /   C n 1 / 

Where e and g are given by Eq. (10) and Eq. (11). The polynomials F(s) and P(s) are then

F ( s )   s / j 
n

(20)
P( s ) s / j  C n  j / s 
n

Ignoring ±j or –1 multiplying factors in Eq. (20) and renormalizations, F(s) reduces to egsn,

and P(s) to a Chebyshev polynomial with all the terms positive and the coefficients in reverse
order. The magnitude characteristic of this approximation is maximally flat at w = 0, due to

the n attenuation zeros at s = 0, and so is similar in the passband to a Butterworth

approximation. In the stopband, it presents a series of transmission zeros, at frequencies that

are the inverse of the roots of the corresponding Chebyshev polynomial. Between adjacent
transmission zeros, there are gain maxima reaching the magnitude of –Amin dB. Without a

renormalization, the stopband starts at 1 rd/s, and the passband ends where the magnitude of

the characteristic function, Eq. (19), reaches e

1 1
p  
Cn
1
 1 
cosh  cosh  1 
(21)
n 
Odd-order filters present a single transmission zero at infinity, and even-order filters end with
a constant gain –Amin at w = ¥. From Eq. (1) and Eq. (19), the attenuation in decibels for a

normalized inverse Chebyshev approximation is:


    
2

A 10 log1     (22)
 C 1 /  
  n 

The gains for some normalized inverse Chebyshev approximations are plotted in Fig. (5). A

frequency scaling by the factor given by Eq. (21) was applied, to make the passband end at w

= 1.

The selectivity of the inverse Chebyshev approximation is the same as the corresponding
Chebyshev approximation, for the same Amax and Amin. This can be verified by calculating

the ratio wp/wr for both approximations. For the normalized Chebyshev approximation, wp =

1, and wr occurs when eCn(wr) = g. For the normalized inverse Chebyshev approximation, wr

= 1, and wp occurs when (eg)/Cn(1/wp) = e. In both cases, the resulting ratio is wr/wp = Cn–
1(g). Equation (11) can be used to compute the required degree.

The transmission zero frequencies are the frequencies that make Eq. (19) infinite:

C n 1 /  k  cosn cos  1 1 /  k  0 
1
k  , k 0, 1,..., n  1 (23)
  1  2k 
cos 
2 n 

The pole frequencies are found by solving Eq. (2) with F(s) and P(s) given by Eq. (20):

2n 2n 2
2 s  s  j
E ( s ) E (  s )        Cn   (24)
 j  j s

The roots of this equation are the solutions of:

 j
C n   j  (25)
s
By observing the similarity of this equation to Eq. (14), the roots of E(s)E(–s) can be

obtained as the complex inverses of the values given be Eq. (18), with e replaced by 1/(eg).

They lie in a curve that is not an ellipse. E(s) is constructed from the roots with negative real

parts, that are distributed in a pattern that resembles a circle shifted to the left side of the

origin.

The similarity of the passband response with the Butterworth response makes the phase

characteristics of the inverse Chebyshev filters much closer to linear than those of the

Chebyshev filters. The Qs of the poles are also significantly lower for the same gain

specifications.

Realization of inverse Chebyshev filters

The realization based on LC doubly terminated ladder structures is also convenient for inverse

Chebyshev filters, by the same reasons mentioned for the direct approximation. In this case,

the passband sensitivities are low due to the nth-order attenuation zero at s = 0, that results in

the nullification of the first n derivatives of the filter gain in relation to all the reactive

elements at s = 0, and keeps the gain errors small in all the passband. Stopband errors are also

small, because the transmission zero frequencies depend only on simple LC series or parallel

resonant circuits. The usual structures used are shown in fig. 4(b).

Those realizations are possible only for the odd-order cases, because those structures can't

realize the constant gain at infinity that occurs in the even-order approximations (realizations

with transformers or with negative elements are possible). Even-order modified

approximations can be obtained by using instead of the Chebyshev polynomials, polynomials

obtained by the application, to the Chebyshev polynomials, of the Moebius transformation

(4,6):
 2 x 2  x z21 k 
x  2
; x z1 cos max (26)
1  x z1 2n

where kmax the greatest odd integer that is less than the filter order n. This transformation

moves the pair of roots closer to the origin of an even-order Chebyshev polynomial to the

origin. If the resulting polynomials are used to generate polynomial approximations, starting

from Eq. (7), the results are filters with two attenuation zeros at the origin, that are realizable

as a doubly terminated ladder filter with equal terminations, a convenience in passive

realizations. If the same polynomials are used in inverse polynomial approximations, starting

from Eq. (19), the results are filters with two transmission zeros at the infinity, that now are

realizable by doubly terminated LC structures. The direct and inverse approximations

obtained in this way have the same selectivity, slightly smaller than in the original case.

Table 3 lists polynomials, poles, zeros, frequency and Q of the poles, and LC doubly

terminated realizations for some inverse Chebyshev filters. The filters were scaled in

frequency to make the passband end at 1 rad/s. The even-order realizations were obtained

from modified approximations, and are listed separately in table 4. The structures are a mix

of the two forms in Fig. 4(b). Note that some realizations are missing. These are cases where

the zero-shifting technique for the realization of LC doubly terminated ladder filters fails. For

inverse Chebyshev filters, and other inverse polynomial filters, there is a minimum value of
Amin for each order that turns the realization in this form possible (7).

Other similar approximations

Different approximations with uniform passband or stopband ripple, somewhat less selective,

can be generated by reducing the number or the amplitude of the oscillations in a Chebyshev-

like polynomial, and generating the approximations starting from eqs. (7) or (19),

numerically (7).
A case particularly interesting results if the last oscillations of the polynomial value end in 0

instead of ±1. This creates double roots close to x = ±1 in the polynomial. In a polynomial

approximation, the higher frequency passband minimum disappears, replaced by a second-

order maximum close to the passband border. In an LC doubly-terminated realization, the

maximum power transfer at this frequency causes the nullification of the first two derivatives

of the gain in relation to the reactive elements, substantially reducing the gain error at the

passband border. In an inverse polynomial approximation, this causes the joining of the first

two transmission zeros, as a double transmission zero, what increases the attenuation and

reduces the error at the stopband beginning, allowing also a symmetrical realization for orders

5 and 7.

Other variations arise from the shifting of roots to the origin. This is also best done

numerically. Odd- (even-) order polynomial approximations with any odd (even) number of

attenuation zeros at w = 0, up to the approximation’s order (in the last case resulting a

Butterworth approximation), can be generated. The same polynomials generate inverse

polynomial approximations with any odd (even) number of transmission zeros at infinity.

In all cases, the Q of the poles is reduced and the phase is closer to linear. Similar techniques

can also be applied to elliptic approximations. For example, a low-pass elliptic approximation

can be transformed into a Chebyshev approximation by the shifting of all the transmission

zeros to infinity, or into an inverse Chebyshev approximation by the shifting of all the

attenuation zeros to the origin. There are many variations between these extremes.
Bibliography

1. A. Antoniou, Digital Filters: Analysis, Design, and Applications, New York: McGraw-

Hill,1993.

2. P. L. Chebyshev, Théorie des mécanismes connus sous le nom de parallelogrammes,

Oeuvres, Vol. I, St. Petersburg, 1899.

3. M. E. Van Valkenburg, Analog Filter Design, Holt, Rinehart and Winston, New

York,1982.

4. R. W. Daniels, Approximation Methods for Electronic Filter Design. McGraw-Hill, New

York, 1974.

5. H. J. Orchard, Inductorless Filters, Electronics Letters, vol. 2, pp. 224-225, September

1966.

6. G. C. Temes and J. W LaPatra, Circuit Synthesis and Design, McGraw-Hill Kogakusha,

Tokyo, 1977.

7. L. Weinberg, Network Analysis and Synthesis, McGraw-Hill, New York, 1962.

8. A. C. M. de Queiroz and L. P. Calôba, An approximation algorithm for irregular-ripple

filters, IEEE International Telecommunications Symposium, Rio de Janeiro, Brazil, pp. 430-

433, September 1990.

Table 1. Chebyshev Polynomials

n polynomial

0 1

1 x

2 2x – 1

3 4x3 – 3x

4 8x4 – 8x2 + 1

5 16x5 – 20x3 + 5x

6 32x6 – 48x4 +18x – 1

7 64x7 –112x5 + 56x3 –7x

8 128x8 – 256x6 + 160x4 – 32x2 + 1

9 256x9 – 576x7 + 432x5 – 120x3 + 9x

10 512x10 – 1280x8 + 1120x6 –400x4 + 50x2 – 1

11 1024x11 – 2816x9 + 2816x7 – 1232x5 + 220x3 – 11x

12 2048x12 – 6144x10 + 6912x8 – 3584x6 + 840x4 – 72x2 + 1

Table 2. Normalized Chebyshev filters with Amax = 1 dB and p = 1 rad/s.

Polynomials E(s)
n a0 a1 a2 a3 a4 a5 a6 a7 a8 a9 a10
1 1.96523 1.00000
2 1.10251 1.09773 1.00000
3 0.49131 1.23841 0.98834 1.00000
4 0.27563 0.74262 1.45392 0.95281 1.00000
5 0.12283 0.58053 0.97440 1.68882 0.93682 1.00000
6 0.06891 0.30708 0.93935 1.20214 1.93082 0.92825 1.00000
7 0.03071 0.21367 0.54862 1.35754 1.42879 2.17608 0.92312 1.00000
8 0.01723 0.10734 0.44783 0.84682 1.83690 1.65516 2.42303 0.91981 1.00000
9 0.00768 0.07060 0.24419 0.78631 1.20161 2.37812 1.88148 2.67095 0.91755 1.00000
10 0.00431 0.03450 0.18245 0.45539 1.24449 1.61299 2.98151 2.10785 2.91947 0.91593 1.00000

Poles
n re/im 1 w/Q 1 re/im 2 w/Q 2 re/im 3 w/Q 3 re/im 4 w/Q 4 re/im 5 w/Q 5
1 -1.96523

2 -0.54887 1.05000
0.89513 0.95652
3 -0.24709 0.99710 -0.49417
0.96600 2.01772
4 -0.13954 0.99323 -0.33687 0.52858
0.98338 3.55904 0.40733 0.78455
5 -0.08946 0.99414 -0.23421 0.65521 -0.28949
0.99011 5.55644 0.61192 1.39879
6 -0.06218 0.99536 -0.16988 0.74681 -0.23206 0.35314
0.99341 8.00369 0.72723 2.19802 0.26618 0.76087
7 -0.04571 0.99633 -0.12807 0.80837 -0.18507 0.48005 -0.20541
0.99528 10.89866 0.79816 3.15586 0.44294 1.29693
8 -0.03501 0.99707 -0.09970 0.85061 -0.14920 0.58383 -0.17600 0.26507
0.99645 14.24045 0.84475 4.26608 0.56444 1.95649 0.19821 0.75304
9 -0.02767 0.99761 -0.07967 0.88056 -0.12205 0.66224 -0.14972 0.37731 -0.15933
0.99723 18.02865 0.87695 5.52663 0.65090 2.71289 0.34633 1.26004
10 -0.02241 0.99803 -0.06505 0.90245 -0.10132 0.72148 -0.12767 0.47606 -0.14152 0.21214
0.99778 22.26303 0.90011 6.93669 0.71433 3.56051 0.45863 1.86449 0.15803 0.74950

Polynomials P(s)
n ctt a0
1 1.96523 1.00000
2 0.98261 1.00000
3 0.49131 1.00000
4 0.24565 1.00000
5 0.12283 1.00000
6 0.06141 1.00000
7 0.03071 1.00000
8 0.01535 1.00000
9 0.00768 1.00000
10 0.00384 1.00000

Doubly terminated LC ladder realizations

n Rg/Rl L/C 1 L/C 2 L/C 3 L/C 4 L/C 5 L/C 6 L/C 7 L/C 8 L/C 9 L/C 10
1 1.00000
1.00000 1.01769
2 1.63087 1.11716
0.61317 1.11716
3 1.00000 0.99410
1.00000 2.02359 2.02359
4 1.63087 1.73596 1.28708
0.61317 1.28708 1.73596
5 1.00000 1.09111 1.09111
1.00000 2.13488 3.00092 2.13488
6 1.63087 1.80069 1.87840 1.32113
0.61317 1.32113 1.87840 1.80069
7 1.00000 1.11151 1.17352 1.11151
1.00000 2.16656 3.09364 3.09364 2.16656
8 1.63087 1.82022 1.93073 1.90742 1.33325
0.61317 1.33325 1.90742 1.93073 1.82022
9 1.00000 1.11918 1.18967 1.18967 1.11918
1.00000 2.17972 3.12143 3.17463 3.12143 2.17972
10 1.63087 1.82874 1.94609 1.95541 1.91837 1.33890
0.61317 1.33890 1.91837 1.95541 1.94609 1.82874

Table 3. Normalized inverse Chebyshev filters with Amax = 1 dB and Amin = 50 dB, and p =

1 rad/s.

Polynomials E(s)
n a0 a1 a2 a3 a4 a5 a6 a7 a8 a9 a10
 1 1.96523 1.00000
2 1.96838 1.98099 1.00000
3 2.01667 3.14909 2.51015 1.00000
4 2.19786 4.52937 4.90289 3.13118 1.00000
5 2.60322 6.42983 8.61345 7.26320 3.81151 1.00000
6 3.35081 9.35051 14.61162 14.91369 10.30744 4.54023 1.00000
7 4.64002 14.09440 24.72451 29.03373 24.18372 14.09633 5.30979 1.00000
8 6.82650 22.03426 42.29782 55.31092 52.89124 37.20009 18.68307 6.11268 1.00000
9 10.54882 35.60372 73.49954 104.68294 111.48145 90.07839 54.81844 24.10445 6.94337 1.00000
10 16.95789 59.19226 129.80937 198.24216 230.34722 207.44800 145.48766 77.89699 0.39330 7.79647 1.00000

Poles
n re/im 1 w/Q 1 re/im 2 w/Q 2 re/im 3 w/Q 3 re/im 4 w/Q 4 re/im 5 w/Q 5
1 -1.96523
0.00000
2 -0.99049 1.40299
0.99363 0.70823
3 -0.61468 1.25481 -1.28079
1.09395 1.02071
4 -0.42297 1.18385 -1.14262 1.25229
1.10571 1.39945 0.51249 0.54799
5 -0.30648 1.13993 -0.94418 1.23656 -1.31018
1.09795 1.85969 0.79849 0.65483
6 -0.23016 1.10962 -0.75398 1.21506 -1.28598 1.35770
1.08549 2.41056 0.95283 0.80576 0.43545 0.52789
7 -0.17794 1.08768 -0.59638 1.18959 -1.14085 1.36871 -1.47946
1.07303 3.05632 1.02930 0.99735 0.75619 0.59986
8 -0.47425 1.16431 -0.14101 1.07137 -0.95398 1.34983 -1.48710 1.55173
1.06334 1.22752 1.06205 3.79891 0.95496 0.70747 0.44316 0.52173
9 -0.38185 1.14152 -0.77805 1.31643 -0.11413 1.05899 -1.34453 1.56247 -1.70623
1.07575 1.49471 1.06189 0.84597 1.05282 4.63922 0.79596 0.58105
10 -0.63221 1.27960 -0.31203 1.12193 -0.09407 1.04944 -1.13939 1.53032 -1.72054 1.78611
1.11252 1.01201 1.07766 1.79777 1.04521 5.57772 1.02161 0.67155 0.47954 0.51906

Polynomials P(s)
n multiplier a0 a2 a4 a6 a8 a10
1 1.96523 1.00000
2 0.00316 622.45615 1.00000
3 0.05144 39.20309 1.00000
4 0.00316 695.02278 74.56663 1.00000
5 0.03477 74.86195 19.34709 1.00000
6 0.00316 1059.61979 494.96516 57.80151 1.00000
7 0.03463 133.99401 95.81988 19.57753 1.00000
8 0.00316 2158.72730 2130.49651 657.07341 64.84805 1.00000
9 0.03786 278.65997 354.39519 150.23800 23.58892 1.00000
10 0.00316 5362.55604 8380.91576 4584.36462 1023.53040 79.98165 1.00000

Zeros
n w1 w2 w3 w4 w5
1
2 24.94907
3 6.26124
4 7.97788 3.30455
5 3.74162 2.31245
6 6.92368 2.53424 1.85520
7 3.60546 2.00088 1.60458
8 7.29689 2.56233 1.71209 1.45144
9 3.88896 2.06927 1.53587 1.35062
10 8.08496 2.78589 1.78865 1.41948 1.28053

LC doubly terminated realizations

n Rg/Rl L/C 1 L/C 2 L/C 3 L/C 4 L/C 5 L/C 6 L/C 7
1 1.00000
1.00000 1.01769
2 1.00000
1.00000
3 1.00000 1.56153
1.00000 0.78077 0.01634 0.78077
4 1.00000
1.00000
5 1.00000 1.16364 1.30631
1.00000 0.37813 0.16071 1.62010 0.05468 0.47172
6 1.00000
1.00000
7 1.00000 0.72897 1.34370 0.96491
1.00000 0.09574 0.34265 1.32044 0.28905 1.32059 0.07972 0.30081
Table 4. Normalized even-order modified inverse Chebyshev filters with two transmission
zeros at infinity, with Amax = 1 dB, Amin = 50 dB, and p = 1 rad/s.

Polynomials E(s)
n a0 a1 a2 a3 a4 a5 a6 a7 a8 a9 a10
2 1.96523 1.98254 1.00000
4 2.12934 4.47598 4.86847 3.12041 1.00000
6 3.14547 9.02141 14.23655 14.65051 10.18872 4.51414 1.00000
8 6.32795 20.98707 40.68275 53.69811 51.69862 36.58972 18.48009 6.07949 1.00000
10 15.69992 56.17036 124.1801 191.3464 223.6989 202.6490 142.8395 76.84994 30.12162 7.76165 1.00000

Poles
n re/im 1 w/Q 1 re/im 2 w/Q 2 re/im 3 w/Q 3 re/im 4 w/Q 4 re/im 5 w/Q 5
2 -0.99127 1.40187
0.99127 0.70711
4 -0.43134 1.18419 -1.12886 1.23225
1.10284 1.37268 0.49409 0.54580
6 -0.23626 1.11107 -0.76275 1.20632 -1.25806 1.32324
1.08566 2.35141 0.93457 0.79077 0.41016 0.52590
8 -0.14421 1.07247 -0.48399 1.16315 -0.96075 1.33672 -1.45079 1.50858
1.06273 3.71848 1.05767 1.20162 0.92940 0.69566 0.41357 0.51992
10 -0.64341 1.27784 -0.31781 1.12245 -0.09573 1.05011 -1.14584 1.51514 -1.67803 1.73625
1.10404 0.99303 1.07652 1.76590 1.04574 5.48452 0.99132 0.66115 0.44586 0.51735

Polynomials P(s)
n ctt a0 a2 a4 a6 a8
2 1.96523 1.00000
4 0.16412 12.97454 1.00000
6 0.11931 26.36278 10.89186 1.00000
8 0.13145 48.13911 44.73326 12.54437 1.00000
10 0.16119 97.39855 147.0191 76.50032 15.68797 1.00000

Zeros
n w1 w2 w3 w4
2
4 3.60202
6 2.69467 1.90542
8 2.71078 1.74464 1.46706
10 2.94484 1.82001 1.43081 1.28694

LC doubly terminated ladder realizations

n Rg/Rl L/C 1 L/C 2 L/C 3 L/C 4 L/C 5 L/C 6 L/C 7 L/C 8
2 1.00000 1.00881
1.00000 1.00881
4 1.00000 1.51207 0.05275 0.58997
1.00000 0.64094 1.46110
6 1.00000 0.87386 1.67233 0.13065 0.32187
1.00000 0.17880 0.31519 1.63514 1.05413
8 1.00000 0.67581 0.32023 1.34317 0.38303 1.12998 0.18178 0.16760
1.00000 0.32898 1.02594 1.21303 0.74862
 C1 C2 C3

C4 C5 C6

Fig. 1. Plots of the first six Chebyshev polynomials Cn(x). The squares limit the region –1 £

x£1, –1 £ Cn(x) £ 1, where the polynomial value oscillates.

 |T ( j  )| dB
0 1
0
1 
3
5

6
4
–1 2
(a)
|T ( j )| dB
1 10
0 
–1
1

10
–100 (b)

Fig. 2. Passband gain (a) and stopband gain (b) for the first normalized Chebyshev

approximations with 1 dB passband ripple. Observe the uniform passband ripple and the

monotonic stopband gain decrease.

 j
–1 1
cosh ( 1n sinh

)
–1 1
sinh ( 1n sinh

)

 2n
n

n


n


Fig. 3. Localization of the poles in a normalized Chebyshev low-pass approximation (seventh

order, in the case). The pole locations can be obtained as shown.

 Rg L2 Ln

Rl Rl
C1 C3 Cn
(a)
L2
Rg

Rl
C1 C2 C3 Cn

Rg L1 L3 Ln

L2
Rl

C2
(b)

Fig. 4. LC doubly terminated ladder realizations for Chebyshev filters, in the direct form (a),

and in the inverse form (b). These classical realizations continue to be the best prototypes for

active realizations, due to their low sensitivity to errors in the element values.
 |T ( j  )| dB
0 1
0

10
1 2

–1
(a)
|T ( j )| dB
1 100
0 
–1

–100 (b)

Fig. 5. Passband gain (a) and stopband gain (b) for the first normalized inverse Chebyshev
approximations with Amax = 1 dB and Amin = 50 dB. Observe the maximally flat passband

and the uniform stopband ripple.