Weeks 9-13 Overview

Active filter theory:

ECTE212 Filter characteristics and design parameters; and
Electronics Filter transfer function;
Filter types, primarily:
Week 3 Butterworth; and
Dr. Daniel R. Franklin Chebychev Type I
Filter transformation; and
Filter realisation.

ECTE212ElectronicsWeek 3 – p. 1 ECTE212ElectronicsWeek 3 – p. 2

Notes: Notes: The second half of the course is somewhat different to the first half. We
will look at some practical and useful circuits from a system perspective. You will
learn some design skills as well as gaining some deeper insight into the material
you covered in the first half of the course.

Refer to Chapter 12 of the 5th Edition of Sedra and Smith.

Weeks 9-13 Overview Analog Filters
Oscillators: A filter is
Oscillation: the Barkhausen criterion; and A system which is deliberately designed to
Various oscillator realisations change the magnitude and/or phase of a signal in
a known, predictable way at different frequencies;
Negative feedback and circuit stability:
Generally classified as low-pass, high-pass,
Requirements for stability;
band-pass, band-stop/reject and all-pass; and
Stabilisation of unstable amplifier circuits
Characterised by a set of descriptive parameters,
Power amplifiers: determined by the problem you are trying to solve.
Class A, B, AB, C and D amplifiers. An analog filter is a continuous-time system which
performs this function on a continuous function of
time (e.g. voltage, current).

ECTE212ElectronicsWeek 3 – p. 3 ECTE212ElectronicsWeek 3 – p. 4

Notes: As the focus of this half of the course is on circuit design, you will see some Notes: When you think about it, filters are everywhere. If you listen to someone
circuits used in real systems. Much of the content is oriented around frequency- playing the guitar, the ‘digital’ information, i.e. the notation, is converted into an
domain analysis of circuits - you may see some relationship between the material analog signal by the guitarist (in a non-linear way - some sort of product term is
covered in this course and that covered in your analog control subject. In fact, usually produced since there are harmonics present in the output signal). The in-
much of electronics can be considered as an application of control theory (both strument itself is a filter; like the filters we are looking at, it has poles (resonances)
analog and digital!). and zeros (but a well-designed instrument should make these imperceptible). The
air between the guitar and your ear is effectively a low-pass filter - higher frequen-
cies are attenuated more as distance increases (consider someone driving past
you playing doof-doof music - you only hear the low-frequency components when
they are relatively distant). Your ear is a band-pass filter (you can’t hear much be-
low 20 Hz, and most people are insensitive to frequencies above about 20 kHz).
Finally, the auditory centres of your brain itself constitute a very complex non-linear
filter, which turns the signal into something you perceive as music. Sorry, I just took
all the magic out of it ;-)
Why? Key Filter Parameters
There are many situations in which a filter is useful: Any LTI system is a filter; however, useful filters
Selecting one signal from many (e.g. a radio generally can be specified by a set of descriptive
tuner); parameters:
Compensating for unwanted filtering in the Upper and lower cut-off frequencies (usually 3 dB
environment (e.g. an equaliser); below passband gain);
Enhancing part of a signal (e.g. bass-boost); Rate of rolloff (sharpness of cut-off slope);
Digital-to-Analog and Analog-to-Digital conversion Passband and stopband ripple;
(anti-aliasing); Stopband attenuation, passband gain; and
Eliminating a noise source at a certain frequency Quality factor, damping factor;
(e.g. 50 Hz mains noise)

ECTE212ElectronicsWeek 3 – p. 5 ECTE212ElectronicsWeek 3 – p. 6

Notes: Filters are one of the fundamental parts of your set of basic electronic Notes: Depending on what you’re trying to do with your filter, some of these pa-
building-blocks. rameters may not be important. The design techniques we are going to explore
are fairly easy to automate - you can easily find Matlab code or even spreadsheet
templates to produce a required set of component values for a given set of design
criteria. Sadly you can’t do this in the exam!

Not all parameters apply to all filters; e.g. Q-factor only makes sense for band-stop
or band-pass filters, and some filters do not have either passband or stopband
ripple.
Filter Parameters LTI Analog Filters
Passband Ripple = 2 dB
Most useful filters can be modelled as a linear
0 time-invariant system:
Stopband Attenuation = 24 dB
−5 Passband = 1.5 MHz
−10 p(s) a0 + a1 s + a2 s2 + · · · + aM sM
Transition band = 0.23 MHz H(s) = =
−15 q(s) b0 + b1 s + b2 s2 + · · · + sN −1
|H(w)| (dB)

−20
Stopband
In reality, all practical filters are non-linear and
−25
Spectral
time-varying (due to temperature, pressure,
−30 Nulls corrosion, aging, etc); however, the good news is
that we can usually ignore this...
−35

−40
0 0.5 1 1.5 2 2.5 3
Frequency (MHz)
ECTE212ElectronicsWeek 3 – p. 7 ECTE212ElectronicsWeek 3 – p. 8

Notes: Notes: Physical circuits also behave differently at high frequencies - once you
get over a few MHz you need to be very careful about the circuit layout. The
skin effect, in which high-frequency current through a conductor is non-uniformly
distributed, means that resistance, capacitance and inductance themselves (not
just reactance) vary with frequency. Designing radio frequency (RF) circuits with
these considerations is an art as well as a science and requires a great deal of
experience (plus computer simulation and experimentation).
Poles, Zeros Analog Filter Topology
Solutions to p(s) = 0 are called zeros, while Several standard filter topologies have been
solutions to q(s) = 0 are called poles. developed to simplify parameter-based filter design,
The filter is stable if the real part of all poles is by factorising p(s) and q(s) into a product of linear
negative (i.e. poles are on the left-hand side of the
and quadratic terms. The main ones are
s-plane) and the order of p(s) is less than the order Butterworth
of q(s). If this is not the case, the filter may be Chebychev type I and II
unstable and act as an oscillator; Elliptic
The filter has spectral nulls at frequencies Bessel
corresponding to the zeros. Given a sinusoidal input
at this frequency, you will have zero output.
Generally in filter design, the zeros are purely
imaginary.
ECTE212ElectronicsWeek 3 – p. 9 ECTE212ElectronicsWeek 3 – p. 10

Notes: When you are designing a circuit, you may need to consider component Notes: Each design has advantages and disadvantages. You need to choose an
tolerance. If your poles are very close to the imaginary axis, and your components appropriate filter based on consideration of its ability to satisfy your design criteria
have 10% tolerance, you may find that when the circuit is constructed, sometimes with minimum complexity, maximum stability and maximum ease of implementa-
it will operate correctly, sometimes it will become unstable. You may also find tion. It is never an easy decision, nor is there always one design which is obviously
some components (semiconductors in particular) are temperature-sensitive - this superior.
can also push your circuit from being stable to being unstable.

It is also important to realise that all circuits are filters - including amplifiers. This
means that if you want to ensure that your amplifier is stable, you will need to be
very careful about where you position the poles - and if you have a pole which is in
the “danger zone” near the imaginary axis, you may need to cancel it with a zero
to improve stability. You will learn more about this in coming weeks.

Another interesting point - if you place all of your filter zeros at infinity, you will not
have any spectral nulls in the stop-band.

In realisable analog filters, all complex poles and zeros are complex conjugates.
General Observations Butterworth Filters
Sharper cut-off requires a higher-order q(s); A nice, simple (!) all-pole filter design to start with:
however... No pass-band or stop-band ripple;
Higher-order polynomials are more susceptible to Magnitude of the transfer function is of the form
stability problems!
1
It is not necessary to set stop-band attenuation to |H(jω)| = r  2N (1)
below the noise floor; 1 + ǫ2 ω
ωp
Real analog filters must be constructed from
available components: where N is the filter order, ωp is the passband
You cannot buy a 197.513Ω resistor, and even if edge frequency and ǫ is a function of the required
you could, it is unlikely to always maintain that attenuation at ω = ωp .
value...
Capacitors tend to be more ‘ideal’ than inductors.
ECTE212ElectronicsWeek 3 – p. 11 ECTE212ElectronicsWeek 3 – p. 12

Notes: Although we are focusing here on low-pass filters, it is straightforward to Notes: Various major IC vendors sell integrated Butterworth filters on a single chip,
transform a given low-pass filter design into high-pass, band-stop or band-pass typically 6th-order or a bit higher. If you find one that is suitable for your design, it
form. will probably greatly reduce your component count and manufacturing cost.

Implementation with practical components requires some thought; if you design a See Sedra and Smith, 5th ed., pp 1091-1095 or 4th ed., pp 892-897 for more
system requiring capacitors smaller than a few tens of picofarads, then the circuit details and examples.
probably won’t work as the inter-wire capacitances will become significant. Simi-
larly very small and very large inductances and resistances are best avoided. A
good idea is to check component availability (and prices!) in an electronics catalog
before designing your filter so you know what is and is not available.
Butterworth Characteristics Butterworth Characteristics
From the magnitude of the transfer function given in Inverting this equation,
1, the gain at DC (i.e., when H(ω)) is exactly one; q
Apass
ǫ = 10 10 − 1 (2)
Also, when ω = ωp , the attenuation relative to the DC
gain (i.e. |H(0)| = 1) is Attenuation at the edge of the stop-band ωs :
"  2N #
ωs
A(ωs ) = −|H(ωs )| dB = 10 log10 1 + ǫ2
Apass = −|H(ωp )| (dB) = 10 log10 (1 + ǫ2 ) (dB) ωp
(3)
Thus, for a given ωp , Apass is the amount of variation
Given a desired minimum stop-band attenuation
in the passband, expressed in dB.
A(ωs ), this equation can
n be solved find N :
too
 A(ωs )

log ǫ12 10 10 − 1
N =  ωs

 (4)
 2 log ωp 
ECTE212ElectronicsWeek 3 – p. 13 ECTE212ElectronicsWeek 3 – p. 14

Notes: Apass is the attenuation at the edge of the passband. Notes: You can use any desired base for the logarithm. N is the minimum fil-
ter order required to satisfy the design requirements (i.e. that at ωp we have
an attenuation of exactly Apass , and at ωs we have an attenuation of at least
Astop = A(ωs ) (since N must be an integer in a physical circuit - thus N must
always be chosen to exceed the minimum design specifications. The ⌈x⌉ operator
means the ceiling or next-highest integer to the real value x.
Butterworth Filter Poles Butterworth Design Process
P1 jω Determine ǫ from Equation 2;
P2
π
π Poles are located Find the lowest integer N which satisfies
2N

P3 π
N symmetrically on the A(ωs ) ≥ Astop by using Equation 4
N
ω0
left-hand side of a circle
of radius ω0 , separated Determine pole locations from Figure 15 and
σ
by π/N radians: Equation 5;
  N1 The transfer function is given by
1
ω0 = ωp (5)
Kω0N
ǫ H(s) = (6)
(s − p1 )(s − p2 )(s − p3 )...(s − pN )
where K is the desired DC gain (if you want a
non-unity DC gain): |H(0)| = K
ECTE212ElectronicsWeek 3 – p. 15 ECTE212ElectronicsWeek 3 – p. 16

Notes: Rather than memorise the formula, you may be better off remembering the Notes:
graphical arrangement of poles.
S&S Example 11.1 Example: Butterworth LPF
1
Find the Butterworth transfer function that meets the ω0 = 2π × 10 × 103 (1/0.50885) 9 = 6.773 × 104 rad/s.
following low-pass filter specifications: fp = 10 kHz, There are 9 poles, π/9 radians apart from each
Apass = 1 dB, fs = 15 kHz, DC gain = 1, Astop = 25 dB. other, with the first at 5π/9 radians.
Combining conjugate factors in the denominator
p by 1determining ǫ. Equation 2 gives
Begin
(except for the 5th pole which is purely real), we
ǫ = 10 10 − 1 = 0.50885. arrive at the transfer function:
Find N using Equation 4. For our parameters, this 3.0 × 1043
H(s) =
yields N = ⌈8.76⌉ = 9. (s + 6.773 × 104 )(s2 + 1.2729 × 105 s + 4.587 × 109 )
1 1
As a self check, calculate A(ωs ) for N = 9, and we × 2 ×
(s + 1.0377 × 105 s + 4.587 × 109 ) (s2 + 6.7730 × 104 s + 4.587 × 109 )
find that the actual stop-band attenuation is 25.8 dB. 1
× 2
This is better than 25 dB, so N = 9 is correct. (s + 2.3516 × 104 s + 4.587 × 109 )

ECTE212ElectronicsWeek 3 – p. 17 ECTE212ElectronicsWeek 3 – p. 18

Notes: Notes:
Example: Butterworth LPF Chebyshev Filters
0 Another straightforward (!) filter design;
|H(jf)| (dB)

−50
−100
Unlike the Butterworth filter, this one has:
−150 Monotonic stopband, non-zero passband ripple
−200 (Type I - all-pole);
1000 10000 100000 Monotonic passband, non-zero stopband ripple
Frequency (Hz)
0
(Type II - pole/zero).
Phase (H(jf)) (deg)

−100
−200 Chebyshev filters require about half the order to
−300
−400 achieve the same specifications;
−500
−600 Design process is quite similar to Butterworth;
−700
1000 10000 100000
Frequency (Hz)

ECTE212ElectronicsWeek 3 – p. 19 ECTE212ElectronicsWeek 3 – p. 20

Notes: Notes: Refer to Sedra and Smith, Microelectronic Circuits 5th Ed., pp 1095-1098.
Magnitude Response (Type I) Chebyshev (I) Characteristics
Passband magnitude response: Attenuation at the edge of the passband (and
maximum gain variation over the passband):
1
|H(jω)| = r    (7)
2 2
1 + ǫ cos N cos −1 ω Apass = 10 log10 (1 + ǫ2 ) (9)
ωp
Re-arranging, we have (exactly as for Butterworth):
Stopband magnitude response: q
Apass
1 ǫ = 10 10 − 1 (10)
|H(jω)| = r    (8)
2 2 −1 ω Attenuation at the edge of the stop-band ωs :
1 + ǫ cosh N cosh ωp
   
2 2 −1 ωs
A = 10 log10 1 + ǫ cosh N cosh (11)
ωp
ECTE212ElectronicsWeek 3 – p. 21 ECTE212ElectronicsWeek 3 – p. 22

Notes: As for Butterworth filters, N is the filter order, ωp is the passband edge, ωs Notes:
is the stopband edge and ǫ determines maximum passband transmission.

The hyperbolic trigonometry functions are defined as

ex − e−x
sinh x =
2
and

ex + e−x
cosh x =
2
Most calculators have built-in hyperbolic trigonometric functions.
Chebyshev (I) Characteristics Chebyshev (I) Characteristics
As for Butterworth filters, Equation 11 may be solved The DC gain of a Chebyshev filter depends on the
to determine the minimum integer N required to filter order:
satisfy A(ωs ) ≥ Astop : Odd-order LP filters have unity DC gain
 q ! Even order LP filters have maximum pass-band
A(ωs )
−1 10 10−1 deviation at DC
 cosh ǫ 
  Number of passband maxima and minima equal
N =
   
 (12)
 cosh−1 ωs  filter order
 ω p 
 
For the Butterworth example we have already seen,
this yields ⌈4.4109⌉ = 5.
Once you have N and ǫ you can determine the pole
locations. ECTE212ElectronicsWeek 3 – p. 23 ECTE212ElectronicsWeek 3 – p. 24

Notes: Notes:
Chebyshev (I) Filter Poles Chebyshev (I) Filter Poles
    jω
2k − 1 π 1 −1 1 1000
Pk = −ωp sin sinh sinh 6th Order
N 2 N ǫ 10th Order
    15th Order
2k − 1 π 1 1 (13)
+ jωp cos cosh sinh−1 500

N 2 N ǫ
k = 1, 2, . . . N σ
0

−500

−1000 1 kHz passband

0.1 dB passband ripple
ECTE212ElectronicsWeek 3 – p. 25
−1000 −500 0 500 1000 ECTE212ElectronicsWeek 3 – p. 26

Notes: Notes: Poles are now not on a circular segment - rather it is elliptical, with the
major and minor axes as given by:

β2 + 1 β2 − 1
r1 = ωp , r2 = ωp
2β 2β

where

√ !2N
1+ 1 + ǫ2
β=
ǫ

Geometrically the location may be determined by drawing the poles of equivalent

Butterworth filters with radius r1 , r2 as determined above, the imaginary part is
the same as the corresponding pole on the r1 circle, and the real part is the same
as the pole on the r2 circle. Or you can just use the formula.
Chebyshev (I) Design Process S&S (5th Ed.) Example 12.2
Determine ǫ from Equation 10; Find the Chebyshev (I) transfer function that meets
Find the lowest integer N which satisfies the following low-pass filter specifications (exactly as
A(ωs ) ≥ Astop ; for the Butterworth filter in the previous lecture):

Determine pole locations from Equation 13; thus fp = 10 kHz;

Apass = 1 dB;
The transfer function (with |H(0)| = K) is given by
fs = 15 kHz;
DC gain = 1; and
KωpN
H(s) = N −1 (14) Astop = 25 dB
ǫ2 (s − p1 )(s − p2 )(s − p3 )...(s − pN )
In this case, Apass is the maximum passband ripple.

ECTE212ElectronicsWeek 3 – p. 27 ECTE212ElectronicsWeek 3 – p. 28

Notes: As you can see, the design process is virtually identical to that of the But- Notes:
terworth filter - but with more complex equations.
Example: Chebyshev LPF Example: Chebyshev LPF
Solution: From Equation 13, substituting k = 1, 2 . . . 5, we have

Begin by determining
p ǫ. As before, Equation 10
1 p1 , p5 = ωp (−0.0895 ± j0.9901) = −5.6235 × 103 ± j6.2210 × 104
gives ǫ = 10 − 1 = 0.50885.
10
p2 , p4 = ωp (−0.2342 ± j0.6119) = −1.4715 × 104 ± j3.8447 × 104
Solve Equation 12. As mentioned earlier, for the
p3 = ωp (−0.2895) = −1.8190 × 104
given parameters, this yields N = 5. Substituting
N = 5 yields A(ωs ) = 29.9 dB. This is better than 25
dB, so our calculation is correct.

ECTE212ElectronicsWeek 3 – p. 29 ECTE212ElectronicsWeek 3 – p. 30

Notes: Notes:
Example: Chebyshev LPF Example: Chebyshev LPF
Combining conjugate factors in the denominator (except 0
−20
for the 3rd pole which is purely real), we arrive at the

|H(jf)| (dB)
−40
−60
transfer function: −80
−100
−120
1.2028 × 1023 −140
1000 10000 100000
H(s) =
(s + 1.8190 × 104 )(s2 + 2.9430 × 104 s + 1.6948 × 109 ) Frequency (Hz)
1 0

Phase (H(jf)) (deg)

× 2 −50
s + 1.1241 × 104 s + 3.9017 × 109 −100
−150
−200
−250
−300
−350
−400
−450
1000 10000 100000
Frequency (Hz)

ECTE212ElectronicsWeek 3 – p. 31 ECTE212ElectronicsWeek 3 – p. 32

Notes: Notes:
Other Filter Types Filter Transformation
The other main class of analog filter is Elliptic So far we have seen how to develop transfer
(Cauer) filters; functions for two types of all-pole low-pass filters;
These have equal amounts of ripple in passband What about other filter types?
and stopband - they also require the smallest order Fortunately, there are several approaches to get
of any filter type in order to meet a given from LP to LP, HP, BP, BR filters with arbitrary cut-off
specification. frequencies;
However, they are based on elliptic integrals which
must be evaluated numerically - phase response is
relatively nonlinear;
If linear phase is required, use a Bessel filter.

ECTE212ElectronicsWeek 3 – p. 33 ECTE212ElectronicsWeek 3 – p. 34

Notes: Not all filters are equally suitable for all applications. Although the Bessel Notes:
filter exhibits poor transition performance (requiring a high order than even a But-
terworth filter) to achieve a given cut-off, for some applications (audio in particular)
it is highly desirable to have a linear pass-band response (what this means is that in
the passband, the output is simply a perfect time-delayed version of the input). For
many applications, however, the Elliptic filter is considered superior. The design
process for such a filter is basically the same as above, but with more complicated
equations (requiring numerical evaluation of functions).
LP-to-? Transformation Example: LP-BP Transformation
Transform the single-pole butterworth low-pass filter
LP(ωp ) to ?(ωp′ ) S-Substitution New Band Edges
ωp
given by
Low-pass s→ ωp′ s ωp′
ωp ωp′ ωp
High-pass s→ ωp′ H(s) =
s
s2 +ωl ωu
s + ωp
Band-pass s→ ωp s(ω u −ωl )
ωl , ωu
s(ωu −ωl ) into a bandpass filter with upper and lower band edge
Band-stop s→ ωp s2 +ωl ωu ωl , ωu frequencies ωu and ωl respectively.
Table 1: Lowpass to anything transformations

ECTE212ElectronicsWeek 3 – p. 35 ECTE212ElectronicsWeek 3 – p. 36

Notes: The table is taken from Proakis and Manolakis, Digital Signal Processing - Notes: Problem taken from Proakis and Manolakis, Example 8.4.1.
Principles, Algorithms and Applications, 3rd ed., Prentice Hall, 1996
Example: LP-BP Transformation Filter Realisation
Solution: Actually implementing any of the above H(s)
Using the substitution from Table 1, we have requires assembly of many building blocks;
A number of standard designs are available for
ωp realising first-order and second-order blocks;
H ′ (s) = s2 +ωu ωl
ωp s(ωu −ωl ) + ωp An arbitrary-order filter (of any variety) can be built
(ωu − ωl )s by cascading these blocks;
=
s2 + (ωu − ωl )s + ωu ωl The choice of block implementation depends on the
problem you are trying to solve.
Therefore, the resulting bandpass filter has one zero
and two complex-valued poles.

ECTE212ElectronicsWeek 3 – p. 37 ECTE212ElectronicsWeek 3 – p. 38

Notes: Notes:
Basic First-Order Blocks Second-Order Blocks
C There are a very large number of second-order filter
blocks available, including:
Passive RLC resonators
R2 R2
R1 C R1 Active RC + simulated L circuits (Antoniou)
− −

Vi + Vi +
Two-integrator loop biquadratic (KHN,
Vo Vo
Tow-Thomas)
Single-Amplfier Biquads (Bridged-T, Sallen and
Key)
1 This will be covered in detail next week.
CR1 sR
R1
2

H(s) = − 1 (LP) H(s) = − 1 (HP)

s+ R2 C s+ R1 C

ECTE212ElectronicsWeek 3 – p. 39 ECTE212ElectronicsWeek 3 – p. 40

Notes: In both cases, the equations relating the physical circuit to the transfer Notes:
a
function are quite simple. For the LPF, R2 C = ω1 , with the DC gain: ω0 =
0 0
R2
−R . For the HPF, the equivalent expressions are R1 C = ω1 and for the high-
1 0
a1 R2
frequency gain: ω
0
= − R1 . Also note that a general pole-zero system can
be made using a single op-amp by using a parallel RC circuit in both feedback
and input paths. However such a filter tends to have low input impedence at high
frequencies and is therefore normally avoided. You can even make a single-op-
amp all-pass filter - see Figure 11.14 in S&S for more details.

Each first-order block gives you a flat response in the passband and a -20 dB/decade
attenuation rolloff in the stopband.