Ch03-H8518.

tex 19/9/2007 15: 13 page 37

CHAPTER 3
FILTER
Design

F
ilters occur so frequently in the instrumentation and
communications industries that no book covering the
field of RF circuit design could be complete without at f3dB
least one chapter devoted to the subject. Indeed, entire

Attenuation
books have been written on the art of filter design alone,
so this single chapter cannot possibly cover all aspects of all types
of filters. But it will familiarize you with the characteristics of
four of the most commonly used filters and will enable you to
design very quickly and easily a filter that will meet, or exceed,
most of the common filter requirements that you will encounter.
We will cover Butterworth, Chebyshev, and Bessel filters in all Frequency
of their common configurations: low-pass, high-pass, bandpass, (A) Low-pass
and bandstop. We will learn how to take advantage of the atten-
uation characteristics unique to each type of filter. Finally, we
will learn how to design some very powerful filters in as little
as 5 minutes by merely looking through a catalog to choose a
f3dB
design to suit your needs.
Attenuation

BA C K G R O U N D
In Chapter 2, the concept of resonance was explored and we
determined the effects that component value changes had on
resonant circuit operation. You should now be somewhat famil-
iar with the methods that are used in analyzing passive resonant Frequency
circuits to find quantities, such as loaded Q, insertion loss, and (B) High-pass
bandwidth. You should also be capable of designing one- or
two-resonator circuits for any loaded Q desired (or, at least,
determine why you cannot). Quite a few of the filter applica-
tions that you will encounter, however, cannot be satisfied with
f3dB f3dB
the simple bandpass arrangement given in Chapter 2. There are
Attenuation

occasions when, instead of passing a certain band of frequen-

cies while rejecting frequencies above and below (bandpass), we
would like to attenuate a small band of frequencies while pass-
ing all others. This type of filter is called, appropriately enough,
a bandstop filter. Still other requirements call for a low-pass or
high-pass response. The characteristic curves for these responses
are shown in Fig. 3-1. The low-pass filter will allow all signals
below a certain cutoff frequency to pass while attenuating all Frequency
others. A high-pass filter’s response is the mirror-image of the
low-pass response and attenuates all signals below a certain cut-
off frequency while allowing those above cutoff to pass. These FIG. 3-1. Typical filter response curves.
Ch03-H8518.tex 19/9/2007 15: 13 page 38

38 RF CIRCUIT DESIGN

types of response simply cannot be handled very well with the The loaded Q of this filter is dependent upon the individual
two-resonator bandpass designs of Chapter 2. Q’s of the series leg and the shunt leg where, assuming perfect
In this chapter, we will use the low-pass filter as our workhorse, components,
as all other responses will be derived from it. So let’s take a quick XL
Q1 = (Eq. 3-2)
look at a simple low-pass filter and examine its characteristics. Rs
Fig. 3-2 is an example of a very simple two-pole, or second-order and,
low-pass filter. The order of a filter is determined by the slope RL
of the attenuation curve it presents in the stopband. A second- Q2 = (Eq. 3-3)
Xc
order filter is one whose rolloff is a function of the frequency
squared, or 12 dB per octave. A third-order filter causes a rolloff and the total Q is:
that is proportional to frequency cubed, or 18 dB per octave. Q 1 Q2
Thus, the order of a filter can be equated with the number of Qtotal = (Eq. 3-4)
Q1 + Q 2
significant reactive elements that it presents to the source as the
signal deviates from the passband. If the total Q of the circuit is greater than about 0.5, then for
The circuit of Fig. 3-2 can be analyzed in much the same manner optimum transfer of power from the source to the load, Q1 should
as was done in Chapter 2. For instance, an examination of the equal Q2 . In this case, at the peak frequency, the response will
effects of loaded Q on the response would yield the family of approach 0-dB insertion loss. If the total Q of the network is
curves shown in Fig. 3-3. Surprisingly, even this circuit config- less than about 0.5, there will be no peak in the response and,
uration can cause a peak in the response. This is due to the fact for optimum transfer of power, Rs should equal RL . The peaking
that at some frequency, the inductor and capacitor will become of the filter’s response is commonly called ripple (defined in
resonant and, thus, peak the response if the loaded Q is high Chapter 2) and can vary considerably from one filter design to the
enough. The resonant frequency can be determined from next depending on the application. As shown, the two-element
filter exhibits only one response peak at the edge of the passband.
1
Fr = √ (Eq. 3-1) It can be shown that the number of peaks within the passband is
2π LC directly related to the number of elements in the filter by:
For low values of loaded Q, however, no response peak will be Number of Peaks = N − 1
noticed.
where
N = the number of elements.
Rs L

Thus, the three-element low-pass filter of Fig. 3-4 should exhibit

two response peaks as shown in Fig. 3-5. This is true only if the
C RL

Rs L

C C RL
FIG. 3-2. A simple low-pass filter.

10
FIG. 3-4. Three-element low-pass filter.
Attenuation (dB)

Q

20
0.
5

0 dB
1.
0

30 3 dB Ripple
Attenuation
2.
0

40
f3dB
5.
0

50
0.1 0.2 0.5 1.0 2.0 5.0 10
Frequency (f/fc) Frequency

FIG. 3-3. Typical two-pole filter response curves. FIG. 3-5. Typical response of a three-element low-pass filter.
Ch03-H8518.tex 19/9/2007 15: 13 page 39

Normalization and the Low-Pass Prototype 39

0 profound effect on the Q of the filter and, therefore, on

the passband ripple and shape factor of the filter. If a
10
filter is inserted between two resistance values for which
Attenuation (dB)

20 it was not designed, the performance will suffer to an

extent, depending upon the degree of error in the

Q


30 terminating impedance values.

1. 2 .
25 5
40 4. The final attenuation slope of the response is dependent
upon the order of the network. The order of the network

5. 0.0
50 is equal to the number of reactive elements in the

0
1
low-pass filter. Thus, a second-order network
60 (2 elements) falls off at a final attenuation slope of
0.1 0.2 0.5 1.0 2.0 5.0 10
Frequency (f/fc) 12 dB per octave, a third-order network (3 elements) at
the rate of 18 dB per octave, and so on, with the addition
of 6 dB per octave per element.
FIG. 3-6. Curves showing frequency response vs. loaded Q for
three-element low-pass filters.
M O D E R N F I LT E R D E S I G N
loaded Q is greater than one. Typical response curves for various Modern filter design has evolved through the years from a subject
values of loaded Q for the circuit given in Fig. 3-4 are shown in known only to specialists in the field (because of the advanced
Fig. 3-6. For all odd-order networks, the response at DC and at mathematics involved) to a practical well-organized catalog of
the upper edge of the passband approaches 0 dB with dips in the ready-to-use circuits available to anyone with a knowledge of
response between the two frequencies. All even-order networks eighth grade level math. In fact, an average individual with abso-
will produce an insertion loss at DC equal to the amount of lutely no prior practical filter design experience should be able
passband ripple in dB. Keep in mind, however, that either of to sit down, read this chapter, and within 30 minutes be able to
these two networks, if designed for low values of loaded Q, can design a practical high-pass, low-pass, bandpass, or bandstop
be made to exhibit little or no passband ripple. But, as you can filter to his specifications. It sounds simple and it is—once a few
see from Figs. 3-3 and 3-6, the elimination of passband ripple basic rules are memorized.
can be made only at the expense of bandwidth. The smaller the The approach we will take in all of the designs in this chap-
ripple that is allowed, the wider the bandwidth becomes and, ter will be to make use of the myriad of normalized low-pass
therefore, selectivity suffers. Optimum flatness in the passband prototypes that are now available to the designer. The actual
occurs when the loaded Q of the three-element circuit is equal design procedure is, therefore, nothing more than determining
to one (1). Any value of loaded Q that is less than one will cause your requirements and then finding a filter in a catalog that satis-
the response to roll off noticeably even at very low frequencies, fies these requirements. Each normalized element value is then
within the defined passband. Thus, not only is the selectivity scaled to the frequency and impedance you desire, and then trans-
poorer but the passband insertion loss is too. In an application formed to the type of response (bandpass, high-pass, bandstop)
where there is not much signal to begin with, an even further that you wish. With practice, the procedure becomes very simple
decrease in signal strength could be disastrous. and soon you will be defining and designing filters.
Now that we have taken a quick look at two representative low- The concept of normalization may at first seem foreign to the
pass filters and their associated responses, let’s discuss filters in person who is a newcomer to the field of filter design, and the
general: idea of transforming a low-pass filter into one that will give one
of the other three types of responses might seem absurd. The best
1. High-Q filters tend to exhibit a far greater initial slope
advice I can give (to anyone not familiar with these practices and
toward the stopband than their low-Q counterparts with
who might feel a bit skeptical at this point) is to press on. The only
the same number of elements. Thus, at any frequency in
way to truly realize the beauty and simplicity of this approach
the stopband, the attenuation will be greater for a high-Q
is to try a few actual designs. Once you try a few, you will be
filter than for one with a lower Q. The penalty for this
hooked, and any other approach to filter design will suddenly
improvement is the increase in passband ripple that must
seem tedious and unnecessarily complicated.
occur as a result.
2. Low-Q filters tend to have the flattest passband response
but their initial attenuation slope at the band edge is N O R MA L I ZAT I O N A N D T H E
small. Thus, the penalty for the reduced passband ripple L OW-PA S S P R OT OTY P E
is a decrease in the initial stopband attenuation.
In order to offer a catalog of useful filter circuits to the elec-
3. As with the resonant circuits discussed in Chapter 2, the tronic filter designer, it became necessary to standardize the
source and load resistors loading a filter will have a presentation of the material. Obviously, in practice, it would
Ch03-H8518.tex 19/9/2007 15: 13 page 40

40 RF CIRCUIT DESIGN

be extremely difficult to compare the performance and evalu-

ate the usefulness of two filter networks if they were operating 0 dB
under two totally different sets of circumstances. Similarly, the
3 dB
presentation of any comparative design information for filters, if

Attenuation
not standardized, would be totally useless. This concept of stan-
dardization or normalization, then, is merely a tool used by filter v
1
experts to present all filter design and performance information
in a manner useful to circuit designers. Normalization assures
the designer of the capability of comparing the performance of
any two filter types when given the same operating conditions. Frequency (v)

All of the catalogued filters in this chapter are low-pass fil-

ters normalized for a cutoff frequency of one radian per second
FIG. 3-8. The Butterworth response.
(0.159 Hz) and for source and load resistors of one ohm. A char-
acteristic response of such a filter is shown in Fig. 3-7. The circuit
used to generate this response is called the low-pass prototype. Since the Butterworth response is only a medium-Q filter, its
initial attenuation steepness is not as good as some filters but
it is better than others. This characteristic often causes the
0 dB
Butterworth response to be called a middle-of-the-road design.
Ripple The attenuation of a Butterworth filter is given by
Attenuation

 2n 
ω
AdB = 10 log 1 + (Eq. 3-5)
v
1 ωc

where
Frequency (v) ω = the frequency at which the attenuation is desired,
ωc = the cutoff frequency (ω3dB ) of the filter,
n = the number of elements in the filter.
FIG. 3-7. Normalized low-pass response.
If Equation 3-5 is evaluated at various frequencies for various
Obviously, the design of a filter with such a low cutoff frequency numbers of elements, a family of curves is generated which will
would require component values much larger than those we are give a very good graphical representation of the attenuation pro-
accustomed to working with; capacitor values would be in farads vided by any order of filter at any frequency. This information
rather than microfarads and picofarads, and the inductor values is illustrated in Fig. 3-9. Thus, from Fig. 3-9, a 5-element (fifth
would be in henries rather than in microhenries and nanohenries. order) Butterworth filter will provide an attenuation of approxi-
But once we choose a suitable low-pass prototype from the cata- mately 30 dB at a frequency equal to twice the cutoff frequency
log, we can change the impedance level and cutoff frequency of
the filter to any value we wish through a simple process called 0
scaling. The net result of this process is a practical filter design
12
with realizable component values.
24 n
2

F I LT E R TY P E S 36
3
Attenuation (dB)

Many of the filters used today bear the names of the men who 48
4
developed them. In this section, we will take a look at three 60
such filters and examine their attenuation characteristics. Their 5
72
relative merits will be discussed and their low-pass proto-
84 6
types presented. The three filter types discussed will be the
Butterworth, Chebyshev, and Bessel responses. 96 7
108
The Butterworth Response 120
The Butterworth filter is a medium-Q filter that is used in designs 1.0 1.5 2.0 2.5 3.0 3.5 4.0 5 6 7 8 9 10
that require the amplitude response of the filter to be as flat Frequency Ratio (f/fc)
as possible. The Butterworth response is the flattest passband
response available and contains no ripple. The typical response
of such a filter might look like that of Fig. 3-8. FIG. 3-9. Attenuation characteristics for Butterworth filters.
Ch03-H8518.tex 19/9/2007 15: 13 page 41

Filter Types 41

of the filter. Notice here that the frequency axis is normalized

to ω/ωc and the graph begins at the cutoff (−3 dB) point. This L2 L4
1
graph is extremely useful as it provides you with a method of
determining, at a glance, the order of a filter needed to meet a
given attenuation specification. A brief example should illustrate C1 C3 1
this point (Example 3-1).

n C1 L2 C3 L4 C5 L6 C7
EXAMPLE 3-1 2 1.414 1.414
How many elements are required to design a Butterworth 3 1.000 2.000 1.000
filter with a cutoff frequency of 50 MHz, if the filter must 4 0.765 1.848 1.848 0.765
provide at least 50 dB of attenuation at 150 MHz?
5 0.618 1.618 2.000 1.618 0.618

Solution 6 0.518 1.414 1.932 1.932 1.414 0.518

The first step in the solution is to find the ratio of 7 0.445 1.247 1.802 2.000 1.802 1.247 0.445
ω/ωc = f/fc . n L1 C2 L3 C4 L5 C6 L7
f 150 MHz
= L1 L3
fc 50 MHz 1
=3
Thus, at three times the cutoff frequency, the response C2 C4 1
must be down by at least 50 dB. Referring to Fig. 3-9, it is
seen very quickly that a minimum of 6 elements is
required to meet this design goal. At an f /f c of 3, a
6-element design would provide approximately 57 dB of TABLE 3-1. Butterworth Equal Termination Low-Pass Prototype Element
attenuation, while a 5-element design would provide only Values (Rs = RL )
about 47 dB, which is not quite good enough.

1.848 0.765
1

The element values for a normalized Butterworth low-pass filter

operating between equal 1-ohm terminations (source and load) 0.765 1.848 1
can be found by
(2k − 1)π
Ak = 2 sin , k = 1, 2, . . . n (Eq. 3-6)
2n
where
FIG. 3-10. A four-element Butterworth low-pass prototype circuit.
n is the number of elements,
Ak is the kth reactance in the ladder and may be either an
of the prototype ladder network. The 1-ohm load resistor is then
inductor or capacitor.
placed directly across the output of the filter.
The term (2k − 1)π/2n is in radians. We can use Equation 3-6 to Remember that the cutoff frequency of each filter is 1 radian per
generate our first entry into the catalog of low-pass prototypes second, or 0.159 Hz. Each capacitor value given is in farads, and
shown in Table 3-1. The placement of each component of the each inductor value is in henries. The network will later be scaled
filter is shown immediately above and below the table. to the impedance and frequency that is desired through a simple
The rules for interpreting Butterworth tables are simple. The multiplication and division process. The component values will
schematic shown above the table is used whenever the ratio Rs /RL then appear much more realistic.
is calculated as the design criteria. The table is read from the Occasionally, we have the need to design a filter that will operate
top down. Alternately, when RL /Rs is calculated, the schematic between two unequal terminations as shown in Fig. 3-11. In this
below the table is used. Then, the element designators in the case, the circuit is normalized for a load resistance of 1 ohm,
table are read from the bottom up. Thus, a four-element low- while taking what we get for the source resistance. Dividing both
pass prototype could appear as shown in Fig. 3-10. Note here the load and source resistor by 10 will yield a load resistance of
that the element values not given in Table 3-1 are simply left out 1 ohm and a source resistance of 5 ohms as shown in Fig. 3-12.
Ch03-H8518.tex 19/9/2007 15: 13 page 42

42 RF CIRCUIT DESIGN

Rs n Rs /RL C1 L2 C3 L4
Filter
50  0.300 1.838 0.440 5.363
RL 10 
0.200 2.669 0.284 7.910
0.100 5.167 0.138 15.455
∞ 1.500 1.333 0.500
4 1.111 0.466 1.592 1.744 1.469
FIG. 3-11. Unequal terminations.
1.250 0.388 1.695 1.511 1.811
Rs 1.429 0.325 1.862 1.291 2.175
Filter
5 1.667 0.269 2.103 1.082 2.613
RL 1 2.000 0.218 2.452 0.883 3.187
2.500 0.169 2.986 0.691 4.009
3.333 0.124 3.883 0.507 5.338
5.000 0.080 5.684 0.331 7.940
FIG. 3-12. Normalized unequal terminations.
10.000 0.039 11.094 0.162 15.642
∞ 1.531 1.577 1.082 0.383
We can use the normalized terminating resistors to help us find
n RL /Rs L1 C2 L3 C4
a low-pass prototype circuit.
Table 3-2 is a list of Butterworth low-pass prototype values L1 L3
Rs
for various ratios of source to load impedance (Rs /RL ). The
C2 C4 RL

L2 L4
Rs

C1 C3 RL TABLE 3-2A. (Continued)

schematic shown above the table is used when Rs /RL is cal-

n Rs /RL C1 L2 C3 L4 culated, and the element values are read down from the top of
2 1.111 1.035 1.835 the table. Alternately, when RL /Rs is calculated, the schematic
1.250 0.849 2.121
below the table is used while reading up from the bottom of the
table to get the element values (Example 3-2).
1.429 0.697 2.439
Obviously, all possible ratios of source to load resistance could
1.667 0.566 2.828 not possibly fit on a chart of this size. This, of course, leaves
2.000 0.448 3.346 the potential problem of not being able to find the ratio that you
2.500 0.342 4.095 need for a particular design task. The solution to this dilemma
is to simply choose a ratio that most closely matches the ratio
3.333 0.245 5.313 you need to complete the design. For ratios of 100:1 or so, the
5.000 0.156 7.707 best results are obtained if you assume this value to be so high
10.000 0.074 14.814 for practical purposes as to be infinite. Since, in these instances,
you are only approximating the ratio of source to load resistance,
∞ 1.414 0.707
the filter derived will only approximate the response that was
3 0.900 0.808 1.633 1.599 originally intended. This is usually not too much of a problem.
0.800 0.844 1.384 1.926
0.700 0.915 1.165 2.277 The Chebyshev Response
0.600 1.023 0.965 2.702 The Chebyshev filter is a high-Q filter that is used when: (1) a
0.500 1.181 0.779 3.261 steeper initial descent into the stopband is required, and (2) the
0.400 1.425 0.604 4.064
passband response is no longer required to be flat. With this type
of requirement, ripple can be allowed in the passband. As more
ripple is introduced, the initial slope at the beginning of the stop-
TABLE 3-2A. Butterworth Low-Pass Prototype Element Values band is increased and produces a more rectangular attenuation
Ch03-H8518.tex 19/9/2007 15: 13 page 43

Filter Types 43

Rs L2 L4 L6

C1 C3 C5 C7 RL

n Rs /RL C1 L2 C3 L4 C5 L6 C7
5 0.900 0.442 1.027 1.910 1.756 1.389
0.800 0.470 0.866 2.061 1.544 1.738
0.700 0.517 0.731 2.285 1.333 2.108
0.600 0.586 0.609 2.600 1.126 2.552
0.500 0.686 0.496 3.051 0.924 3.133
0.400 0.838 0.388 3.736 0.727 3.965
0.300 1.094 0.285 4.884 0.537 5.307
0.200 1.608 0.186 7.185 0.352 7.935
0.100 3.512 0.091 14.095 0.173 15.710
∞ 1.545 1.694 1.382 0.894 0.309
6 1.111 0.289 1.040 1.322 2.054 1.744 1.335
1.250 0.245 1.116 1.126 2.239 1.550 1.688
1.429 0.207 1.236 0.957 2.499 1.346 2.062
1.667 0.173 1.407 0.801 2.858 1.143 2.509
2.000 0.141 1.653 0.654 3.369 0.942 3.094
2.500 0.111 2.028 0.514 4.141 0.745 3.931
3.333 0.082 2.656 0.379 5.433 0.552 5.280
5.000 0.054 3.917 0.248 8.020 0.363 7.922
10.000 0.026 7.705 0.122 15.786 0.179 15.738
∞ 1.553 1.759 1.553 1.202 0.758 0.259
7 0.900 0.299 0.711 1.404 1.489 2.125 1.727 1.296
0.800 0.322 0.606 1.517 1.278 2.334 1.546 1.652
0.700 0.357 0.515 1.688 1.091 2.618 1.350 2.028
0.600 0.408 0.432 1.928 0.917 3.005 1.150 2.477
0.500 0.480 0.354 2.273 0.751 3.553 0.951 3.064
0.400 0.590 0.278 2.795 0.592 4.380 0.754 3.904
0.300 0.775 0.206 3.671 0.437 5.761 0.560 5.258
0.200 1.145 0.135 5.427 0.287 8.526 0.369 7.908
0.100 2.257 0.067 10.700 0.142 16.822 0.182 15.748
∞ 1.558 1.799 1.659 1.397 1.055 0.656 0.223
n RL /Rs L1 C2 L3 C4 L5 C6 L7

L1 L3 L5 L7
Rs

C2 C4 C6 RL

TABLE 3-2B. Butterworth Low-Pass Prototype Element Values

Ch03-H8518.tex 19/9/2007 15: 13 page 44

44 RF CIRCUIT DESIGN

n Chebyshev Polynomial
EXAMPLE 3-2  
ω
1
Find the low-pass prototype value for an n = 4 ωc

Butterworth filter with unequal terminations:  2

2 2 ω
−1
RS = 50 ohms, RL = 100 ohms. ωc
 3  
Solution 3 4 ω
ωc
−3 ω
ωc

Normalizing the two terminations for RL = 1 ohm will yield  4  2

4 8 ω
−8 ω
+1
a value of Rs = 0.5. Reading down from the top of ωc ωc

Table 3-2, for an n = 4 low-pass prototype value, we  5  3  

see that there is no Rs /RL = 0.5 ratio listed. Our second 5 16 ω
ωc
− 20 ω
ωc
+5 ω
ωc
choice, then, is to take the value of RL /Rs = 2, and read up  6  4  2
from the bottom of the table while using the schematic 6 32 ω
ωc
− 48 ω
ωc
+ 18 ω
ωc
−1
below the table as the form for the low-pass prototype  7  5  3  
values. This approach results in the low-pass prototype 7 64 ω
ωc
− 112 ω
ωc
+ 58 ω
ωc
−7 ω
ωc
circuit of Fig. 3-13.

TABLE 3-3. Chebyshev Polynomials to the Order n

0.218 0.883
0.5

The attenuation of a Chebyshev filter can be found by making a

2.452 3.187 1 few simple but tiresome calculations, and can be expressed as:
   
ω
AdB = 10 log 1 + ε2 Cn2 (Eq. 3-7)
ωc
where
FIG. 3-13. Low-pass prototype circuit for Example 3-2.  
ω
Cn2 is the Chebyshev polynomial to the order n evaluated
ω
 c
ω
at .
ωc
The Chebyshev polynomials for the first seven orders are given
0 in Table 3-3. The parameter ε is given by:
10
Attenuation (dB)

ε= 10 RdB /10 − 1 (Eq. 3-8)

20 Butterworth
Response
where
30
3 dB RdB is the passband ripple in decibels.
40
Chebyshev      
Response ω ω ω
Note that is not the same as . The quantity
ωc ωc ωc
1 2 3 4 can be found by defining another parameter:
Frequency (f/fc)  
1 −1 1
B = cosh (Eq. 3-9)
n ε
FIG. 3-14. Comparison of three-element Chebyshev and Butterworth
responses.
where
n = the order of the filter,
ε = the parameter defined in Equation 3-8,
cosh−1 = the inverse hyperbolic cosine of the quantity in
curve when compared to the rounded Butterworth response. This parentheses.
comparison is made in Fig. 3-14. Both curves are for n = 3 filters.
Finally, we have:
The Chebyshev response shown has 3 dB of passband ripple and    
produces a 10-dB improvement in stopband attenuation over the ω ω
= cosh B (Eq. 3-10)
Butterworth filter. ωc ωc
Ch03-H8518.tex 19/9/2007 15: 13 page 45

Filter Types 45

0 0
12 12
n
2 24 n
2
24
36 36

Attenuation (dB)
Attenuation (dB)

3 3
48 48

60 4 60 4
72
72
5 84 5
84
96
96 6 6
108
108 7
7 120
120 1.0 1.5 2.0 2.5 3.0 3.54.0 5 6 7 8 9 10
1.0 1.5 2.0 2.5 3.0 3.5 4.0 5 6 7 8 9 10 Frequency Ratio (f/fc)
Frequency Ratio (f/fc )

FIG. 3-16. Attenuation characteristics for a Chebyshev filter with 0.1-dB

FIG. 3-15. Attenuation characteristics for a Chebyshev filter with 0.01-dB
ripple.
ripple
0
12
where n
2
  24
ω
= the ratio of the frequency of interest to the cutoff 36 3
ωc
Attenuation (dB)

frequency,
4
cosh = the hyperbolic cosine. 60
5
72
If your calculator does not have hyperbolic and inverse hyper-
bolic functions, they can be manually determined from the 84 6
following relations: 96
7
−x 108
cosh x = 0.5(e + e )
x
120
1.0 1.5 2.0 2.5 3.0 3.5 4.0 5 6 7 8 9 10
and
Frequency Ratio (f/fc)

cosh−1 x = ln (x ± x 2 − 1)

The preceding equations yield families of attenuation curves, FIG. 3-17. Attenuation characteristics for a Chebyshev filter with 0.5-dB
each classified according to the amount of ripple allowed in ripple.
the passband. Several of these families of curves are shown in
0
Figs. 3-15 through 3-18, and include 0.01-dB, 0.1-dB, 0.5-dB,
and 1.0-dB ripple. Each curve begins at ω/ωc = 1, which is the 12
n
2
normalized cutoff, or 3-dB frequency. The passband ripple is, 24
therefore, not shown. 36 3
Attenuation (dB)

If other families of attenuation curves are needed with different 48 4

values of passband ripple, the preceding Chebyshev equations 60
can be used to derive them. The problem in Example 3-3 5
72
illustrates this. 6
84
Obviously, performing the calculations of Example 3-3 for 7
96
various values of ω/ωc , ripple, and filter order is a very time-
108
consuming chore unless a programmable calculator or computer
is available to do most of the work for you. 120
1.0 1.5 2.0 2.5 3.0 3.5 4.0 5 6 7 8 9 10
The low-pass prototype element values corresponding to the Frequency Ratio (f/fc)
Chebyshev responses of Figs. 3-15 through 3-18 are given in
Tables 3-4 through 3-7. Note that the Chebyshev prototype
values could not be separated into two distinct sets of tables FIG. 3-18. Attenuation characteristics for a Chebyshev filter with 1-dB
covering the equal and unequal termination cases, as was done ripple.
Ch03-H8518.tex 19/9/2007 15: 13 page 46

46 RF CIRCUIT DESIGN

n Rs /RL C1 L2 C3 L4
EXAMPLE 3-3
1.429 0.759 2.344
Find the attenuation of a 4-element, 2.5-dB ripple,
1.667 0.609 2.750
low-pass Chebyshev filter at ω/ωc = 2.5.
2.000 0.479 3.277
Solution 2.500 0.363 4.033
First evaluate the parameter: 3.333 0.259 5.255

ε = 102.5/10 − 1 5.000 0.164 7.650

= 0.882 10.000 0.078 14.749

∞ 1.412 0.742
Next, find B.
   3 1.000 1.181 1.821 1.181
1 1
B= cosh−1
4 0.882 0.900 1.092 1.660 1.480

= 0.1279 0.800 1.097 1.443 1.806

0.700 1.160 1.228 2.165
Then, (ω/ωc ) is:
  0.600 1.274 1.024 2.598
ω
= 2.5 cosh.1279
ωc 0.500 1.452 0.829 3.164
= 2.5204 0.400 1.734 0.645 3.974
0.300 2.216 0.470 5.280
Finally, we evaluate the fourth order (n = 4) Chebyshev
polynomial at (ω/ωc ) = 2.52. 0.200 3.193 0.305 7.834
   4  2 0.100 6.141 0.148 15.390
ω ω ω
C 2n =8 −8 +1 ∞ 1.501 1.433 0.591
ωc ωc ωc
4 1.100 0.950 1.938 1.761 1.046
= 8(2.5204)4 − 8(2.5204)2 + 1
1.111 0.854 1.946 1.744 1.165
= 273.05
1.250 0.618 2.075 1.542 1.617
We can now evaluate the final equation. 1.429 0.495 2.279 1.334 2.008
   
ω 1.667 0.398 2.571 1.128 2.461
AdB = 10 log10 1 + ε2 C2n
ωc 2.000 0.316 2.994 0.926 3.045
2 2 2.500 0.242 3.641 0.729 3.875
= 10 log10 [1 + (0.882) (273.05) ]
3.333 0.174 4.727 0.538 5.209
= 47.63 dB
5.000 0.112 6.910 0.352 7.813
Thus, at an ω/ωc , of 2.5, you can expect 47.63 dB of
10.000 0.054 13.469 0.173 15.510
attenuation for this filter.
∞ 1.529 1.694 1.312 0.523
n RL /Rs L1 C2 L3 C4

L2 L4 L1 L3
Rs Rs

C1 C3 RL C2 C4 RL

n RS /RL C1 L2 C3 L4
2 1.101 1.347 1.483 TABLE 3-4A. (Continued)

1.111 1.247 1.595

1.250 0.943 1.997
for the Butterworth prototypes. This is because the even order
(n = 2, 4, 6, . . .) Chebyshev filters cannot have equal termina-
tions. The source and load must always be different for proper
TABLE 3-4A. Chebyshev Low-Pass Element Values for 0.01-dB Ripple. operation as shown in the tables.
Ch03-H8518.tex 19/9/2007 15: 13 page 47

Rs L2 L4 L6

C1 C3 C5 C7 RL

n Rs /RL C1 L2 C3 L4 C5 L6 C7
5 1.000 0.977 1.685 2.037 1.685 0.977
0.900 0.880 1.456 2.174 1.641 1.274
0.800 0.877 1.235 2.379 1.499 1.607
0.700 0.926 1.040 2.658 1.323 1.977
0.600 1.019 0.863 3.041 1.135 2.424
0.500 1.166 0.699 3.584 0.942 3.009
0.400 1.398 0.544 4.403 0.749 3.845
0.300 1.797 0.398 5.772 0.557 5.193
0.200 2.604 0.259 8.514 0.368 7.826
0.100 5.041 0.127 16.741 0.182 15.613
∞ 1.547 1.795 1.645 1.237 0.488
6 1.101 0.851 1.796 1.841 2.027 1.631 0.937
1.111 0.760 1.782 1.775 2.094 1.638 1.053
1.250 0.545 1.864 1.489 2.403 1.507 1.504
1.429 0.436 2.038 1.266 2.735 1.332 1.899
1.667 0.351 2.298 1.061 3.167 1.145 2.357
2.000 0.279 2.678 0.867 3.768 0.954 2.948
2.500 0.214 3.261 0.682 4.667 0.761 3.790
3.333 0.155 4.245 0.503 6.163 0.568 5.143
5.000 0.100 6.223 0.330 9.151 0.376 7.785
10.000 0.048 12.171 0.162 18.105 0.187 15.595
∞ 1.551 1.847 1.790 1.598 1.190 0.469
7 1.000 0.913 1.595 2.002 1.870 2.002 1.595 0.913
0.900 0.816 1.362 2.089 1.722 2.202 1.581 1.206
0.800 0.811 1.150 2.262 1.525 2.465 1.464 1.538
0.700 0.857 0.967 2.516 1.323 2.802 1.307 1.910
0.600 0.943 0.803 2.872 1.124 3.250 1.131 2.359
0.500 1.080 0.650 3.382 0.928 3.875 0.947 2.948
0.400 1.297 0.507 4.156 0.735 4.812 0.758 3.790
0.300 1.669 0.372 5.454 0.546 6.370 0.568 5.148
0.200 2.242 0.242 8.057 0.360 9.484 0.378 7.802
0.100 4.701 0.119 15.872 0.178 18.818 0.188 15.652
∞ 1.559 1.867 1.866 1.765 1.563 1.161 0.456
n RL /Rs L1 C2 L3 C4 L5 C6 L7

L1 L3 L5 L7
Rs

C2 C4 C6 RL

TABLE 3-4B. Chebyshev Low-Pass Element Values for 0.01-dB Ripple.

Ch03-H8518.tex 19/9/2007 15: 13 page 48

48 RF CIRCUIT DESIGN

The rules used for interpreting the Butterworth tables apply here
L2 L4 also. The schematic shown above the table is used, and the
Rs
element designators are read down from the top, when the ratio
Rs /RL is calculated as a design criteria. Alternately, with RL /Rs
C1 C3 RL calculations, use the schematic given below the table and read
the element designators upwards from the bottom of the table.
Example 3-4 is a practice problem for use in understanding the
procedure.
n Rs /RL C1 L2 C3 L4
2 1.355 1.209 1.638
1.429 0.977 1.982 EXAMPLE 3-4
1.667 0.733 2.489 Find the low-pass prototype values for an n = 5, 0.1-dB
2.000 0.560 3.054 ripple, Chebyshev filter if the source resistance you are
designing for is 50 ohms and the load resistance is
2.500 0.417 3.827
250 ohms.
3.333 0.293 5.050
5.000 0.184 7.426 Solution
10.000 0.087 14.433 Normalization of the source and load resistors yields an
∞ 1.391 0.819 Rs /RL = 0.2. A look at Table 3-5, for a 0.1-dB ripple filter
3 1.000 1.433 1.594 1.433 with an n = 5 and an Rs /RL = 0.2, yields the circuit values
shown in Fig. 3-19.
0.900 1.426 1.494 1.622
0.800 1.451 1.356 1.871
0.295 0.366
0.700 1.521 1.193 2.190
L2 L4
0.600 1.648 1.017 2.603 0.2

0.500 1.853 0.838 3.159

0.400 2.186 0.660 3.968 3.546 9.127 7.889 1
0.300 2.763 0.486 5.279 C1 C3 C5

0.200 3.942 0.317 7.850

1.100 7.512 0.155 15.466
∞ 1.513 1.510 0.716
FIG. 3-19. Low-pass prototype circuit for Example 3-4.
4 1.355 0.992 2.148 1.585 1.341
1.429 0.779 2.348 1.429 1.700
1.667 0.576 2.730 1.185 2.243
2.000 0.440 3.227 0.967 2.856 It should be mentioned here that equations could have been pre-
sented in this section for deriving the element values for the
2.500 0.329 3.961 0.760 3.698
Chebyshev low-pass prototypes. The equations are extremely
3.333 0.233 5.178 0.560 5.030 long and tedious, however, and there would be little to be gained
5.000 0.148 7.607 0.367 7.614 from their presentation.
10.000 0.070 14.887 0.180 15.230
∞ 1.511 1.768 1.455 0.673 The Bessel Filter
n RL /Rs L1 C2 L3 C4
The initial stopband attenuation of the Bessel filter is very poor
and can be approximated by:

Rs
L1 L3  2
ω
AdB = 3 (Eq. 3-11)
ωc
C2 C4 RL

This expression, however, is not very accurate above an ω/ωc

that is equal to about 2. For values of ω/ωc greater than
2, a straight-line approximation of 6 dB per octave per ele-
TABLE 3-5A. Chebyshev Low-Pass Prototype Element Values for 0.1-dB ment can be made. This yields the family of curves shown in
Ripple Fig. 3-20.
Ch03-H8518.tex 19/9/2007 15: 13 page 49

L2 L4 L6
Rs

C1 C3 C5 C7 RL

n Rs /RL C1 L2 C3 L4 C5 L6 C7
5 1.000 1.301 1.556 2.241 1.556 1.301
0.900 1.285 1.433 2.380 1.488 1.488
0.800 1.300 1.282 2.582 1.382 1.738
0.700 1.358 1.117 2.868 1.244 2.062
0.600 1.470 0.947 3.269 1.085 2.484
0.500 1.654 0.778 3.845 0.913 3.055
0.400 1.954 0.612 4.720 0.733 3.886
0.300 2.477 0.451 6.196 0.550 5.237
0.200 3.546 0.295 9.127 0.366 7.889
0.100 6.787 0.115 17.957 0.182 15.745
∞ 1.561 1.807 1.766 1.417 0.651
6 1.355 0.942 2.080 1.659 2.247 1.534 1.277
1.429 0.735 2.249 1.454 2.544 1.405 1.629
1.667 0.542 2.600 1.183 3.064 1.185 2.174
2.000 0.414 3.068 0.958 3.712 0.979 2.794
2.500 0.310 3.765 0.749 4.651 0.778 3.645
3.333 0.220 4.927 0.551 6.195 0.580 4.996
5.000 0.139 7.250 0.361 9.261 0.384 7.618
10.000 0.067 14.220 0.178 18.427 0.190 15.350
∞ 1.534 1.884 1.831 1.749 1.394 0.638
7 1.000 1.262 1.520 2.239 1.680 2.239 1.520 1.262
0.900 1.242 1.395 2.361 1.578 2.397 1.459 1.447
0.800 1.255 1.245 2.548 1.443 2.624 1.362 1.697
0.700 1.310 1.083 2.819 1.283 2.942 1.233 2.021
0.600 1.417 0.917 3.205 1.209 3.384 1.081 2.444
0.500 1.595 0.753 3.764 0.928 4.015 0.914 3.018
0.400 1.885 0.593 4.618 0.742 4.970 0.738 3.855
0.300 2.392 0.437 6.054 0.556 6.569 0.557 5.217
0.200 3.428 0.286 8.937 0.369 9.770 0.372 7.890
0.100 6.570 0.141 17.603 0.184 19.376 0.186 15.813
∞ 1.575 1.858 1.921 1.827 1.734 1.379 0.631
n RL /Rs L1 C2 L3 C4 L5 C6 L7

Rs L1 L3 L5 L7

C2 C4 C6 RL

TABLE 3-5B. Chebyshev Low-Pass Prototype Element Values for 0.1-dB Ripple
Ch03-H8518.tex 19/9/2007 15: 13 page 50

50 RF CIRCUIT DESIGN

0
L2 L4
Rs

20
C1 C3 RL n

40 3
n Rs /RL C1 L2 C3 L4

Attenuation (dB)
2 1.984 0.983 1.950
4
2.000 0.909 2.103
2.500 0.564 3.165 60 5
3.333 0.375 4.411 6
5.000 0.228 6.700
7
10.000 0.105 13.322 80
∞ 1.307 0.975
3 1.000 1.864 1.280 1.834
0.900 1.918 1.209 2.026 100
0.800 1.997 1.120 2.237
0.700 2.114 1.015 2.517
0.500 2.557 0.759 3.436 120
1 2 3 4 5 6 8 10
0.400 2.985 0.615 4.242
Frequency (f/fc)
0.300 3.729 0.463 5.576
0.200 5.254 0.309 8.225
0.100 9.890 0.153 16.118 FIG. 3-20. Attenuation characteristics of Bessel filters.
∞ 1.572 1.518 0.932
phase nonlinearity results in distortion of wideband signals due
4 1.984 0.920 2.586 1.304 1.826
to the widely varying time delays associated with the different
2.000 0.845 2.720 1.238 1.985
spectral components of the signal. Bessel filters, on the other
2.500 0.516 3.766 0.869 3.121 hand, with their maximally flat (constant) group delay are able
3.333 0.344 5.120 0.621 4.480 to pass wideband signals with a minimum of distortion, while
5.000 0.210 7.708 0.400 6.987 still providing some selectivity.
10.000 0.098 15.352 0.194 14.262
∞ 1.436 1.889 1.521 0.913 The low-pass prototype element values for the Bessel filter are
n RL /Rs L1 C2 L3 C4
given in Table 3-8. Table 3-8 tabulates the prototype element
values for various ratios of source to load resistance.
L1 L3
Rs
F R E Q U E N CY A N D I M P E DA N C E S CA L I N G
Once you specify the filter, choose the appropriate attenuation
C2 C4 RL
response, and write down the low-pass prototype values, the
next step is to transform the prototype circuit into a usable fil-
ter. Remember, the cutoff frequency of the prototype circuit is
0.159 Hz (ω = 1 rad/sec), and it operates between a source and
a load resistance that are normalized so that RL = 1 ohm.
TABLE 3-6A. Chebyshev Low-Pass Prototype Element Values for 0.5-dB
Ripple The transformation is effected through the following formulas:
Cn
C= (Eq. 3-12)
But why would anyone deliberately design a filter with very 2πfc R
poor initial stopband attenuation characteristics? The Bessel fil- and
ter was originally optimized to obtain a maximally flat group RLn
L= (Eq. 3-13)
delay or linear phase characteristic in the filter’s passband. Thus, 2πfc
selectivity or stopband attenuation is not a primary concern
when dealing with the Bessel filter. In high- and medium-Q where
filters, such as the Chebyshev and Butterworth filters, the phase C = the final capacitor value,
response is extremely nonlinear over the filter’s passband. This L = the final inductor value,
Ch03-H8518.tex 19/9/2007 15: 13 page 51

Frequency and Impedance Scaling 51

L2 L4 L6
Rs

C1 C3 C5 C7 RL

n Rs /RL C1 L2 C3 L4 C5 L6 C7
5 1.000 1.807 1.303 2.691 1.303 1.807
0.900 1.854 1.222 2.849 1.238 1.970
0.800 1.926 1.126 3.060 1.157 2.185
0.700 2.035 1.015 3.353 1.058 2.470
0.600 2.200 0.890 3.765 0.942 2.861
0.500 2.457 0.754 4.367 0.810 3.414
0.400 2.870 0.609 5.296 0.664 4.245
0.300 3.588 0.459 6.871 0.508 5.625
0.200 5.064 0.306 10.054 0.343 8.367
0.100 9.556 0.153 19.647 0.173 16.574
∞ 1.630 1.740 1.922 1.514 0.903
6 1.984 0.905 2.577 1.368 2.713 1.299 1.796
2.000 0.830 2.704 1.291 2.872 1.237 1.956
2.500 0.506 3.722 0.890 4.109 0.881 3.103
3.333 0.337 5.055 0.632 5.699 0.635 4.481
5.000 0.206 7.615 0.406 8.732 0.412 7.031
10.000 0.096 15.186 0.197 17.681 0.202 14.433
7 1.000 1.790 1.296 2.718 1.385 2.718 1.296 1.790
0.900 1.835 1.215 2.869 1.308 2.883 1.234 1.953
0.800 1.905 1.118 3.076 1.215 3.107 1.155 2.168
0.700 2.011 1.007 3.364 1.105 3.416 1.058 2.455
0.600 2.174 0.882 3.772 0.979 3.852 0.944 2.848
0.500 2.428 0.747 4.370 0.838 2.289 0.814 3.405
0.400 2.835 0.604 5.295 0.685 5.470 0.669 4.243
0.300 3.546 0.455 6.867 0.522 7.134 0.513 5.635
0.200 5.007 0.303 10.049 0.352 10.496 0.348 8.404
0.100 9.456 0.151 19.649 0.178 20.631 0.176 16.665
∞ 1.646 1.777 2.031 1.789 1.924 1.503 0.895
n RL /Rs L1 C2 L3 C4 L5 C6 L7

Rs L1 L3 L5 L7

C2 C4 C6 RL

TABLE 3-6B. Chebyshev Low-Pass Prototype Element Values for 0.5-dB Ripple
Ch03-H8518.tex 19/9/2007 15: 13 page 52

52 RF CIRCUIT DESIGN

EXAMPLE 3-5
L2 L4
Rs
Scale the low-pass prototype values of Fig. 3-19
(Example 3-4) to a cutoff frequency of 50 MHz and a load
C1 C3 RL resistance of 250 ohms.
Solution
Use Equations 3-12 and 3-13 to scale each component as
n Rs /RL C1 L2 C3 L4 follows:
3.546
2 3.000 0.572 3.132 C1 =
2π(50 × 106 )(250)
4.000 0.365 4.600 = 45 pF
8.000 0.157 9.658 9.127
C3 =
∞ 1.213 1.109 2π(50 × 106 )(250)
3 1.000 2.216 1.088 2.216 = 116 pF
0.500 4.431 0.817 2.216 7.889
C5 =
0.333 6.647 0.726 2.216 2π(50 × 106 )(250)

0.250 8.862 0.680 2.216 = 100 pF

0.125 17.725 0.612 2.216 (250)(0.295)

L2 =
2π(50 × 106 )
∞ 1.652 1.460 1.108
= 235 nH
4 3.000 0.653 4.411 0.814 2.535
(250)(0.366)
4.000 0.452 7.083 0.612 2.848 L4 =
2π(50 × 106 )
8.000 0.209 17.164 0.428 3.281 = 291 nH
∞ 1.350 2.010 1.488 1.106 The source resistance is scaled by multiplying its
n RL /Rs L1 C2 L3 C4 normalized value by the final value of the load resistor.
Rs(final) = 0.2(250)
L1 L3
Rs = 50 ohms
The final circuit appears in Fig. 3-21.
C2 C4 RL

50  235 nH 291 nH

45 pF 116 pF 100 pF 250 

TABLE 3-7A. Chebyshev Low-Pass Prototype Element Values for 1.0-dB
Ripple

Cn = a low-pass prototype element value,

FIG. 3-21. Low-pass filter circuit for Example 3-5.
Ln = a low-pass prototype element value,
R = the final load resistor value,
fc = the final cutoff frequency. 2. Normalize the frequencies of interest by dividing them
The normalized low-pass prototype source resistor must also be by the cutoff frequency of the filter. This step forces your
transformed to its final value by multiplying it by the final value data to be in the same form as that of the attenuation
of the load resistor (Example 3-5). Thus, the ratio of the two curves of this chapter, where the 3-dB point on the
always remains the same. curve is:
The process for designing a low-pass filter is a very simple one f
=1
which involves the following procedure: fc

1. Define the response you need by specifying the required 3. Determine the maximum amount of ripple that you can
attenuation characteristics at selected frequencies. allow in the passband. Remember, the greater the amount
Ch03-H8518.tex 19/9/2007 15: 13 page 53

High-Pass Filter Design 53

L2 L4 L6
Rs

C1 C3 C5 C7 RL

n Rs /RL C1 L2 C3 L4 C5 L6 C7
5 1.000 2.207 1.128 3.103 1.128 2.207
0.500 4.414 0.565 4.653 1.128 2.207
0.333 6.622 0.376 6.205 1.128 2.207
0.250 8.829 0.282 7.756 1.128 2.207
0.125 17.657 0.141 13.961 1.128 2.207
∞ 1.721 1.645 2.061 1.493 1.103
6 3.000 0.679 3.873 0.771 4.711 0.969 2.406
4.000 0.481 5.644 0.476 7.351 0.849 2.582
8.000 0.227 12.310 0.198 16.740 0.726 2.800
∞ 1.378 2.097 1.690 2.074 1.494 1.102
7 1.000 2.204 1.131 3.147 1.194 3.147 1.131 2.204
0.500 4.408 0.566 6.293 0.895 3.147 1.131 2.204
0.333 6.612 0.377 9.441 0.796 3.147 1.131 2.204
0.250 8.815 0.283 12.588 0.747 3.147 1.131 2.204
0.125 17.631 0.141 25.175 0.671 3.147 1.131 2.204
∞ 1.741 1.677 2.155 1.703 2.079 1.494 1.102
n RL /Rs L1 C2 L3 C4 L5 C6 L7

L1 L3 L5 L7
Rs

C2 C4 C6 RL

TABLE 3-7B. Chebyshev Low-Pass Prototype Element Values for 1.0-dB Ripple

of ripple allowed, the more selective the filter is. Higher Example 3-6 diagrams the process of designing a low-pass filter
values of ripple may allow you to eliminate a few using the preceding steps.
components.
4. Match the normalized attenuation characteristics (Steps 1
and 2) with the attenuation curves provided in this
chapter. Allow yourself a small “fudge-factor” for good H I G H-PA S S F I LT E R D E S I G N
measure. This step reveals the minimum number of Once you have learned the mechanics of low-pass filter design,
circuit elements that you can get away with–given a high-pass design becomes a snap. You can use all of the atten-
certain filter type. uation response curves presented, thus far, for the low-pass
filters by simply inverting the will produce an attenuation of
5. Find the low-pass prototype values in the tables.
about 60 dB at an f /fc of 3 (Fig. 3-16). If you were working
6. Scale all elements to the frequency and impedance of the instead with a high-pass filter of the same size and type, you
final design. could still use Fig. 3-16 to tell you that at an f /fc of 1/3 (or,
Ch03-H8518.tex 19/9/2007 15: 13 page 54

54 RF CIRCUIT DESIGN

n Rs /RL C1 L2 C3 L4
L2 L4
Rs ∞ 1.501 0.978 0.613 0.211
n RL /Rs L1 C2 L3 C4
C1 C3 RL
L1 L3
Rs

C2 C4 RL
n Rs /RL C1 L2 C3 L4
2 1.000 0.576 2.148
1.111 0.508 2.310
1.250 0.443 2.510 TABLE 3-8A. (Continued)
1.429 0.380 2.764
1.667 0.319 3.099
2.000 0.260 3.565 EXAMPLE 3-6
2.500 0.203 4.258 Design a low-pass filter to meet the following
3.333 0.149 5.405 specifications:
5.000 0.097 7.688 fc = 35 MHz,
10.000 0.047 14.510
Response greater than 60 dB down at 105 MHz,
∞ 1.362 0.454 Maximally flat passband—no ripple,
3 1.000 0.337 0.971 2.203
Rs = 50 ohms,
0.900 0.371 0.865 2.375
0.800 0.412 0.761 2.587 RL = 500 ohms.

0.700 0.466 0.658 2.858

Solution
0.600 0.537 0.558 3.216
The need for a maximally flat passband automatically
0.500 0.635 0.459 3.714 indicates that the design must be a Butterworth response.
0.400 0.783 0.362 4.457 The first step in the design process is to normalize
everything. Thus,
0.300 1.028 0.267 5.689
0.200 1.518 0.175 8.140 Rs 50
=
RL 500
0.100 2.983 0.086 15.470
= 0.1
∞ 1.463 0.843 0.293
4 1.000 0.233 0.673 1.082 2.240 Next, normalize the frequencies of interest so that they
may be found in the graph of Fig. 3-9. Thus, we have:
1.111 0.209 0.742 0.967 2.414
1.250 0.184 0.829 0.853 2.630 f60dB 105 MHz
=
1.429 0.160 0.941 0.741 2.907 f3dB 35 MHz
=3
1.667 0.136 1.089 0.630 3.273
2.000 0.112 1.295 0.520 3.782 We next look at Fig. 3-9 and find a response that is down
at least 60 dB at a frequency ratio of f /f c = 3. Fig. 3-9
2.500 0.089 1.604 0.412 4.543 indicates that it will take a minimum of 7 elements to
3.333 0.066 2.117 0.306 5.805 provide the attenuation specified. Referring to the catalog
5.000 0.043 3.142 0.201 8.319 of Butterworth low-pass prototype values given in
Table 3-2 yields the prototype circuit of Fig. 3-22.
10.000 0.021 6.209 0.099 15.837
Continued on next page

TABLE 3-8A. Bessel Low-Pass Prototype Element Values

Ch03-H8518.tex 19/9/2007 15: 13 page 55

High-Pass Filter Design 55

After finding the response that satisfies all of the require-

Example 3-6.—Cont ments, the next step is to simply refer to the tables of low-pass
prototype values and copy down the prototype values that are
0.067 0.142 0.182
L2 L4 L6 called for. High-pass values for the elements are then obtained
0.100 directly from the low-pass prototype values as follows (refer to
Fig. 3-24):
2.257 10.700 16.822 15.748
1.000
C1 C3 C5 C7 1.821
L2
1

1.181 1.181
FIG. 3-22. Low-pass prototype circuit for Example 3-6. C1 C3 1

We then scale these values using Equations 3-12 and

3-13. The first two values are worked out for you.
(A) Low-pass prototype circuit
2.257
C1 =
2π(35 × 106 )500 0.549
= 21 pF 1 C2

(500)(0.067)
L2 = 0.847 0.847
2π(35 × 106 ) 1
L1 L3
= 152 nH
Similarly,

C3 = 97 pF,
(B) Equivalent high-pass prototype circuit
C5 = 153 pF,
C7 = 143 pF,
FIG 3-24. Low-pass to high-pass filter transformation.
L4 = 323 nH,
L6 = 414 nH, Simply replace each filter element with an element of the
RS = 50 ohms, opposite type and with a reciprocal value. Thus, L1 of Fig. 3-
24B is equal to 1/C1 of Fig. 3-24A. Likewise, C2 = 1/L2 and
RL = 500 ohms.
L3 = 1/C3 . Stated another way, if the low-pass prototype indi-
cates a capacitor of 1.181 farads, then use an inductor with
The final circuit is shown in Fig. 3-23.
a value of 1/1.181 = 0.847 henry, instead, for a high-pass
design. However, the source and load resistors should not be
50  152 nH 323 nH 414 nH altered.

21 pF 97 pF 153 pF 143 pF 500  The transformation process results in an attenuation character-

istic for the high-pass filter that is an exact mirror image of the
low-pass attenuation characteristic. The ripple, if there is any,
remains the same and the magnitude of the slope of the stopband
(or passband) skirts remains the same. Example 3-7 illustrates
FIG. 3-23. Low-pass filter circuit for Example 3-6.
the design of high-pass filters.
A closer look at the filter designed in Example 3-7 reveals that it
is symmetric. Indeed, all filters given for the equal termination
fc /f = 3) a 5-element, 0.1-dB-ripple Chebyshev high-pass fil- class are symmetric. The equal termination class of filter thus
ter will also produce an attenuation of 60 dB. This is obviously yields a circuit that is easier to design (fewer calculations) and,
more convenient than having to refer to more than one set of in most cases, cheaper to build for a high-volume product, due
curves. to the number of equal valued components.
Ch03-H8518.tex 19/9/2007 15: 13 page 56

L2 L4 L6
Rs

C1 C3 C5 C7 RL

n RS /RL C1 L2 C3 L4 C5 L6 C7
5 1.000 0.174 0.507 0.804 1.111 2.258
0.900 0.193 0.454 0.889 0.995 2.433
0.800 0.215 0.402 0.996 0.879 2.650
0.700 0.245 0.349 1.132 0.764 2.927
0.600 0.284 0.298 1.314 0.651 3.295
0.500 0.338 0.247 1.567 0.538 3.808
0.400 0.419 0.196 1.946 0.427 4.573
0.300 0.555 0.146 2.577 0.317 5.843
0.200 0.825 0.096 3.835 0.210 8.375
0.100 1.635 0.048 7.604 0.104 15.949
∞ 1.513 1.023 0.753 0.473 0.162
6 1.000 0.137 0.400 0.639 0.854 1.113 2.265
1.111 0.122 0.443 0.573 0.946 0.996 2.439
1.250 0.108 0.496 0.508 1.060 0.881 2.655
1.429 0.094 0.564 0.442 1.207 0.767 2.933
1.667 0.080 0.655 0.378 1.402 0.653 3.300
2.000 0.067 0.782 0.313 1.675 0.541 3.812
2.500 0.053 0.973 0.249 2.084 0.429 4.577
3.333 0.040 1.289 0.186 2.763 0.319 5.847
5.000 0.026 1.289 0.123 4.120 0.211 8.378
10.000 0.013 3.815 0.061 8.186 0.105 15.951
∞ 1.512 1.033 0.813 0.607 0.379 0.129
7 1.000 0.111 0.326 0.525 0.702 0.869 1.105 2.266
0.900 0.122 0.292 0.582 0.630 0.963 0.990 2.440
0.800 0.137 0.259 6.652 0.559 1.080 0.875 2.656
0.700 0.156 0.226 0.743 0.487 1.231 0.762 2.932
0.600 0.182 0.193 0.863 0.416 1.431 0.649 3.298
0.500 0.217 0.160 1.032 0.346 1.711 0.537 3.809
0.400 0.270 0.127 1.285 0.276 2.130 0.427 4.572
0.300 0.358 0.095 1.705 0.206 2.828 0.318 5.838
0.200 0.534 0.063 2.545 0.137 4.221 0.210 8.362
0.100 1.061 0.031 5.062 0.068 8.397 0.104 15.917
∞ 1.509 1.029 0.835 0.675 0.503 0.311 0.105
n RL /Rs L1 C2 L3 C4 L5 C6 L7

L1 L3 L5 L7
Rs

C2 C4 C6 RL

TABLE 3-8B. Bessel Low-Pass Prototype Element Values

Ch03-H8518.tex 19/9/2007 15: 13 page 57

Bandpass Filter Design 57

T H E D UA L N ETW O R K
Thus far, we have been referring to the group of low-pass pro- EXAMPLE 3-7
totype element value tables presented, and then we choose the Design an LC high-pass filter with an fc of 60 MHz and a
schematic that is located either above or below the tables for minimum attenuation of 40 dB at 30 MHz. The source and
the form of the filter that we are designing, depending on the load resistance are equal at 300 ohms. Assume that a
value of RL /Rs . Either form of the filter will produce exactly the 0.5-dB passband ripple is tolerable.
same attenuation, phase, and group-delay characteristics, and
each form is called the dual of the other. Solution
First, normalize the attenuation requirements so that the
Any filter network in a ladder arrangement, such as the ones
low-pass attenuation curves may be used.
presented in this chapter, can be changed into its dual form by
application of the following rules: f 30 MHz
=
fc 60 MHz
1. Change all inductors to capacitors, and vice-versa, = 0.5
without changing element values. Thus, 3 henries
Inverting, we get:
becomes 3 farads.
fc
2. Change all resistances into conductances, and vice-versa, =2
f
with the value unchanged. Thus, 3 ohms becomes
3 mhos, or 13 ohm. Now, select a normalized low-pass filter that offers at least
40-dB attenuation at a ratio of f c /f = 2. Reference to
3. Change all shunt branches to series branches, and vice Fig. 3.17 (attenuation response of 0.5-dB-ripple
versa. Chebyshev filters) indicates that a normalized n = 5
4. Change all elements in series with each other into Chebyshev will provide the needed attenuation. Table 3-6
elements that are in parallel with each other. contains the element values for the corresponding
network. The normalized low-pass prototype circuit is
5. Change all voltage sources into current sources, and vice
shown in Fig. 3-25A. Note that the schematic below
versa.
Table 3-6B was chosen as the low-pass prototype circuit
rather than the schematic above the table. The reason for
Fig. 3-26 shows a ladder network and its dual representation. doing this will become obvious after the next step. Keep in
mind, however, that the ratio of Rs /RL is the same as the
Dual networks are convenient, in the case of equal terminations,
ratio of RL /Rs , and is unity. Therefore, it does not matter
if you desire to change the topology of the filter without changing
which form is used for the prototype circuit.
the response. It is most often used, as shown in Example 3-7, to
eliminate an unnecessary inductor which might have crept into Next, transform the low-pass circuit to a high-pass
the design through some other transformation process. Inductors network by replacing each inductor with a capacitor, and
are typically more lower-Q devices than capacitors and, there- vice versa, using reciprocal element values as shown in
fore, exhibit higher losses. These losses tend to cause insertion Fig. 3-25B. Note here that, had we begun with the
loss, in addition to generally degrading the overall performance low-pass prototype circuit shown above Table 3-6B, this
of the filter. The number of inductors in any network should, transformation would have yielded a filter containing
therefore, be reduced whenever possible. three inductors rather than the two shown in Fig. 3-25B.
The object in any of these filter designs is to reduce the
A little experimentation with dual networks having unequal ter-
number of inductors in the final design. More on this later.
minations will reveal that you can quickly get yourself into
The final step in the design process is to scale the network
trouble if you are not careful. This is especially true if the load
in both impedance and frequency using Equations 3-12
and source resistance are a design criteria and cannot be changed
and 3-13. The first two calculations are done
to suit the needs of your filter. Remember, when the dual of a net-
for you.
work with unequal terminations is taken, then the terminations
must, by definition, change value as shown in Fig. 3-26. 1
C1 = 1.807
2π(60 × 106 )(300)
BA N D PA S S F I LT E R D E S I G N = 4.9 pF
The low-pass prototype circuits and response curves given in this  
1
chapter can also be used in the design of bandpass filters. This 300
1.303
is done through a simple transformation process similar to what L2 =
2π(60 × 106 )
was done in the high-pass case.
The most difficult task awaiting the designer of a bandpass filter, = 611 nH
if the design is to be derived from the low-pass prototype, is in
Ch03-H8518.tex 19/9/2007 15: 13 page 58

58 RF CIRCUIT DESIGN

6
Example 3-7.—Cont
4 2
The remaining values are:
5
C3 = 3.3 pF 2
3 5
C5 = 4.9 pF
3
L4 = 611 nH
(A) A reprsentative ladder network
1.807 2.691 1.807
1 3
L1 L3 L5
Rs
3

2
1.303 1.303 1
C2 C4 RL 5
1/4 2 1/5

(A) Normalized low-pass filter circuit (B) Its dual form

1 1/1.807 1/2.691 1/1.807

C1 C3 C5 FIG. 3-26. Duality.
Rs

1/1.303 1/1.303 1 at a frequency or bandwidth of 4 kHz (f /fc = 2), then the response
L2 L4 RL of the bandpass network would be down 30 dB at a bandwidth of
4 kHz. Thus, the normalized f /fc axis of the low-pass attenuation
curves becomes a ratio of bandwidths rather than frequencies,
(B) High-pass transformation

Rs C1 C3 C5

300  4.9 pF 3.3 pF 4.9 pF 0

3 dB BW1
L2 611 nH L4 611 nH RL 300 

30 dB BW2
(C) Frequency and impedance-scaled filter circuit

(A) Low-pass prototype response

FIG. 3-25. High-pass filter design for Example 3-7.

The final filter circuit is given in Fig. 3-25C. 0

3 dB BW1

specifying the bandpass attenuation characteristics in terms of

the low-pass response curves. A method for doing this is shown 30 dB BW2
by the curves in Fig. 3-27. As you can see, when a low-pass
design is transformed into a bandpass design, the attenuation
bandwidth ratios remain the same. This means that a low-pass (B) Bandpass response
filter with a 3-dB cutoff frequency, or a bandwidth of 2 kHz,
would transform into a bandpass filter with a 3-dB bandwidth of
2 kHz. If the response of the low-pass network were down 30 dB FIG. 3-27. Low-pass to bandpass transformation bandwidths.
Ch03-H8518.tex 19/9/2007 15: 13 page 59

Bandpass Filter Design 59

such that:
0 45 MHz
BW f 3 dB 75 MHz
= (Eq. 3-14)
BWc fc
where
40 dB 125 MHz
BW = the bandwidth at the required value of attenuation,
BW c = the 3-dB bandwidth of the bandpass filter.
f3 f1 f0 f2 f4

Often a bandpass response is not specified, as in Example 3-8.

Instead, the requirements are often given as attenuation values
at specified frequencies as shown by the curve in Fig. 3-28. In FIG. 3-28. Typical bandpass specifications.
this case, you must transform the stated requirements into infor-
mation that takes the form of Equation 3-14. As an example, 3 3 3
consider Fig. 3-28. How do we convert the data that is given into
the bandwidth ratios we need? Before we can answer that, we
have to find f3 . Use the following method. 2 2 2 2 2 2

EXAMPLE 3-8
Find the Butterworth low-pass prototype circuit which,
FIG. 3-29. Low-pass to bandpass circuit transformation.
when transformed, would satisfy the following bandpass
filter requirements:
frequency of the response curve shown in Fig. 3-28 must be
BW3dB = 2 MHz
BW40dB = 6 MHz

fo = (45)(75) MHz
Solution = 58.1 MHz

Note that we are not concerned with the center frequency

We can use Equation 3-15 again to find f3 .
of the bandpass response just yet. We are only concerned
with the relationship between the above requirements
and the low-pass response curves. Using Equation 3-14, 58.1 = f3 (125)
we have:
BW f BW40dB or,
= =
BWc fc BW3dB f3 = 27 MHz
6 MHz
= Now that f3 is known, the data of Fig. 3-28 can be put into the
2 MHz
form of Equation 3-14.
=3
Therefore, turn to the Butterworth response curves shown BW40dB 125 MHz − 27 MHz
=
in Fig. 3-9 and find a prototype value that will provide BW3dB 75 MHz − 45 MHz
40 dB of attenuation at an f /f c = 3. The curves indicate a = 3.27
5-element Butterworth filter will provide the needed
attenuation. To find a low-pass prototype curve that will satisfy these require-
ments, simply refer to any of the pertinent graphs presented in
this chapter and find a response that will provide 40 dB of atten-
The frequency response of a bandpass filter exhibits geometric uation at an f /fc of 3.27. (A fourth-order or better Butterworth
symmetry. That is, it is only symmetric when plotted on a loga- filter will do quite nicely.)
rithmic scale. The center frequency of a geometrically symmetric The actual transformation from the low-pass to the bandpass con-
filter is given by the formula: figuration is accomplished by resonating each low-pass element

f o = fa fb (Eq. 3-15) with an element of the opposite type and of the same value. All
shunt elements of the low-pass prototype circuit become parallel-
where fa and fb are any two frequencies (one above and one below resonant circuits, and all series elements become series-resonant
the passband) having equal attenuation. Therefore, the center circuits. This process is illustrated in Fig. 3-30.
Ch03-H8518.tex 19/9/2007 15: 13 page 60

60 RF CIRCUIT DESIGN

BA N D-R E J E CT I O N F I LT E R D E S I G N
0 Band-rejection filters are very similar in design approach to the
3 dB bandpass filter of the last section. Only, in this case, we want to
reject a certain group of frequencies as shown by the curves in
Fig. 3-30.

50 dB The band-reject filter lends itself well to the low-pass prototype
design approach using the same procedures as were used for the
bandpass design. First, define the bandstop requirements in terms
f1 f2 f3 f4 of the low-pass attenuation curves. This is done by using the
inverse of Equation 3-14. Thus, referring to Fig. 3-30, we have:
BWc f4 − f 1
FIG. 3-30. Typical band-rejection filter curves. =
BW f3 − f2

To complete the design, the transformed filter is then frequency- This sets the attenuation characteristic that is needed and allows
and impedance-scaled using the following formulas. For the you to read directly off the low-pass attenuation curves by substi-
parallel-resonant branches, tuting BWc /BW for fc /f on the normalized frequency axis. Once
the number of elements that are required in the low-pass proto-
Cn type circuit is determined, the low-pass network is transformed
C= (Eq. 3-16)
2πRB into a band-reject configuration as follows:
RB Each shunt element in the low-pass prototype circuit is replaced
L= (Eq. 3-17)
2πf02 Ln by a shunt series-resonant circuit, and each series-element is
replaced by a series parallel-resonant circuit.
and, for the series-resonant branches, This is shown in Fig. 3-31. Note that both elements in each of
the resonant circuits have the same normalized value.
B
C= (Eq. 3-18)
2πf02 Cn R
6
RLn
L= (Eq. 3-19) 6
2πB
where, in all cases, 3 6 9
R = the final load impedance, 3 9 3 9
B = the 3-dB bandwidth of the final design,
fo = the geometric center frequency of the final design,
L n = the normalized inductor bandpass element values,
Cn = the normalized capacitor bandpass element values.
Example 3-9 furnishes one final example of the procedure for FIG. 3-31. Low-pass to band-reject transformation.
designing a bandpass filter.
Once the prototype circuit has been transformed into its band-
reject configuration, it is then scaled in impedance and frequency
S U M MA RY O F T H E BA N D PA S S using the following formulas. For all series-resonant circuits:
F I LT E R D E S I G N P R O C E D U R E
Cn
1. Transform the bandpass requirements into an equivalent C= (Eq. 3-20)
low-pass requirement using Equation 3-14. 2πRB
RB
2. Refer to the low-pass attenuation curves provided in order L= (Eq. 3-21)
to find a response that meets the requirements of Step 1. 2πf02 Ln
3. Find the corresponding low-pass prototype and write it For all parallel-resonant circuits:
down.
B
4. Transform the low-pass network into a bandpass C= (Eq. 3-22)
configuration. 2πf02 RCn

5. Scale the bandpass configuration in both impedance and RLn

L= (Eq. 3-23)
frequency using Equations 3-16 through 3-19. 2πB
Ch03-H8518.tex 19/9/2007 15: 13 page 61

The Effects of Finite Q 61

0.817
0.500
Example 3-9
Design a bandpass filter with the following requirements:
4.431 2.216 1.000
fo = 75 MHz Passband Ripple = 1 dB
BW3dB = 7 MHz Rs = 50 ohms
BW45dB = 35 MHz RL = 100 ohms (A) Low-pass prototype circuit

Solution 0.500 0.817 0.817

Using Equation 3-14:

BW45dB 35
= 4.431 4.431 2.216 2.216 1.000
BW3dB 7
=5
Substitute this value for f /f c in the low-pass attenuation (B) Bandpass transformation
curves for the 1-dB-ripple Chebyshev response shown in
Fig. 3-18. This reveals that a 3-element filter will provide
about 50 dB of attenuation at an f /f c = 5, which is more 50  2.4 pF 1.86 mH
than adequate. The corresponding element values for this
1007 pF 4.47 nH 504 pF 8.93 nH 100 
filter can be found in Table 3-7 for an Rs /RL = 0.5 and an
n = 3. This yields the low-pass prototype circuit of
Fig. 3-32A which is transformed into the bandpass
prototype circuit of Fig. 3-32B. Finally, using (C) Final circuit with frequency and impedance scaled
Equations 3-16 through 3-19, we obtain the final circuit
that is shown in Fig. 3-32C. The calculations follow. Using
Equations 3-16 and 3-17: FIG. 3-32. Bandpass filter design for Example 3-9.
4.431
C1 =
2π(100)(7 × 106 ) T H E E F F E CT S O F F I N IT E Q
= 1007 pF Thus far in this chapter, we have assumed the inductors and
(100)(7 × 106 ) capacitors used in the designs to be lossless. Indeed, all of
L1 =
2π(75 × 106 )2 (4.431) the response curves presented in this chapter are based on that
= 4.47 nH assumption. But we know from our previous study of Chapters 1
and 2 that even though capacitors can be approximated as hav-
Using Equations 3-18 and 3-19: ing infinite Q, inductors cannot, and the effects of the finite-Q
7 × 106 inductor must be taken into account in any filter design.
C2 =
2π(75 × 106 )2 (0.817)100 The use of finite element Q in a design intended for loss-
= 2.4 pF less elements causes the following unwanted effects (refer to
(100)(0.817) Fig. 3-33):
L2 =
2π(7 × 106 )
= 1.86 µH 1. Insertion loss of the filter is increased whereas the final
stopband attenuation does not change. The relative
Similarly, attenuation between the two is decreased.
C3 = 504 pF 2. At frequencies in the vicinity of cutoff (fc ), the response
L3 = 8.93 nH becomes more rounded and usually results in an
attenuation greater than the 3 dB that was originally
intended.
3. Ripple that was designed into the passband will be
where, in all cases, reduced. If the element Q is sufficiently low, ripple will
B = the 3-dB bandwidth, be totally eliminated.
R = the final load resistance, 4. For band-reject filters, the attenuation in the stop-band
fo = the geometric center frequency, becomes finite. This, coupled with an increase in
Cn = the normalized capacitor band-reject element value, passband insertion loss, decreases the relative
Ln = the normalized inductor band-reject element value. attenuation significantly.
Ch03-H8518.tex 19/9/2007 15: 13 page 62

62 RF CIRCUIT DESIGN

Filter Type Minimum Element Q Required

Ideal Bessel 3
Insertion Loss
Butterworth 15

Low Q 0.01-dB Chebyshev 24

0.1-dB Chebyshev 39
0.5-dB Chebyshev 57
1-dB Chebyshev 75

TABLE 3-9. Filter Elemental-Q Requirements

FIG. 3-33. The effect of finite-Q elements on filter response.

the ideal response to a degree depending upon the element Q.

Regardless of the gloomy predictions outlined above, however,
It is, therefore, highly recommended that you make it a habit to
it is possible to design filters, using the approach outlined in this
use only the highest-Q components available.
chapter, that very closely resemble the ideal response of each
network. The key is to use the highest-Q inductors available for The insertion loss of the filters presented in this chapter can be
the given task. Table 3-9 outlines the recommended minimum calculated in the same manner as was used in Chapter 2. Simply
element-Q requirements for the filters presented in this chapter. replace each reactive element with resistor values corresponding
Keep in mind, however, that any time a low-Q component is to the Q of the element and, then, exercise the voltage division
used, the actual attenuation response of the network strays from rule from source to load.