Filters

0 Overview
A filter is a two-port device used to control the frequency response at a certain point in a system by
providing transmission at frequencies within the passband of the filter and attenuation in its stopband.
It can be classified by magnitude response as low-pass filter (LPF), high-pass filter (HPF), band-pass
filter (BPF), and band-stop filter (BSF). It has wide range of applications including:
 Desired frequency band selection and unwanted band rejection (i.e., SNR improvement and
Interference reduction)
 Noise reduction
 Channel selection in mobile and satellite communications
1 Image Impedance
In a two-port network, if two impedances Z1i and Z2i are such that Z1i is the driving point impedance at
port 1 with impedance Z2i is connected across port 2 and Z2i is the driving point impedance at port 2
with impedance Z1i is connected across port 1, then the impedances Z1i and Z2i are called the image
impedances of the network. For symmetrical network, image impedances are equal to each other, i.e.,
Z1i = Z2i, and is called the characteristic or iterative impedance Z0.
Consider networks shown in Fig. 1, the driving
point impedance at port 1 is given by
V1 AZ 2 i  B
Z1i   .
I1 CZ 2 i  D
Likewise, the driving point impedance at port 2 is
given by
V2 DZ1i  B (a)
Z 2i   .
I 2 CZ1i  A
Here, ABCD denote the transmission parameters.
Solving the two equations above yields
AB BD
Z1i  ; Z 2i  .
CD AC
Since the open-circuit input impedance Zioc and
the short–circuit input impedance Zisc are given by (b)
A B Fig. 1: Image impedance
Z ioc  ; Z isc  ,
C D
the image impedance at port 1 can be rewritten as
Z1i  Z ioc Z isc .
Likewise, since the open-circuit output impedance Zooc and the short–circuit input impedance Zosc are
given by
D B
Z ooc  ; Z osc  ,
C A
the image impedance at port 2 can be rewritten as
Z 2i  Z ooc Z osc .
2 Symmetric T and  networks
Consider a T network interposed between a generator with internal impedance Z1i and a load
impedance of Z2i, as shown in Fig. 2. It is desired that the maximum power transfer occurs, i.e., the
impedance at 1,1’ terminals into which the generator supplies power be equal to Z1i, and the
impedance at 2,2’ terminals be equal to Z2i. Hence,

1
 Z 3 ( Z 2  Z 2i )
Z1in  Z1   Z1i and
Z 2  Z 3  Z 2i
Z ( Z  Z1i )
Z2  3 1  Z 2i .
Z1  Z 3  Z1i
Solving both equations yields
( Z1  Z 3 )(Z1Z 2  Z 2 Z3  Z1Z 3 )
Z1i  ; Fig. 2: A T-network interposed between load
Z 2  Z3 and source
.
( Z 2  Z3 )( Z1Z 2  Z 2 Z 3  Z1Z3 )
Z 2i 
Z1  Z3
Note also that
Z1i  Z1oc Z1sc ; Z 2i  Z 2oc Z 2 sc , as before.
When Z1=Z2, i.e., two series arms of a T-network are equal, Z1/2 Z1/2
the network is said to be symmetric. For symmetric
networks, Z1i = Z2i = Z0 (characteristic impedance). Filter Z2
networks are usually set up as symmetrical sections of T or 
types (Fig. 3(a), Fig. 4(a)). T section can be considered as
built up of unsymmetrical L-half sections (Fig. 3(b)). For the
(a) T-section
T network shown in Fig. 3(a) terminated by its characteristic
impedance Z0, the input impedance is given by
Z1 Z 2 ( Z1 / 2  Z 0 )
Z1in   .
2 Z1 / 2  Z 2  Z 0
With proper choice of Z0, it is possible to make Z1in=Z0,
Z1 Z 2 ( Z1 / 2  Z 0 )
Z1in  Z 0    Z12 / 4  Z1Z 2 .
2 Z1 / 2  Z 2  Z 0
(b) two L sections
Hence, for symmetrical T-section, Z0 is given by Fig. 3: A symmetrical T-section
Z 0T  Z12 / 4  Z1Z 2  Z1Z 2 (1  Z1 / 4 Z 2 ) .
Again, from open and short-circuit measurements for the symmetrical T section,
Z Z /2
Z1oc  Z 2 oc  Z oc  Z1 / 2  Z 2 ; Z1sc  Z 2 sc  Z sc  Z1 / 2  1 2 and Z1oc Z1sc  Z12 / 4  Z1Z 2  Z0T
2

Z1 / 2  Z 2
Thus, Z 0T  Z oc Z sc .
Likewise,  section can be considered as built up of
unsymmetrical L-half sections (Fig. 4(b)). For the 
network shown in Fig. 4(a) terminated by its
characteristic impedance Z0, the input impedance is
given by
 2Z 2 Z 0  (a)  -section
 Z1  2 Z 2
 2 Z 2  Z 0 
Z1in  .
2Z 2 Z 0
Z1   2Z 2
2Z 2  Z 0
Requiring Z1in=Z0 leads to
Z1 Z 2 ZZ
Z 0   1 2.
1  Z1 / 4 Z 2 Z 0T (b) two L sections
It can also be shown that Fig. 4: A symmetrical  -section
Z 0  Z oc Z sc .

2
A series connection of several T or  networks leads to so-called ladder networks, as shown in Fig. 5
(a)-(d). Terminal half-section matching is obtained by connecting the ends of the T-network with the
half sections of the -network (Fig. 4 (b)), i.e., connect terminals 2,2’ of Fig. 4(b) with terminals a,a’
of Fig. 5(a) and 3,3’ with b,b’. Similarly, for the -network of Fig. 5(c), terminal matching is to be
done by the half-sections of the T-network (Fig. 3(b)), i.e., connecting terminals 2,2’ to c,c’ and 1,1’
to d,d’.

(a) Ladder networks of T

sections

(b) Equivalent network

of (a)

(c) Ladder networks of 

sections

(d) Equivalent network

of (c)

Fig. 5: Ladder networks made of T-sections and -sections.

3 Propagation Constant
Under Z0 termination, input and output impedances are equal, i.e.,
Z 0  V1 / I1  V2 /( I 2 ) ,
then V1 / V2  I1 /(  I 2 )  e  ,
where  is a complex number and is defined as
    j ,
where  , ,  are propagation constant, attenuation constant, and phase constant, respectively.
Furthermore,
V1 / V2  I1 /(I 2 )  A  I1 / I 2 e j  e  j .
For n number of sections cascaded, with all of them having the same Z0 value, the ratio of currents can
be written as
I1 I I I
 2  L  n1  2  e 1  e 2  L  e n  e .
 I2  I3  In  In
The overall propagation constant  can be expressed as
  1   2  L   n .
4 Properties of Symmetrical Network
For a symmetrical T-section terminated with a load Z0 and fed with a generator E0, as shown in Fig. 6,

3
E 0  ( Z1 / 2  Z 2 ) I1  Z 2 I 2
0  Z 2 I1  ( Z1 / 2  Z 2  Z 0 ) I 2 .
I1 Z / 2  Z2  Z0
Thus,  1  e , or
 I2 Z2
Fig. 6 Symmetrical network terminated by Z0
Z 0  Z 2 (e  1)  Z1 / 2 .

Applying the previous result Z 0  Z12 / 4  Z1 Z 2  Z1 Z 2 (1  Z1 / 4 Z 2 ) yields

 2
Z 02  Z 2 (e  1)  Z1 / 2  Z 22 (e  1) 2  Z1Z 2 (e  1)  Z12 / 4  Z12 / 4  Z1Z 2 .
After simplification,
Z 2 (e  1) 2  Z1e   0 or e 2  2e   1  ( Z1 / Z 2 )e  . Hence,

e  e  
 cosh   1  Z1 / 2 Z 2 .
2
Since, cosh 2   sinh 2   1 ,

sinh2   cosh2   1  (1  Z1 / 2Z 2 )2  1  Z 12 / 4Z 22  Z1 / Z 2  Z 02 / Z 22  sinh  Z0 / Z2 ,

Z0
and tanh   .
Z1 / 2  Z 2
Using the half-angle identity,

sinh( ) 
1
cosh( )  1)  1 1  Z1  1  Z1 .
2 2 2  2Z 2  4Z 2
2
I  Z / 2  Z2  Z0 Z  Z  Z
Again, 1  e  1  1  1   1   1 ,
 I2 Z2 2Z 2  2Z 2  Z 2
  Z1 
2 
 Z1 Z1 
So,   ln 1      .
 2Z 2  2Z 2  Z 2 
 

5 Filter Fundamentals
The purpose of a filter network is to pass a desired frequency band without loss and stop or
completely attenuate all undesired frequency bands. Since     j ,  = 0 means there is no
attenuation in transmission with only a phase shift, i.e., |I1| = |I2| and the operation is in the pass band.
If  > 0, then |I1| > |I2|, i.e., the attenuation occurs and the operation is in the stop band.
Recall that

 Z1    j     
sinh( )   sinh   sinh cos  j cosh sin .
2 4Z 2  2  2 2 2 2
Case I When Z1 and Z2 are of the same type of reactances, then Z1/4 Z2 > 0 and sinh(/2) is real, i.e.,
(i) cosh( / 2) sin(  / 2)  0 or sin(  / 2)  0;   n , n  0,2,4, K
(ii) sinh( / 2) cos( / 2)  Z1 / 4Z2 .

4
Therefore, cos(/2) = 1 as sin(/2) =0. Hence,

sinh( / 2)  Z1 / 4Z2 ;   2 sinh1 Z1 / 4Z 2 .

Case II If Z1 and Z2 are of the opposite type of reactances, then Z1/4 Z2 is negative, i.e., Z1/4 Z2 < 0 and
obviously, Z1 / 4Z 2 is imaginary. Therefore, the following conditions must be satisfied:

(i) j cosh( / 2) sin( / 2)  Z1 / 4Z2

(ii) sinh( / 2) cos(  / 2)  0
Two conditions may arise
(a) sinh( / 2)  0, i.e.,   0 when   0 and

j sin( / 2)  Z1 / 4Z 2 Q cosh( / 2)  1.
This signifies the region of zero attenuation or pass band which is limited by the upper limit of the
sine term, i.e., sin(/2) = |1|, or it is required that
 1  Z1 / 4Z 2  0 .
The phase angle in the pass band is given by
  2 sin1  Z1 / 4Z 2 .
(b) cos(/2) = 0; therefore sin(/2) = ±1;  = (2n-1) when  ≠ 0 and
j cosh( / 2)  Z1 / 4Z 2  cosh( / 2)   Z1 / 4Z 2 ;   2 cosh1  Z1 / 4Z 2 .
Since hyperbolic cosine has no value below 1, the condition for stop band is Z1/4 Z2 < -1. The
frequencies at which the network changes from pass band to stop band and vice versa are called the
cut-off frequencies. These frequencies occur when
Z1 / 4Z 2  0, or Z1  0 and Z1 / 4Z 2  1, or Z1  4Z 2 ,
where Z1 and Z2 are of the opposite type of reactances.
For symmetrical T- and -network made up entirely of pure reactances, Z0 is given by
Z0T   X1 X 2 (1  X1 / 4 X 2 ) ; Z 0   X 1 X 2 / Z 0T .
Table 1 summarizes the two bands, namely the pass band and the stop band with respect to the
different values of X1/4X2.
Table 1
X1/4X2 0 to -1 -1 to -
Band Pass Stop
1
 0 2 cosh X1 / 4X 2
 2 sin 1 X 1 / 4 X 2 
Z0T positively real purely reactive
In a pass band, Z0 is real and positive. If the network is terminated with a resistive Z0 = R0, then the
input impedance is R0 and the network will accept and transmit power to the resistive load without
loss. If the network is fed by a generator having an internal impedance R0, then the system will be
matched and the maximum power transfer occurs. In a stop band, Z0 is reactive. If the network is
terminated in its reactive Z0, it may transmit voltage or current with 90 phase difference between
input and output with considerable attenuation.

6 The constant-k Filters

In constant-k filters, Z1 and Z2 are of opposite reactances. Then
Z1 Z 2  k 2 ,
where k is a constant.
6.1 Low-Pass Filters

5
For low-pass filters, Z1=jL, Z2=1/jC, then
Z1 Z 2  L / C  Rk2  k 2 .
The cut-off frequency can be found from
Z1 j L  2 LC 2
  jC   1  c  .
4Z 2 4 4 LC
Fig. 7 shows the low-pass T-section filter.
Fig. 7 Low-pass T-section
The characteristic impedance of the T-section and -section are given by
Z 0T  ( L / C )(1   2 LC / 4)  Rk 1  ( / c ) 2  Rk 1  ( f / f c ) 2 and
Z 0  Rk / 1  ( f / f c ) 2 .
Design Procedure To determine the values of L, C, the value of Rk, i.e., the characteristic impedance
at zero frequency, and the cut-off frequency are required. Then, from
L / C  Rk2 and LC  1 / f c ,
L, C can be calculated.
Low-pass Filter Example Design a low-pass filter with cut-off frequency of 1 MHz, and the
characteristic impedance of 100 .
low-pass filter : m = 1 : Characteristic Impedance
1000

900

800

700

600
|Z0T|[  ]

500

400

300

200

100

0
0 2 4 6 8
10 10 10 10 10
f [Hz]

low-pass filter : m = 1 : frequency response

low-pass filter : m = 1 : attenuation & phase constants 0

-5

5 -10

-15

2 -20
|e-|[dB]

-25


-30

-35

-40

-45

-50
0
0 2 4 6 0 2 4 6
10 10 10 10 10 10 10 10
f [Hz] f [Hz]

6.2 High-Pass Filters

For high-pass filters, Z1=1/jC, Z2=jL, then
Z1 Z 2  L / C  Rk2  k 2 .
The cut-off frequency can be found from
Z1 1 1 1 1
    1  c  .
4Z 2 jL 4 jC 4 LC
2
2 LC Fig. 8 High-pass T-section
Fig. 8 shows the high-pass T-section filter.
The characteristic impedance of the T-section and -section are given by

6
Z 0T  ( L / C )(1  1 / 4 2 LC )  Rk 1  (c /  ) 2  Rk 1  ( f c / f ) 2 and
Z 0  Rk / 1  ( f c / f ) 2 .
Design Procedure To determine the values of L, C, the value of Rk, i.e., the characteristic impedance
at infinite frequency, and the cut-off frequency are required. Then, from
L / C  Rk2 and LC  1 / 4f c ,
L, C can be calculated.
High-pass Filter Example Design a high-pass filter with cut-off frequency of 1 MHz, and the
characteristic impedance of 100 .
high-pass filter : m = 1 : Characteristic Impedance
1000

900

800

700

600

|Z0T|[  ]
500

400

300

200

100

0
0 2 4 6 8
10 10 10 10 10
f [Hz]

high-pass filter : m = 1 : frequency response

high-pass filter : m = 1 : attenuation & phase constants
0
4

-50

20
-100
|e- |[dB]

2


-150

-200

0 0 -250
0 2 4 6 8 0 2 4 6 8
10 10 10 10 10 10 10 10 10 10
f [Hz] f [Hz]

6.3 Band-Pass Filters

For band-pass filters, Z1 is a series LC circuit, i.e., Z1= j (L1 -
1/C1), and Z2 is a parallel LC circuit, i.e., Z2=jL2 // 1/jC2 ,
as shown in Fig. 9. The condition for the band-pass filter is
that both series and parallel LC circuits have equal resonant
frequencies, i.e.,
 02 L1C1  1   02 L2 C 2 or L1C1  L2C2 . Then,
L2 (1   2 L1C1 ) L2 Fig. 9 Band-pass T-section
Z1Z 2    Rk2  k 2 .
C1 (1   2 L2C 2 ) C1
The cut-off frequency can be found from
Z 1   4 Z 2  Z 12   4 Z 1 Z 2  4 Rk2  Z 1   j 2 Rk .
Hence, Z1 at lower cut-off frequency fL is equal to –Z1 at upper cut-off frequency fH, i.e.,
1 / LC1  L L1  H L1  1 / H C1 , or 1   L2 L1C1  ( L /  H )( H2 L1C1  1) .
Using  02  1 / L1C1 yields
1   L2 /  02  ( L /  H )( H2 /  02  1) or  0   L H .

7
Also, from
Z1   j 2 Rk  Z1    Z1    4 jRk , or  H L1  1 /  H C1  ( L L1  1 /  L C1 )  4 Rk ,
H L

one can derive the condition

 H   L  2 Rk C1 02 .
Design Procedure To determine the values of L1, C1, L2, and C2, one needs to specify the center
frequency, the bandwidth and the desired characteristic impedance, then using the following
procedures:
1. Determine C1 from  H   L  2 Rk C1 02 .
2. Determine L1 from L1  1 /  02 C1 .
3. Determine L2 from L2 = k2C1, since L2/C1= k2.
4. Determine C2 from C 2  1 /  02 L2 .
Band-pass Filter Example Design a band-pass filter with center frequency of 100 MHz, the bandwidth
of 20 MHz, and the characteristic impedance of 100 .
band-pass filter : m = 1 : Characteristic Impedance
700

600

500

400
|Z0T|[ ]

300

200

100

0
8
10
f [Hz]

band-pass filter : m = 1 : attenuation & phase constants band-pass filter : m = 1 : frequency response
0
5
-5

-10

-15
2
-20
|e- |[dB]



-25

-30

-35

-40

-45
0 0
8 8
10 10
f [Hz] f [Hz]

6.4 Band-Stop Filters (or Band-elimination filters, Band-rejection filters)

For band-stop filters, Z1 is a parallel LC circuit, i.e., Z1= jL1 //
1/jC1, and Z2 is a series LC circuit, i.e., Z2= j (L2 - 1/C2), as
shown in Fig. 10. The condition for the band-stop filter is that
both series and parallel LC circuits have equal resonant
frequencies, i.e.,
 02 L1C1  1   02 L2 C 2 or L1C1  L2C2 . Then,
L1 (1   2 L1C1 ) L
Z1 Z 2   1  Rk2  k 2 .
C 2 (1   L2C 2 ) C 2
2 Fig. 10 Band-stop T-section
The cut-off frequency can be found from

8
Z 1  4 Z 2  4 Z 22  Z 1 Z 2  Rk2  Z 2   jR k / 2 .
Hence, Z2 at lower cut-off frequency fL is equal to –Z2 at upper cut-off frequency fH, i.e.,
1 / LC2  L L2  H L2  1 / H C2 , or 1   L2 L2C2  ( L /  H )( H2 L2 C2  1) .
Using 02  1 / L2C 2 yields
1   L2 /  02  ( L /  H )( H2 /  02  1) or  0   L H .
Also, from
Z 2   jRk / 2  Z 2    Z 2    jRk , or  H L2  1 /  H C 2  ( L L2  1 /  L C2 )  Rk ,
H L

one can derive the condition

 H   L  Rk C 2 02 / 2 .
Design Procedure To determine the values of L1, C1, L2, and C2, one needs to specify the center
frequency, the bandwidth and the desired characteristic impedance, then using the following
procedures:
1. Determine C2 from  H   L  Rk C 2 02 / 2 .
2. Determine L2 from L2  1 /  02 C 2 .
3. Determine L1 from L1 = k2C2, since L1/C2= k2.
4. Determine C1 from C1  1 /  02 L1 .
Band-stop Filter Example Design a band-stop filter with center frequency of 100 MHz, the bandwidth
of 20 MHz, and the characteristic impedance of 100 .
band-stop filter : m = 1 : Characteristic Impedance
1000

900

800

700

600
|Z0T|[  ]

500

400

300

200

100

0
8
10
f [Hz]

band-stop filter : m = 1 : attenuation & phase constants band-stop filter : m = 1 : frequency response
0
10
-10

-20

-30
2
-40
|e- |[dB]



-50

-60

-70

-80

-90
0 0
8 8
10 10
f [Hz] f [Hz]

7 The m-derived T-section

The constant-k prototype filter section, though simple, has two major disadvantages, namely (i) the
characteristic impedance varies widely over the pass band so that impedance matching is not possible,
(ii) the cut-off rate is not appreciably high, i.e., the drop-off rate is not sufficiently fast. The cut-off

9
rate may be raised by cascading a number of constant-k sections in series, but this is not economical.
The m-derived filters are designed to achieve this objective.
The approach used here is to introduce a zero frequency into the impedance of the shunt arm.
At this frequency, denoted by f, the shunt arm becomes a short circuit and the attenuation becomes
infinity. If f is chosen to be close to the cut-off frequency, then the cut-off rate can be raised. The
attenuation may be kept at high value throughout the stop band by cascading the constant-k prototype
section with the m-derived section. Now, consider the m-derived T-section, let us assume
Z1 '  mZ1 ,
where 0 < m < 1. Then, solving for Z2’ that achieves the same value of Z0T yields
Z2 1  m2
Z 0T  Z12 / 4  Z1Z 2  Z1 '2 / 4  Z1 ' Z 2 '  m 2 Z1 / 4  mZ1Z 2 ' or Z 2 ' 
2
 Z1 .
m 4m
For a low-pass filter section, Z1=jL, Z2=1/jC, then
Z1’=jmL, Z2=1/jmC+(1-m2) jmL/4m, as shown in Fig. 11.
The resonant frequency of the shunt arm becomes
1  m2 2 4 c2 1  m2
L
 LC  1  2   or 4m
4 (1  m 2 ) LC 1  m 2
  c / 1  m 2 , where c is the cut-off frequency.
Therefore, the smaller the value of m, the sharper the cut-off. Fig. 11 m-derived low-pass T-section
Notice that
Z2 1  m2 1 1  m2 1 4  (1  m 2 ) 2 LC 1 1  (1  m 2 )( 2 / c2 )
Z2 '   Z1   j L   ,
m 4m jmC 4m 4m j C m j C
the pass and stop bands can be characterized as follows:
(a) Pass band  1  Z1 / 4Z 2  0 and   0 .
m 2 2 LC m / c
  2 sin 1  Z1 ' / 4Z 2 '  2 sin 1  2 sin 1 .
1  (1  m )( / c )
2 2 2
1  (1  m 2 )( 2 / c2 )
(b) Stop band -<Z1’/4Z2’<-1 and =(2n-1).
For fc < f < f,
m / c m / c
  2 cosh 1  Z1 / 4Z 2  2 cosh 1  2 cosh 1 .
1  (1  m 2 )( 2 / c2 ) 1  ( 2 / 2 )
For f > f
m / c m / c
  2 cosh 1 Z1 / 4Z 2  2 cosh 1  2 cosh 1 .
(1  m )( /  )  1
2 2 2
c ( 2 / 2 )  1
Similar analysis procedure can be applied to the m-derived high-pass T-section, as shown in Fig. 12.
Here,   c 1  m 2 .
Likewise, the m-derived band-pass T-section is shown in Fig. 13.
Question Find  for the T-section in Fig. 13.
m-derived Low-pass Filter Example Design a low-pass filter
with cut-off frequency of 1 MHz, and the characteristic
impedance of 100 . Here, use m = 0.7.

10
 1 m2
L1
4m
4m
C1
1 m2
4m
C
1 m2

Fig. 12 m-derived high-pass T-section Fig. 13 m-derived band-pass T-section

8 Termination with m-derived half sections
The m-derived T- or -sections can be formed by the splitted m-derived half sections or L-sections, as
shown in Fig. 14. These m-derived half sections, having m = 0.6, are called terminating half sections.

1 m2 1  m2 1  m2
Z1 Z1 1 m2 1 m2
Z1 2m 2m Z1 Z1
4m 2m 2m

(a) (b) (c)

Fig. 14 (a) m-derived T-section (b) m-derived -section (c) m-derived half sections
Zobel discovered that an m-derived half section could be made to change its characteristics with
frequency in such a way that the filter is approximately matched to its load at all frequencies over
most of the pass band.
Now, the image impedance of the left half section at the 1,1’ terminals is given by

Z1i  Z1oc Z1sc 

(1  m ) Z / 2m  2 Z 2 / m (mZ1 / 2)
2
1 2


 1  (1  m 2 ) Z1 / 4 Z 2 Z 0 , 
(1  m ) Z1 / 2m  2 Z 2 / m  mZ1 / 2
2

where Z 0  Z1 Z 2 /(1  Z1 / 4 Z 2 ) . The impedance of the left half section at terminal 2,2’ is
Z 2i  Z 2 oc Z 2 sc  mZ 1 
/ 2  (1  m 2 ) Z1 / 2m  2 Z 2 / m mZ1 / 2  Z1Z 2 (1  Z1 / 4 Z 2 )  Z 0T .
The image impedance at 3,3’ terminals Image impedance of m-derived section
5
is equal to Z0T, and at terminals 4,4’ is
equal to Z1i. For low-pass filters, using 4.5

Z 0  Rk / 1  ( f / f c ) 2 yields 4

Z1i 

Rk 1  (1  m 2 )( f / f c ) 2 . 3.5

1  ( f / f c )2 3
Z1i/Rk

The variation of image impedance as a 2.5

function of f/fc is plotted in Fig. 15. It is 2
m=1
seen that m = 0.6 half section has a m=0.8
nearly constant value of Z1i can be 1.5

obtained over 85% of the pass band. 1

m=0.6

Following the same procedure, the m=0.4

image impedance for high-pass filters 0.5 m=0.2

can be given by
 
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Rk 1  (1  m 2 )( f c / f ) 2
Z1i  f/f
c

1  ( fc / f )2 Fig. 15: Variation of image impedance of m-derived section

11
Fig. 16 summarizes the T- and -sections used for low-pass and high-pass T-section filter designs.
8 Composite Filter Design
By combining in cascade the constant-k, m-derived sharp cut-off, and the m-derived matching
sections, one can realize a filter with the desired attenuation and matching properties. This type of
design is called a composite filter. Fig. 17 shows an example of composite filter design. The constant-
k sections, the m-derived section as well as the matching half -sections are shown in Fig. 16.

Fig. 16 Summary of composite filter design

Fig. 17: A four-stage composite filter

12
9 Reactance Plot
Fig. 18 shows a typical plot of reactances as a function of frequency for low-pass, high-pass, band-
pass, and band-stop constant-k filters. Likewise, Fig. 19 shows reactance plot for m-derived filters
when m = 0.7.
low-pass filter : m = 1 high-pass filter
4 2

2 0
X

X
0 -2

-2 -4
0.5 1 1.5 0.5 1 1.5
f/fc f/fc
band-pass filter band-stop filter
5 5

0 0
X

-5 -5
0.9 0.95 1 1.05 0.9 0.95 1 1.05
f/f0 f/f0
Fig. 18: Reactance plot for constant-k filters. The solid lines denote X1, the dash-dot lines denote X2
and the dash lines denote –4Z2.
low-pass filter : m = 0.7 high-pass filter
6 2

4
0
2
X

-2
0

-2 -4
0.5 1 1.5 0.5 1 1.5
f/fc f/fc
band-pass filter band-stop filter
5 5

0 0
X

-5 -5
0.9 0.95 1 1.05 0.9 0.95 1 1.05
f/f0 f/f0
Fig. 19: Reactance plot for m-derived filters when m = 0.7. The solid lines denote X1, the dash-dot
lines denote X2 and the dash lines denote –4X2.

13
12 Insertion Loss Method
The insertion loss method is based on the attenuation response or insertion loss of a filter. The
insertion loss or power loss ratio of a two-port network is given by:
Power available from the source Pinc 1
PLR   
Power delivered to load Pload 1    2
where  is the reflection coefficient looking into the filter (assume no loss in the filter).
Design of a filter using the insertion-loss approach usually begins by designing a normalized
low-pass prototype (LPP). The LPP is a low-pass filter with source resistance of 1 and cutoff
frequency of 1 Radian/s. Impedance transformation and frequency scaling are then applied to
denormalize the LPP and synthesize different type of filters with different cutoff frequencies. Fig. 20
summarizes the process of filter design by the insertion loss method.

Fig. 20 Summary of filter design by insertion loss method

Now, consider the reflection coefficient at the input port, which is given by
Z ( )  1
   , where the 1 source resistance is assumed.
Z ( )  1

Since V     v(t )e  jt dt and v(t) is a real function, V     V * ( ) . Similar result holds for


V (  ) V * ( )
I() as well. Thus, Z      *  Z * ( ) . Therefore,
I (  ) I ( )
Z (  )  1 Z * ( )  1
      * ( ) . It follows that
Z (  )  1 Z * ( )  1
( )   ( ) * ( )  ( )    * (  )    (  ) ,
2 2

hence ||2 is an even function of . Therefore, it can be written as a polynomial in 2:

M ( 2 )
  
2

M ( 2 )  N ( 2 )
where M and N are real polynomials in 2. Thus, the insertion loss can be rewritten as
M ( 2 )
PLR  1  ,
N ( 2 )
which is the form of physically realizable power loss ratio. This equation is used to specify desirable
filter responses.
Maximally Flat (Butterworth or binomial filter) This type of filter is optimum in the sense that it
provides the flattest possible passband response for a given filter complexity, or order (i.e., number of
passive elements). For a low-pass filter, it is specified by
PLR  1  k 2 2 N ,
where N denotes the order of the filter. The pass band extends from  = 0 to  =; at the band edge
the power loss ratio is 1 + k2. Typically, k is chosen to be 1 in order to make the band edge the -3 dB
point. Note that the first (2N-1) derivatives of the power loss ratio are zero at  = 0, resulting in the
maximally flat response.
Equal Ripple (Chebyshev filter) The insertion loss of this low-pass Chebyshev filter is specified by
Chebyshev polynomial as follows

14
PLR  1  k 2TN2   .
This leads to a sharp cutoff with the expense of amplitude ripples in the pass band. The maximum
pass band ripples are given by 1 + k2, thus the pass band ripple level is specified by k2.
Maximally Flat Delay (Bessel-Thomson filter) Flat delay simply implies constant phase velocity,
which in turn implies linear phase, since  = /up. The greatest advantage of this filter is that the
output signal is not distorted, which is desirable in most applications. The insertion loss of this low-
pass filter is specified by
PLR  k 2 BN2   ,
where k is chosen such that the insertion loss at  = 0 is unity and Bn(x) denotes the Bessel
polynomial of order n. The first 4 Bessel polynomials are
B1 ( x )  1  x; B2 ( x )  x 2  3 x  3; B3 ( x )  x 3  6 x 2  15 x  15; B4 ( x )  x 4  10 x 3  45 x 2  105 x  105
Higher-order polynomials can be found using the following recurrence formula:
Bn ( x)  (2n  1) Bn1 ( x)  x 2 Bn2 ( x) .
In fact, the coefficients of Bn(x) can be found directly by formula
(2n  k )! n
ck  n k
2 k!(n  k )!
and Bn ( x )  
k 0
ck x k .

The above insertion loss specification is obtained by setting x = j.

Now, consider the transfer function of the third-order low-pass filter given by
H ( s )  15 /( s 3  6 s 2  15s  15) ; s=j
The phase is given by
15   3
 ( )  arg H ( j )   tan 1 and the group delay becomes
15  6 2
d ( ) 6 4  45 2  225
D ( )    6 ,
d   6 4  45 2  225
which is approximately 1 for small i.e., low frequency range. The higher the order, the broader the
frequency range where group delay is flat.
Low-pass filter prototype
Fig. 21 shows the ladder circuit for low-pass filter prototype with the source impedance of 1 , where
their elements are defined as follows:
generator resistance (Fig. 21a) inductance for series inductors
g0   ; gk   ;
generator conductance (Fig. 21b) k 1 to N capacitance for shunt capacitors
load resistance if g N is a shunt capacitor
g N 1  
load conductanc e if g N is a series inductor
Element values in the figure are different depending on the filter type. They can be determined by
using tables 1, 2, 3, for maximally-flat filter, equal-ripple filter, and maximally-flat time delay,
respectively.

Fig. 21 Ladder circuits for low-pass filter prototype

Also, Fig. 22 shows attenuation versus normalized frequency for maximally flat filter prototypes.
Table 1[1]:

15
Table 2a[1]:

Table 2b:

Table 3[1]:

16
Likewise, Fig. 23 (a), 23 (b) show attenuation versus normalized frequency for equal-ripple filter
prototypes with (a) 0.5 dB ripple level and (b) 3.0 dB ripple level. Fig. 24 shows delay versus
normalized frequency for maximally flat group delay filter prototypes.

Fig. 22[2]

Fig. 23 (a) [2]

Fig. 23 (b) [2]

17
 Fig. 24 [3]
13 Filter Transformations
The low-pass filter prototypes are designed assuming that Rs = 1  and c = 1 rad/s. For practical
filters, these prototypes have to be scaled in terms of impedance and frequency. Furthermore, they can
also be converted to give high-pass, bandpass or bandstop characteristics.
Impedance and Frequency Scaling for Low-Pass Filters
Assuming the source impedance of R0, the new filter component values are obtained by
L ' R0 L ; C '  C / R0 ; R 's  R0 ; R ' L  R0 RL ,
where L, C and RL are the component values for the original prototype.
To change the cutoff frequency of a low-pass filter prototype from unity to c, the following
frequency scaling is required:
   / c ; P 'LR ( )  PLR ( / c ) ,
where c denotes the new cutoff frequency. This transformation can be viewed as a stretching of the
original passband. The new element values are determined by applying the frequency scaling to the
series reactances and shunt susceptances as follows:
 
jX k  j Lk  jL'k ; jBk  j Ck  jC 'k , which shows that the new element values are
c c
Lk Ck
L 'k  ; C 'k  .
c c
Combining the results due to both impedance and frequency scaling yields
R0 Lk Ck
L' k  ; C 'k  .
c R0c
Low-pass to high-pass transformation The frequency substitution
  c /  ; P ' LR ( )  PLR (c /  ) ,
can be used to convert a low-pass response to a high-pass response. This substitution maps  = 0 to 
= ±∞, and vice versa; cutoff occurs when  = ±c. The negative sign is needed to convert inductors
(and capacitors) to realizable capacitors (and inductors). Applying frequency substitution yields
c 1  1
jX k   j Lk  ; jBk   j c Ck  , which shows that the new element values
 jC 'k  jL'k
are
1 1
C 'k  ; L 'k  .
c Lk c C k
Combining the results due to both impedance and frequency substitution yields

18
 1 R0
C 'k  ; L'k  .
R0c Lk c Ck
Example (Low-pass filter design) Design a maximally flat low-pass filter with a cutoff frequency of 2
GHz, impedance of 50 , and at least 15 dB insertion loss at 3 GHz. Compute and plot the amplitude
response and group delay for f = 0 to 4 GHz, and compare with an equal-ripple (3.0 dB ripple) and
linear phase filter having the same order. [2]

Bandpass filter transformation Let 1, 2 denote the edges of the passband, then the low-pass to
bandpass transformation can be accomplished by the following frequency substitution:
0   0  1   0    1
       ;   2 ,
2  1  0     0   0
where  denotes the fractional bandwidth of the passband and 0 denotes the center frequency, which
is chosen to be the geometric mean of 1 and 2, i.e.,
0  12 ,
for simplification purposes. It follows that
1   0  1     1   2  02 
when 0,     0 ; when 1,   0    1   1 ;
  0     0     01 
1   0  1  22  02 
when 2,       1.
  0     02 
Therefore, the pass band exists in the range where the normalized frequency is between -1 and 1, as in
the case of low-pass filter prototype. Applying the frequency substitution in the expressions for series
reactances and shunt susceptances yields
1   0  L L 1 L 
jX k  j    Lk  j k  j 0 k  jL'k  j , L ' k  k , C 'k  ;
  0   0  C ' k  0 0 Lk
1   0  C k C 1 C 
jBk  j   Ck  j  j 0 k  j C ' k  j , C 'k  k , L'k  .
  0   0  L ' k  0 0 Ck

19
Thus, the low-pass filter elements are converted to series resonant circuits (low impedance at
resonance) in the series arms, and to parallel resonant circuits (high impedance at resonance) in the
shunt arms. Notice that both series and parallel resonator elements have a resonant frequency of 0.
Bandstop filter transformation The inverse transformation can be used to obtain a bandstop response.
Thus, the frequency substitution is given by
1
  
    0  , where  and 0 have the same definitions as before. Then following the
 0  
procedure used previously for the bandpass filter, the series inductors of the low-pass filter prototype
are converted to parallel LC circuits having element values given by
Lk 1
L 'k  , C 'k  .
0 0 Lk
Likewise, the shunt capacitors are converted to series LC circuits having element values given by
C k 1
C 'k  , L 'k  .
0 0 Ck
Table 4 summarizes the element transformations from a low-pass filter prototype to a high-pass,
bandpass, or bandstop filter. Notice that the impedance scaling is not included.
Table 4[2]

Example (Bandpass filter design) Design a bandpass filter having a 0.5 dB equal-ripple response, with
N = 3. The center frequency is 1 GHz, the bandwidth is 10%, and the impedance is 50 .[2]

Reference
1. G. L. Mathaei, L. Young, and E. M. T. Jones, Microwave Filters, Impedance-Matching Networks,
and Coupling Structures (Dedham, Mass.: Artech House, 1980)
2. D. M. Pozar, Microwave Engineering Third Edition (Haboken, NJ: John Wiley & Sons, Inc.,
2005)
3. W.-K. Chen, Passive and Active Filters (Haboken, NJ: John Wiley & Sons, Inc., 1986)

20