Signals and Systems, 2009 1

Analog filters

Objectives of this section:

• Learn basics of low-pass, high-pass, band-pass and band-stop (band-reject)

filtering.

• Learn how filter characteristics are related to a filter circuit transfer function, its
poles and zeros.

• Learn about first order and second order active and passive filters, and using them
as building blocks to construct higher order filters.
Signals and Systems, 2009 2

Frequency response of a system

• The steady-state response of a stable LTI system to a sinusoidal input is
determined by the system’s frequency response H(jω) where H(s) is the
system transfer function.

• The steady-state sinusoidal response to the input A cos(ωt + φ) is a sinusoide of

the same frequency.

• The magnitude of the ss response is A|H(jω)|. |H(jω)| determines the

amplification factor or gain of the system at frequency ω .

• The phase of the response is φ + θ(ω); θ(ω) = ∠H(jω) determines a phase

shift between input and steady-state output.

• In general, given an input signal r(t), the output of the system is found as
Z ∞
1
y(t) = Y (jω)ejωt dt, Y (jω) = H(jω)R(jω).
2π −∞

The frequency response H(jω) determines how the system modifies the sinusoidal
content of the input signal.
Signals and Systems, 2009 3

Frequency response plots. Bode diagrams.

Frequency response can be plotted in several ways, e.g.,

• A plot of H(jω) = ReH(jω) + jImH(jω) as a curve in the complex plane id

the Nyquist plot. This kind of plots is important for investigating stability of feedback
systems.

• Plots of |H(jω)| and ∠H(jω) versus ω are called the Bode plot.
Bode Diagram

−5
Example The RLC circuit in one of the −10
System: H

Magnitude (dB)
previous examples was shown to have the −15

−20
Frequency (rad/sec): 2
Magnitude (dB): −9.4 System: H
Frequency (rad/sec): 20

6/5(s+2) Magnitude (dB): −24.6

transfer function H(s) = s2 +2s+2 . Its −25

−30

Bode plot shown on the right was obtained −35

−40

using Matlab: 0

>> s=tf(’s’);
Phase (deg)

>> H=1.2*(s+2)/(sˆ2+5*s+4); −45

System: H
Frequency (rad/sec): 2

>> bode(H) Phase (deg): −45 System: H

Frequency (rad/sec): 20
Phase (deg): −81.5

−90
−2 −1 0 1 2
10 10 10 10 10
Frequency (rad/sec)
Signals and Systems, 2009 4

Frequency axis

• logarithmic scale for ω , i.e., a horizontal distance corresponds to a ratio. The ratio
10 : 1 corresponds to one decade.

Magnitude |H(jω)|

• is expressed in dB, i.e., 20 log10 |H(jω)| (logarithmic gain);

• a vertical distance corresponds to a ratio of magnitudes expressed in dB; i.e., ratio
10 : 1 is +20dB, ratio 1 : 10 is −20dB;

• slopes(gradients) are given in units of dB/decade; i.e., how much logarithmic

magnitude changes when frequency increases/decreases by a factor of 10.

Phase ∠H(jω)

• Multiples of 360◦ do not matter;

Signals and Systems, 2009 5

Sketching approximate magnitude Bode plots

First, consider the case where all the poles and zeros of H(s) are real and distinct:
√
10(s + 1)
H(s) = .
s(s + 10)
1. Determine poles and zeros of H(s). They define the corner frequencies.

2. Choose a “reference” frequency, which is at least one decade less than the
smallest root or zero, or at least one decade greater than the largest root or zero.
In this example, let ω = 0.1:
√ √
10|j0.1 + 1| 10
20 log10 H(j0.1) = 20 log10 ≈ 20 log10 = 10 dB.
0.1|0.1j + 10| 0.1 × 10
3. Starting from this point, sketch the plot as shown in the figure, following the
following rules:
• The slope must increase by 20dB/decade ×m at a frequency ω = |z|, if the
system has zero at s = z of multiplicity m.
• The slope must decrease by 20dB/decade ×m at a frequency ω = |p|, if the
system has pole at s = p of multiplicity m.
Signals and Systems, 2009 6

15

10

0 The straight line

approximation
−5

−10

−15
True plot
−20

−25
Corner frequencies
−30

−35
0 1 2
10 10 10
Signals and Systems, 2009 7

For a transfer function that has a pair of complex conjugate poles or zeros, the
procedure is essentially the same. The corner frequencies corresponding to the
complex conjugate poles or zeros are at the resonant frequency ω0 of the poles
(zeros). At this corner frequency the slop changes by ±40 dB/decade (increases at
the zeros, and decreases at the poles).

ω02
For example, for an underdamped system H(s) = 2 ω0 2 , ω 0 > 0,
s + Q s + ω0
Q > 0.5,
20

10

ω02
H(jω) = 0

(jω)2 + j ωQ0 ω ω02

20*log10 |G(jω)|, (dB)

+ −10

1
=  2  
−20

ω ω
1− ω0 +j ω0 Q
−30

−40
−1 0 1
10 10 10
Frequency

20 log10 |H(jω)| ≈ 0 if ω ≪ ω0 and 20 log10 |H(jω)| ≈ straight line with a slop

of −40 dB/decade if ω ≫ ω0 .
Signals and Systems, 2009 8

Rules for sketching asymptotic magnitude Bode plots

1. Identify all corner frequencies:

2. Slope of asymptotic logarithmic gain graph increases/decreases at corner

frequencies. It increases at “numerator” corner frequencies, and decreases at
“denominator” corner frequencies.

3. The increase/decrease is ±20 dB/decade × multiplicity of the corner frequency.

Signals and Systems, 2009 9

Homework exercise 1

Sketch an asymptotic magnitude Bode plot of the transfer function

12500(1 + 0.1s)
G(s) =
s(1 + 0.5s)(s2 + 30s + 2500)
s
5(1 + 10 )
= s s 2 s


s(1 + 2 )( 50 ) + 0.6 50 + 1

Corner frequencies (from the smallest to the largest):

pole at s = 0, corner frequency at 0 rad/sec, slope −20 dB/decade

pole at s = −2, corner frequency at 2 rad/sec, slope changes from −20
dB/decade to −40 dB/decade
zero at s = −10, corner frequency at 10 rad/sec, slope changes from −40
dB/decade to −20 dB/decade;
pair of complex conjugate poles, corner frequency at 50 rad/sec, slope
changes from −20 dB/decade to −60 dB/decade.
Signals and Systems, 2009 10

In order to sketch the magnitude Bode plot, choose a frequency range, say ω from 0.1
rad/sec to 100 rad/sec. Let’s find a point on the low frequency asymptote. Say, for
ω = 0.1 rad/sec,
 
 5 
20 log10 |G(0.1j)| = 20 log10   = 20 log10 5 − 20 log10 0.1 ≈ 34 dB.
0.1j 
Hence, the low-frequency asymptote is a straight line with a slope of −20 dB/decade
passing through the point ω = 0.1 rad/sec, 20 log10 |G| = 34 dB. This line
continues from ω = 0.1 to the next corner frequency of ω = 2 rad/sec.

The slope of the asymptotic Bode plot will change as ω increases past each corner
frequency.

At ω = 2 rad/sec (pole) the slope decreases by −20 dB/decade and becomes −40
dB/decade. Then, at ω = 10 rad/sec (zero) the slope increases by 20 dB/decade, and
finally decreases by −40 dB/decade at ω = 50 rad/sec (pair of complex poles).
Signals and Systems, 2009 11

40

30

20
Asymptotic Bode plot
10
|G(j ω)|
0
Actual Bode plot
10

-10
20log

-20

-30

-40

-50
pole zero 2 poles
-60
-1 0 1 2
10 10 2 10 50 10

Frequency ω , rad/sec
Signals and Systems, 2009 12

10kΩ 1kΩ
Homework exercise 2
Sketch an asymptotic magnitude Bode plot of vin
the frequency response of the circuit 1µF 1µF
−
vout
+
The transfer function is

103 1 + 104 × 10−6 s 1 + s/100

G(s) = − 4 × = −0.1
10 1 + 103 × 10−6 s 1 + s/1000

Corner frequencies 100 rad/sec (zero) and 1000 rad/sec (pole). Choose the frequency
range from 10 rad/sec to 10000 rad/sec.

At ω ≪ 100, dc gain is 20 log10 |G(jω)| ≈ −20 dB, slope is 0 dB/decade. Hence,

at ω ≪ 100 the magnitude Bode plot is approximated by a horizontal straight line
passing through the point ω = 10 rad/sec, −20 dB.

At ω = 100 rad/sec (zero), the slope increases by 20 dB/decade, then at ω = 1000

rad/sec (pole) the slope decreases by 20 dB/decade.
Signals and Systems, 2009 13

10

0
20 log10 |G(jω)|, dB

−5

−10

−15

−20

−25

−30
1 2 3 4
10 10 10 10
Frequency ω, rad/sec
Signals and Systems, 2009 14

Filters
Filter is a system that shapes or modifies the frequency content of a signal or waveform
to suppress or amplify some parts of the input spectrum.

Terminology:

• Analog filters vs Digital filters

Analog filters process the actual input waveform. Digital filters require that the
input waveform is digitized (sampled) by an analog-to-digital converter.

• Active analog filters vs Passive analog filters

Passive filters include only resistors, capacitors and inductors. Such filters cannot
amplify signals, only attenuate. Active filters include resistors, capacitors,
inductors, and opamp amplifier circuits (feedback!).
Signals and Systems, 2009 15

Ideal Filters
Ideally, it is desired that the harmonics of the input spectrum with frequencies in the
filter pass band pass through the filter unmodified, and the harmonics with frequencies
with the frequencies in the stop band are suppressed completely.

|H(jω )|
• Low-pass filter: Components of frequency ω < ωc 1
pass unchanged through the filter, components of
frequency ω > ωc are blocked. The frequency ωc
is the cutoff frequency.
ωc ω

The cutoff frequency separates the pass-band from the stop-band. The output of the
filter contains only harmonics whose frequencies are in the pass-band.

|H(jω )|

• High-pass filter: Components of frequency ω > ωc 1

pass unchanged through the filter, components of
frequency ω < ωc are blocked.
ωc ω
Signals and Systems, 2009 16

|H(jω )|
1
• Band-pass filter: Only components of frequency
ωc1 < ω < ωc2 pass unchanged through the fil-
ter, all other components of the signal are blocked.
ωc ωc ω
1 2

We have two cutoff frequencies in this case, ωc1 and ωc2 .

|H(jω )|
1
• Band-stop filter: Components of frequency ωc1 <
ω < ωc2 are blocked. All other components of the
signal pass unchanged through the filter.
ωc ωc ω
1 2

As we have seen, the impulse response of an ideal filter is given by a sync function, or
is constructed using a sync function. Therefore, ideal filter are non-causal and cannot
be realized in real-time.
Signals and Systems, 2009 17

First-order filter basics

First order filters have transfer functions of the form
kωc ks
Hlp (s) = or Hhp (s) =
s + ωc s + ωc

First order low-pass filters have transfer function

|H lp(j ω)|
Hlp (s), offer little attenuation to the low frequency
content of the input while attenuating (blocking) high-
frequency content. Ideal lowpass filter

The cutoff frequency ωcutoff is a frequency at which:

ω (rad/s, log scale)

1
|H(jωcutoff )| = √ · max |H(jω)| ≈ 0.707 · max |H(jω)|.
2 ω ω

In circuit theory terms, in the pass-band, the average power delivered to a load is at
least 50% of Pmax .
Signals and Systems, 2009 18

20 log10 k

magnitude 20 log10 |H(jω)|, dB

0
1st order LP filter: 3dB=20 log10 0.707

kωc
Hlp (s) = , −5

s + ωc
kωc −10
|Hlp (jω)| = p ,
2
ω + ωc 2

k = |H(0)| is the DC gain. −15

cutoff frequency = corner frequency ω

c

−20
0 1
10 10
frequency, rad/sec (log scale)

To find the cutoff frequency, solve

kωc k
p = √ .
2 2
ω + ωc 2
This gives that the cutoff frequency equals the corner frequency ωc .
Signals and Systems, 2009 19

20 log k
1st order HP filter:

magnitude 20 log10 |H(jω)|, dB

10

ks 3dB=20 log
10
0.707
Hhp (s) = ,
s + ωc −5

kω
|Hhp (jω)| = p ,
2
ω + ωc2 −10

The parameter k characterises −15

the gain of the filter in the high cutoff frequency = corner frequency ωc

frequency band.
−20
0 1
10 10
frequency, rad/sec (log scale)

To find the cutoff frequency, solve

kω k
p = √ .
2 2
ω + ωc 2
This again gives that the cutoff frequency equals the corner frequency ωc .
Signals and Systems, 2009 20

Series RC and RL circuits as low-pass filters

Transfer function is R
1 1/(RC)
H(s) = = .
RCs + 1 s + 1/(RC) vin +
+ v
C out
1 − -
The frequency response H(jω) = .
jRCω + 1
1
The cutoff frequency ωc = RC , the DC gain k = 1 (cannot be adjusted!)

R R/L
Transfer function is H(s) = = . L
Ls + R s + R/L +
+ R
The cutoff frequency ωc = R − vin v
L , the DC gain k = 1 - out
(cannot be adjusted either!)

Read more about RL, RC low-pass filters in the text by Nilsson and Reidel.
Signals and Systems, 2009 21

Series RC and RL circuits as high-pass filters

R
Transfer function is
RCs s + v -
H(s) = = . out
RCs + 1 s + 1/(RC)
vin + C
jRCω −
The frequency response H(jω) = .
jRCω + 1

1
The cutoff frequency ωc = RC , the gain at large frequencies ≈ k = 1.

Transfer function is
vout
+ -
Ls s
H(s) = = .
L
Ls + R s + R/L
The cutoff frequency ωc = R + R
L, − vin
the gain at large frequencies ≈ k = 1.

More about the RC and RL high-pass filters in Chapter 14 of the textbook.

Signals and Systems, 2009 22

Second order analog filters

• The denominator of the transfer function is a quadratic (second order) polynomial.
• provide a better approximation to ideal filters than the first order filters.
• together with first order filters, can be used as building blocks for filters that provide
more accurate approximations to ideal filters.
1

0.9

The frequency response of second-order filters 0.8

1st order LP filter

is characterized by three design parameters: 0.7

• the gain k , 0.6

0.5

• the corner or centre frequency ω0 , and 0.4

• the quality factor Q (or the damping ratio

0.3
2nd order LP filter
0.2

ζ ). 0.1

0
0 1
10 10

Filter circuits are designed by choosing the values of prototype circuit components to
match required values of k , w0 , and Q (the coefficient matching technique).
Signals and Systems, 2009 23

Second order filter transfer functions and frequency responses

Filter Transfer function, H(s) Frequency response, H(jω)

kω02 kω02
Low-pass
s2 + ωQ0 s + ω02 ω02 − ω 2 + j ωQ0 ω
ks2 −kω 2
High-pass
s2 + ωQ0 s + ω02 ω02 − ω 2 + j ωQ0 ω
k ωQ0 s jk ωQ0 ω
Band-pass ω0
s2 + Qs + ω02 ω02 − ω 2 + j ωQ0 ω
k(s2 + ω02 ) k(ω02 − ω 2 )
Band-stop
s2 + ωQ0 s + ω02 ω02 − ω 2 + j ωQ0 ω

1 1
The quality factor Q = 2ζ and the damping ratio ζ= 2Q determine the shape of the
frequency response. For an overdamped stable system, 0 < Q < 0.5 and ζ > 1. For
an underdamped stable system, Q > 0.5 and ζ < 1.
Signals and Systems, 2009 24

Second order low-pass filter

Q=2

Gain |H(jω)|

Q=1.5

Q=1

k
Q=0.707
(second order
0.707 k Butterworth filter)

Q=0.5

ω0
Frequency, rad/sec

The cut-off frequency is determined by Q and ζ and is found by solving the equation:

2 ω04 1
|Hlp (jω)| = ω02 2
= .
(ω02 − ω)2 + 2
Q2 ω
Signals and Systems, 2009 25

Usually it is desired to have a maximally flat frequency response, that ensures a better
bandwidth without having an unwanted resonance.

The maximally flat frequency response is achieved when Q =ζ= √1 = 0.707.

2

LP filters that have maximally flat frequency response are called Butterworth filters.
The second order Butterworth LP filters have the transfer function of the form

kω02 k
√ 2
= 2 √ .
2
s + ω0 2s + ω0 s
ω0 + 2 ωs0 + 1

For 2nd order Butterworth LP filters, the cutoff frequency equals ω0 .

Signals and Systems, 2009 26

Second order high-pass filter

Q=2
Gain |H(jω)| Q=1.5

Q=1

k
0.707k

Q=0.707

Q=0.5

ω0
Frequency, ω, rad/sec

The cut-off frequency is determined by Q and ζ and is found by solving the equation:

2 ω4 1
|Hhp (jω)| = ω02 2
= .
(ω02 − ω)2 + 2
Q2 ω
Signals and Systems, 2009 27

Second order broadband band-pass and band-stop filters

One way of constructing band-pass or
band-stop filters is cascading a lowpass
vin vout
filter and a highpass filter. This results
LP HP
in a broadband filter.
The cascade (series) connection pro-
duces a broadband band-pass filter.

The transfer function of the resulting filter is the product of the transfer functions:
  
kωcL ks
H(s) = HLP (s)HHP (s) =
s + ωcL s + ωcH
k 2 ωcL s
= 2 .
s + (ωcL + ωcH )s + ωcL ωcH
• The resulting transfer function has two real poles at s = −ωcL , s = −ωcH .
Hence, the bode plot has two corner frequencies at ω = ωcL , ω = ωcH .
√
• The centre frequency of the filter is ω0 = ωcL ωcH (ie, the frequency in the
centre of the passband on the Bode plot).
Signals and Systems, 2009 28

The Bode plot of a broadband bandpass filter is the sum of the Bode plots of its LP and
HP components
0 0

−5 −5

−10
−10

−15
−15

−20
−20
−25

−25
−30

−30
−35

ωcH=1 rad/s ω =100 rad/s ωcH=1 rad/s ω =2 rad/s

cL −35 cL
−40

−45 −40
−2 −1 0 1 2 3 4 −2 −1 0 1 2 3 4
10 10 10 10 10 10 10 10 10 10 10 10 10 10

Drawbacks of the above design:

• The maximum gain reduces as ωcL − ωcH reduces.

• The bandwidth is difficult to predict when ωcL − ωcH is too small.
Signals and Systems, 2009 29

The parallel design gives a bandstop filter. LP

vin
Σ
vout
The transfer function is
HP

kωcL ks
H(s) = HLP (s) + HHP (s) = +
s + ωcL s + ωcH
ks2 + 2kωcL s + kωcL ωcH
= 2 .
s + (ωcL + ωcH )s + ωcL ωcH
• The filter has two real poles at s = −ωcL , s = −ωcH , and two corner
frequencies are at ω = ωcL , ω = ωcH .
√
• The centre frequency of the filter is ω0 = ωcL ωcH .
Signals and Systems, 2009 30

5 5

max gain = 0 db max gain = 0 db

0 0

−5 −5

−10 −10

−15 −15

−20 −20

−25
ω =100 rad/s −25
ω =2 rad/s
ω =1 rad/s cH ω =1 rad/s cH
cL cL
−30 −30

−35 −35
−2 −1 0 1 2 3 4 −2 −1 0 1 2 3 4
10 10 10 10 10 10 10 10 10 10 10 10 10 10

Same drawbacks:

• The gain in the stop band increases as ωcL − ωcH reduces, and is more difficult
to predict.
Signals and Systems, 2009 31

Second order narrowband band-pass and band-stop filters

The cascade and parallel designs have a limitation that only a broadband filters will
result from those designs. This limitation is due to the fact that the transfer functions
have real poles, which only possible with the quality factor Q ≤ 1/2.

Using transfer functions with complex conjugate poles gives better control over the
bandwidth through higher quality factors Q.

The following transfer functions realise narrow-band filters:

k ωQ0 s
Band-pass filter ω0 2 ;
s2 + Qs + ω0

k(s2 + ω02 )
Band-stop filter .
s2 + ωQ0 s + ω02
Signals and Systems, 2009 32

Second order narrow-band band-pass filter

k
Q=0.5
• The maximum gain is at
ω = ω0 :
0.707k max |H(jω)| = k,
ω
Q=0.707
• The cutoff frequencies
Gain |H(jω)|

ωc1 , ωc2 are found

Q=1 from the equation
Q=1.5
k
|H(jω)| = √
2
Q=2
• The bandwidth:
ω1 ω0 ω2 BW = ωc2 − ωc1
Frequency, ω, rad/sec
Bandwidth

r 2
ω0 ω0 ω0 √
ωc1,c2 = ± 2Q + 2Q + ω02 , BW = Q, ω0 = ωc1 ωc2
Signals and Systems, 2009 33

Second order narrow-band band-stop filter

k Q=2 • The minimum gain is at

Q=1.5
ω = ω0 :
Q=0.5
Q=0.707 Q=1

min |H(jω)| = 0,
ω
0.707k
• The cutoff frequencies
ωc1 , ωc2 are found
Gain |H(jω)|

Bandwidth
from the equation

k
|H(jω)| = √
2
• The width of the stop
band:
ω1 ω0 ω2
Frequency, ω, rad/sec SB = ωc2 − ωc1
r 2
ω0 ω0 ω0 √
ωc1,c2 = ± 2Q + 2Q + ω02 , SB = Q, ω0 = ωc1 ωc2 .
Signals and Systems, 2009 34

Active filter circuits

Active filter circuits are filter circuits that employ opamps. Advantages of using active
filters are:

• Bandpass and bandstop filters are implemented without using inductors. Such
circuits are smaller in size and do not induce electromagnetic noise.

• Active filter circuits provide control over the filter gain that is not available in the
case of passive filter circuits.

• Connecting a resistive load at the output of the passive filter modifies its cutoff
frequency and pass-band due to loading. This is not the case with active filters due
to the properties of opamps.
Signals and Systems, 2009 35

Example: RC highpass passiver filter

C
The transfer function is
Rf Cs s
H(s) = = . +
Rf Cs + 1 s + 1/(Rf C) vin +
− Rf vout
The cutoff frequency is ωc = R 1C . -
f

Connect a resistive load RL . The transfer function be-

comes
s C
H(s) =  .
Rf RL
s + 1/ Rf +RL C +
vin +
The cutoff frequency changes to − Rf vout RL
-
Rf + RL
ωc = .
Rf RL C
The cutoff frequency increases as load increases, and the bandwidth of the filter
reduces.
Signals and Systems, 2009 36

First order active filters

Filter Circuit System TF Design equations

C cutoff frequency
1
vin R
1 R2 ωc = R2 C ,
− maximum gain
vout R2 /R1 R2
Low-pass, + − |k| = R1
kωc 1 + R2 Cs
s + ωc
cutoff frequency
1
ωc = R1 C
vin R maximum gain
1 C R2 R2 Cs R2
High-pass, − − |k| = R1
ks
vout 1 + R1 Cs
+

s + ωc
Signals and Systems, 2009 37

Example
40

Design an active filter circuit that has 34

35

the asymptotic magnitude Bode plot 30

shown on the right.

20log10|G(jω)|
20 dB/decade
25

• The slope at low frequencies +20 20

dB/decade, so G(jω) must have a 15

jω factor in its numerator.

10
1 2 3 4
10 10 10 10
500
Frequency ω, rad/sec

• At ω = 500 rad/sec, the gradient changes from 20 dB/decade to 0db/Dec, hence

G(jω) must have a real pole at s = −500 (choose a stable circuit). Thus,
jω
G(jω) = ±k ,
jω + 500

• At large ω , 20 log10 |G(jω)| ≈ 20 log10 k = 34 ⇒ k = 1034/20 ≈ 50.

s
Thus, G(s) must be equal to G(s) = ±50 .
s + 500
Signals and Systems, 2009 38

There are many circuits that match this

transfer function. One such circuit is vin R R2
shown on the right: 1 C
−
R2 R2 Cs vout
G(s) = − =− . +
R1 + 1/(Cs) 1 + R1 Cs

Find R1 , R2 and C to match the required G(s). Using the table, we obtain the
relationship between components

1 R2
ωc = = 500 rad/s, k= = 50
R1 C R1
Let C = 1 µF, then R1 = 2 kΩ, R2 = 100 kΩ.
Signals and Systems, 2009 39

Second order active broadband filters

It was observed that such filters can be obtained by cascading an LP filter and an HP
filter. The cascade series connection leads to a broadband pass-band filter:
CL
vin R R2L
1L
CH R2H
−
RF
−
+ A R1H −
+ B R1i
+ vout

Low-pass stage High-pass stage Inverting amplifier

(to adjust gain)

The transfer function of the resulting filter is the product of the transfer functions:
  
−s RF
−ωcL
H(s) = HL (s)HP (s)HI = −
s + ωcH s + ωcL
R1i
−kωcL s
= 2 .
s + (ωcL + ωcH )s + ωcL ωcH
√
Real poles at s = −ωcL , s = −ωcH , centre frequency is at ω0 = ωcL ωcH .
Signals and Systems, 2009 40

CL
The parallel design gives a
Low-pass stage
bandstop filter. R1L R2L
− R1i
+ A

RF
vin
R1i −
B
CH R2H + vout
−
R1H
+ Summing amplifier
The transfer function is
High-pass stage

  
−ωcL −s RF
H(s) = (HL (s) + HP (s))HI = + −
s + ωcL s + ωcH R1i
RF s2 + 2ωcL s + ωcL ωcH
= · 2 .
R1i s + (ωcL + ωcH )s + ωcL ωcH
√
Real poles at s = −ωcL , s = −ωcH , the centre frequency is at ω0 = ωcL ωcH .
Signals and Systems, 2009 41

Second order underdamped active filters. Sallen-Key design.

The cascade series and parallel designs have a limitation that only a broadband filters
will result from those designs. This limitation is due to the fact that the transfer
functions have real poles when the quality factor Q ≤ 1/2 (damping ratio ζ > 1).

Using transfer functions with complex conjugate poles gives better control over the
bandwidth through higher quality factors Q.

2C
As an example, consider a common
positive-feedback second-order active
bandstop filter known as the Sallen-Key R R v2
+
circuit. The circuit transfer function is v3
vin
−
k(s2 + ω02 ) v1 vout
H(s) = 2 ω0 2 ;
s + Q s + ω0 C C
(A-1)R0
1 1 R/2
R0
ω0 = , Q= , k = A.
RC 4 − 2A

A is the gain of the noninverting amplifier; i.e., A > 1. Also, for stability we need
A < 2.
Signals and Systems, 2009 42

• Since the gain of the noniverting amplifier A > 1, then

1 1
Q= > , and Q → ∞ as A → 2.
4 − 2A 2
Hence, controlling the bandwidth and quality factor can be done through changing
the gain A of the noninverting amplifier.
10

A=1.9
• Changes in A do not affect 5

the center frequency ω0 . 0

A=1
• The gain in the pass band −5

is equal to A and is con- −10

trolled by the opamp feed- −15

back resistor.
−20

• Note that |H(jω0 )| = 0. −25

Hence, for any A, the filter
−30
provides very good attenu- ω0=106 rad/s
ation in the vicinity of ω0 . −35

−40
5 6 7
10 10 10
Signals and Systems, 2009 43

Higher order active opamp filters. Butterworth filters.

• By cascading several low order active stages, a sharper transition between the
passband and stopband can be obtained.

• Direct cascading of several filters usually undesirable, since cascading affects

adversely the filter bandwidth. To avoid this, special families of transfer functions
are developed, such as the Butterworth filters, Chebyshev filters, elliptic filters, etc.
Magnitude Bode plot of 6th order LP elliptic and Butterworth filters
ω =300 rad/sec
cutoff

elliptic filter
0 Butterworth filter

−10

• Each of these filter transfer func-

tions attempt to achieve a trade-
Magnitude, dB
−20

off between the ripple in the pass −30

(stop) band and the sharpness of

the roll-off in the transition-band. −40

−50

−60
0 1 2 3 4
10 10 10 10 10
Frequency ω, (rad/sec)
Signals and Systems, 2009 44

Butterworth LP transfer functions are designed to have magnitude frequency

responses that are flat in the pass band, but provide a relatively slow roll-off.

1
|Hn (jω)| = p , 1

1 + (ω/ωc )2n
0.8

n is the order of the filter, ωc is the cutoff frequency,

and Hn (s) is the Butterworth LP transfer function:

Magnitude
0.6

±1
0.4
RC filter

Hn (s) = ; 20th order

Butterworth filter

Dn (s/ωc ) 0.2
4th order
Butterworth filter

0
−1 0 1

the polynomials Dn (s) are given in the table

10 10 10
Frequency, rad/sec

Order, n Denominator, Dn (s)

1 s√+1
2 s2 + 2s + 1
3 (s + 1)(s2 + s + 1)
4 (s2 + 0.765s + 1)(s2 + 1.848s + 1)
High order Butterworth filter circuits are constructed by cascading 2nd order Sallen-Key
low-pass filters and a 1st order low-pass filter.
Signals and Systems, 2009 45

Example Design the third order Butterworth filter with a cutoff frequency of 500
rad/sec.

Solution From the table, Dn (s) = (s + 1)(s2 + s + 1), and since the desired cutoff
frequency is 500 rad/s, the desired filter must have a transfer function

1 k2 5002 k1 500
H(s) = s s 2 s = 2 · .
( 500 + 1)(( 500 ) + 500 + 1) (s + 500s + 5002 ) (s + 500)
k1 , k2 are constants that must be chosen so that k1 k2 = 1.
Signals and Systems, 2009 46

Design the 2nd order stage

C
Use the Sallen-Key low-pass filter cir-
cuit. The transfer function is
+
k2 ω02
2 ω0 2 , vin R R
−
s + Q s + ω0 vout
C
1 1
ω0 = , k2 = A, Q = ,
RC 3−A RB
RA
A = 1 + RB /RA is the gain of the
noninvering amplifier.

k2 5002
To obtain H2 (s) = s2 +500s+5002 , we have to set

1 1
ω0 = = 500, Q = = 1, k2 = A = 2.
RC 3−A
R
Choose RA = RB = 20 kΩ, then the condition A = 1 + RA = 2 is satisfied. Also,
B
choose R = 20 kΩ, C = 0.1 µF to achieve ω0 = 500 rad/s.
Signals and Systems, 2009 47

Design the 1st order stage

Use the low-pass filter from the table on p.36.
The required transfer function is
C
−k1 500 vin R
H1 (s) = . 1 R2
s + 500
−
vout
Since k2 has been chosen to be k2 = 2, then
+
the design equations are

R2 1 1
k1 = = = 0.5, ωc = = 500,
R1 k2 R2 C
Components R1 = 40 kΩ, R2 = 20 kΩ, C = 0.1 µF meet these requirements.
Signals and Systems, 2009 48

Assemble the filter

The transfer function of the designed filter is

2 × 5002 −250
H(s) = 2 · .
s + 500s + 5002 s + 500

0.1uF

0.1uF

+
vin 20k 20k
−
40k 20k
0.1uF
−
vout
20k +
20k