Analog Filter Design

Part. 1: Introduction

P. Bruschi - Analog Filter Design 1

 Definition of Filter
• Electronic filters are linear circuits whose operation is defined
in the frequency domain, i.e. they are introduced to perform
different amplitude and/or phase modifications on different
frequency components.

• Digital Filters (or Numerical Filters) operates on digital (i.e.

coded) signals
• Analog Filters operate on analog signals, i.e. signals where the
information is directly tied to the infinite set of values that a
voltage or a current may assume over a finite interval (range).

P. Bruschi - Analog Filter Design 2

 Brief Filter History: The Basis
Harmonic Analysis
 200 BC: Apollonius of Perga theory of “deferents and epicycles”
(the basis of later (100 AD) Ptolemaic system of astronomy),
maybe anticipated by Babylon mathematicians intuitions

 1822 Joseph Fourier: Théorie Analytique de la Chaleur. Preceded by

Euler, d’Alembert and Bernoulli works on trigonometric interpolation.

Electrical Network Analysis

 1845 Gustav Robert Kirchhoff: Kirchhoff Circuit Laws (KCL, KVL)
 1880-1889 Olivier Heaviside developed the Telegrapher's equations and coined the terms
inductance, conductance, impedance etc.
 1893 Charles P. Steinmetz: “Complex Quantities and Their Use in Electrical Engineering“ In
the same year, Arthur Kennely introduced the complex impedance concept

P. Bruschi - Analog Filter Design 3

 Brief Filter History: Resonance
 Acoustic resonance was know since the invention of the first musical instruments.
 Acoustical (i.e. mechanical) resonance allows selecting and enhancing individual
frequencies. It suggested the first FDM (frequency division multiplexing) application
to telegraph lines based on electrical resonance.
 Electrical resonance was first observed
in the discharge transient of Leiden Jars

Not suitable for telephone

FDM, but sufficient for Radio
tuning

Marconi RTX with syntonic tuning

 1898 – Sir Oliver Lodge: Syntonic tuning

P. Bruschi - Analog Filter Design 4

Brief filter history: The beginning of the electronics era

1904 – John Ambrose Fleming Valve (Vacuum Diode)

1906 – Lee De Forest “Audion” Vacuum Triode

1911 – First Triode-based Amplifiers and Oscillators
1912 – Lee De Forest: cascaded amplifier stages

1915 – G. A. Campbell – K.W. Wagner “wave filters”

first example of filter theory:
Image Parameters Theory

P. Bruschi - Analog Filter Design 5

 Brief Filter History: Towards Modern Filter Theory
 1917 – Edwin Howard Armstrong: First Superhetorodyne Radio
 1920 – Introduction of the term “feedback” (referred to positive FB)
 1927 – Harold Stephen Black: application of negative feedback to amplifiers
 1920-30 – Diffusion of Frequency Division Multiplexing for telephone calls.
 1930 – Stephen Butterworth introduces maximally flat filters.
(with amplifier to separate stages)
 1925-40 – Progressive introduction of the Network Syntesys
approach to filter design. Primary contributors was
Wilhelm Cauer
Vacuum tube era

P. Bruschi - Analog Filter Design 6

 Filter operation
The role of a filter can be:

• Modify the magnitude of different frequency components.

These filters are by far the most commonly used.

• Modify the phase of different frequency components (i.e. to

compensate for an unwanted phase response of a filter of an
amplifier)

Real filters generally change both the phase and magnitude of a signal

P. Bruschi - Analog Filter Design 7

 Analog filters: General properties
Common properties:
 Linearity
 Time Invariance
 Stability (BIBO: Bounded Input -> Bounded Output)

Filters may be:

 Lumped / Distributed
 Active / Passive
 Continuous-time / Discrete-Time

P. Bruschi - Analog Filter Design 8

 Lumped element networks
Lumped element networks are made up of components, whose state
and/or behavior is completely defined by a discrete number of
quantities.
Lumped element networks are simplifications or real systems, which
are spatially distributed (i.e. the relevant quantities are given as function
of space variables, defined over a continuous domain).

Lumped Distributed
(e.g. transmission line)

P. Bruschi - Analog Filter Design 9

 Passive and Active Networks (Linear)

Passive network Active network

Passive network:
 Includes only passive components (resistors capacitors, inductors, transformers)
 When connected to external independent sources, the net energy flux into
the network is always positive

P. Bruschi - Analog Filter Design 10

 Continuous and Discrete Time

Discrete time: the signal assumes significant

Continuous time: the signal at all time
values only at time intervals that form a
values belonging to a continuous interval
countable (i.e. discrete) and ordered set.
are significant
The signal is then a sequence of values s(n)
A sequence can be considered as the result of sampling a continuous-time signal
This relationship is not univocal

P. Bruschi - Analog Filter Design 11

 Time and Value Discretization

Discrete
magnitude
(digital)

Continuous
magnitude
(analog)

Discrete Time Continuous Time

P. Bruschi - Analog Filter Design 12
Filter types according to the ideal magnitude response

fc synonyms:

 Corner frequency
 Cut-off frequency

fCL Corner Freq – Low

fCH Corner Freq – High

P. Bruschi - Analog Filter Design 13

 Real filters: the approximation function

 The Low Pass filter is

the reference for other
types
3 dB
 Frequency are general
given as angular
frequencies (w)
 A Transition Band (TB) is
introduced

wc

Note: wp is generally different from wc (-3 dB frequency)

P. Bruschi - Analog Filter Design 14

Approximation parameters for high-pass, band-pass, band-stop filters

P. Bruschi - Analog Filter Design 15

The approximation problem for

time-continuous analog filters

P. Bruschi - Analog Filter Design 16

 Approximation Problem
The first step is reducing the wide variability in filter characteristics by designing
a “reference filter” from which the actual filter can be derived through
“transformations”.
The reference filter, HN(jw), is:
 Low Pass Gain normalization
 Normalized Characteristic frequency normalization

H ( jw  w N ) w
H N ( jw)  H ( jw)  H ( j 0)  H N ( j )
H ( j 0) wN
wN and H(j0) depend on the actual filter
wP wS
H N ( jw) : w P  w PN  w S  w SN 
wN wN

P. Bruschi - Analog Filter Design 17

 Approximation Problem

In order to classify the filters, it is convenient to define the function |k(jw)| as

follows:
1
| K ( jw) | 
2 1
1 | H N ( jw) | 
2

| H N ( jw) | 2 1 | K ( jw) |2

1 in the PB 0 in the PB
Ideal Case: | H N ( jw) |  
2
| K ( jw) |  
2

0 in the SB  in the SB
1
∞
0 0

P. Bruschi - Analog Filter Design 18

 Approximation Problem

 a  1 in the PB
2

Real Case: | K ( jw) |  

2

  2  1 in the SB

 
 H N ( jw)
AP  20 log 
 H ( j 0)

  10 log 

1

  10 log 1  K ( jw) 2
 H ( jw) 2 

 N   N 

 
AP

Worst case: AP  10 log 1  a 2

a  10 10
1

 
AS
Similarly: AS  10 log 1   2
  10 10
1
P. Bruschi - Analog Filter Design 19
 Approximation Problem
We need to find a function K(s) such that |K(jw)|2 satisfies the conditions:


  a 2
 1 for w  w PN
| K ( jw) |  
2

  2  1 for w  w SN
Clearly:
 There are infinite solutions to this mathematical problem
 Solutions should lead to a feasible HN(s)
 Lumped element filter: HN(s) should be a rational functions:
N (s)
H (s)  Where N(s) and D(s) are polynomial
D(s)
P. Bruschi - Analog Filter Design 20
 Notable cases
 Maximally Flat Magnitude Filters (e.g. Butterworth Filters)
 Chebyshev Filters
 Inverse Chebyshev Filters
 Elliptical Filters
 Bessel Filters

General selection criteria:

 The lower the polynomial order, the better the solution
 Monotonic behavior in the PB can be required
 Asymptotic behavior for w>>wS could be important
 Phase response: a linear response is often required

P. Bruschi - Analog Filter Design 21

 Maximally Flat Magnitude (MFM) Approximation
K ( s )  s n

2
K ( jw)   ( jw) n
 K ( jw)   2 w 2 n

1 1
H N ( jw)  
1  K ( jw)
2
1   2w2 n

1 2 2n 3 4 4n 5 6 6n
H N  jw  1   w   w   w  ....
2 8 16

All derivatives up to the (2n-1)th are zero for w=0.

P. Bruschi - Analog Filter Design 22
Maximally Flat Magnitude (MFM) Approximation

2 1
H N ( jw) 
1   2w2 n

Since HN(s) is the Fourier transform of a real signal

(the impulsive response): HN*(jw)=HN(-jw). Then:

1
H N ( jw ) H N (  jw ) 
1   2w 2 n

P. Bruschi - Analog Filter Design 23

 Maximally Flat Magnitude (MFM) Approximation
1
s  jw  w   js H N ( s) H N ( s) 
 s 
n
1  2 2

1
H N (s) 
D(s)
1 1
H N ( s) H N ( s)  
D ( s) D ( s) 1     s 
2 2
n

Thus, HN(s) is an all-poles function:

The roots of D(s)D(-s) are the solutions of 1+2(-s2)n=0

P. Bruschi - Analog Filter Design 24
 Maximally Flat Magnitude (MFM) Approximation

 s 
2 n 1
  2  s   1
2n n 1 1
e j  n 1 1
1
  2
2 1 n
1 Radius:  
1 n  2k  n  1  
s k    exp  j  k : integer
  2n 

 
odd multiples of 2 n for n even
angle : 
even multiples of  for n odd
 2n

P. Bruschi - Analog Filter Design 25

 Maximally Flat Magnitude (MFM) Approximation
n even n odd

sk

1
H N (s) 

n
k 1
( s  sk )

P. Bruschi - Analog Filter Design 26

 Butterworth Filters
 The Butterworth filter is a MFM filter with =1 1
H N ( jw) 
 It can be shown that the case ≠1 is identical to 1  w2 n
=1 but with simple frequency transformation

Dn (s ) are the Butterworth polynomials

1
Non normalized (actual) filter H ( jw) 
2n
 w 
1   
 wN 

P. Bruschi - Analog Filter Design 27

 Butterworth Filters
1 1
H N ( jw)  H N ( jw) w1   3dB wCN  1
1 w 2n
2

w N  wC

Logarithmic
Magnitude

P. Bruschi - Analog Filter Design 28

 Filter Parameter Determination
AP AS

| K ( jw SN ) |  | K ( jw PN ) | a a  10 10
1   10 10
1

| K ( jw) | w n

 
 
log  
AS
 w SN 
n
 w 
n
 10  1 10
n  
    S        wS 
 w PN  wP   2 log   
Ap
 a
10 10  1   wP 
wP
1

1
 w PN  w N  w C  w P a n
w nPN  a  w PN  a n wN

P. Bruschi - Analog Filter Design 29

 Butterworth filter : example

wS
AP
fP=1 kHz fS   1965
 a  10 10
 1  0.509 a
fS=2 kHz wP fP
AP=1 dB
  3.86  10
AS 6
AS=60 dB   10 10
 1  1000

 
 
log  
n   10 .94  11
 
  wS 

1  2 log   
  wP  
wC  w P a n
 2 f P  1.063

P. Bruschi - Analog Filter Design 30

 Chebyshev Filters

K ( jw)  C n w Cn(w): n-th order Chebyshev polynomial

cos n  arccos w  w  1
C n w   
cosh n  arccos h w  w  1

C n  x   2 xC n 1 ( x )  C n  2

C 0  x   1; C1  x   x C 2 x   2 x 2  1

P. Bruschi - Analog Filter Design 31

 Chebyshev polynomials: properties

cos n  arccos w  w  1
C n w    C n w
2

cosh n  arccos h w  w  1
w 6 3rd

For w=1.2
For 0<w<1 oscillates between 0 and 1
For w=0 : 0 if n odd, 1 if n even
2nd
For w=1 : 1 for every n
1st
For w>0 increase monotonically
w
Chebyshev polynomial have the highest leading term coefficient than any other
Polynomial constrained to be less than 1 (in modulus) for w between 0 and 1

P. Bruschi - Analog Filter Design 32

 Chebyshev filters: Pass Band Attenuation
K ( jw)  C n w H N ( jw) 
1
1   2 C n2 (w)
1
0  w 1   H N ( jw)  1
1  2

1 H ( jw) w
wN  wP   HN( j ) 1
1  2 H ( j 0) wN

AP  AP  1 dB    0.5
e.g. 
a    10 10
1  AP  3 dB    1
P. Bruschi - Analog Filter Design 33
Chebyshev filters: Stop Band Attenuation

K ( jw SN )  C n w SN     cosh n  arccosh w SN   


cosh n  arccosh w SN    
AS

 2 10 10  1
 
 2 AP

 arccosh (  )  10 10
1
n min  
 arccosh (w SN ) 

P. Bruschi - Analog Filter Design 34

 Inverse Chebyshev filters (or Chebyshev type II)
Target: Obtain monotonic behavior in the pass-band (no ripple) and ripple
In the stop-band
1
1 H N ( jw) 
K ( jw)  1
1 1
C n   1
w  C ( )
2 2

w
n

1
H N ( j1) 
1
1 2


P. Bruschi - Analog Filter Design 35

 Elliptic (or Cauer) Filters
1
K ( jw)  Rn w H N ( jw) 
1   2 Rn2 w
Rn : “elliptical” rational function : NR(w) / DR(w)

 n / 2 w 2  w 2pk
M  2 for n even
 k 1 w  w 2

Rn w   
zk
 n 1 / 2 w 2  w 2

w  w2  w2
pk
 M for n odd
 k 1 zk

w pk w zk  w s
M such that Rn (w)  1 for | w | 1
P. Bruschi - Analog Filter Design 36
 Phase, delay, group delay
In order to maintain the shape of a generic signal, the following conditions
must be respected:
 All the significant frequency components of the signal fall into the filter
pass-band, which should be as flat as possible:
 The filter phase response in the pass-band should be of the type:

  wt R
If this condition is fulfilled, the input signal is simply delayed by time tR.

In other words, the group delay should be constant. The group delay
is defined as:
d
G  
dw
P. Bruschi - Analog Filter Design 37
 Bessel Filters
Constant group delay is very important in systems that has to handle digital
transmissions, where signal distortion may result in high BER (Bit Error Rate),
or even in unrecoverable signal.

The Bessel filter is obtained by considering an all-pole function:

H N s  
K
D s 
We start from a generic polynomial D(s), substitute s=jw and than calculate
the phase of HN by:
 Im D  jw 
   arctan  
 Re D  jw  

P. Bruschi - Analog Filter Design 38

 Bessel Filters
The Bessel Filters are derived by:
 Taylor’s expansion of the arctan() functionis calculated obtaining a
polynomial approximation of the phase;
 The first derivative of the phase approximation is calculated
 The constant term of the derivative is set to 1 (group delay=1)
 Higher order terms are set to zero; this corresponds to setting the
derivatives of the group delay to zero. The number of derivatives that can be
nulled depends on the filter order.

 The result is a maximally flat group delay

 Can be used to purposely introduce a delay

P. Bruschi - Analog Filter Design 39

 Bessel filters, examples
Denominators of various order for unity group delay
1 : D ( s )  s  1; K 1 w  3 dB  2n  1 ln( 2)
2 : D ( s )  s 2  3s  3; K 3
Approximate
3 : D ( s )  s 3  6 s 2  15 s  15; K  15 expression for n ≥ 3
........
D n  2 n  1D n 1  s 2 D n  2 (Bessel polynomials, recursive expression)
w 1
w  D  (pass-band group delay after frequency scaling)
wN wN
The Bessel filter is less selective than a Butterworth filter of same
order, but its phase response is much more linear

P. Bruschi - Analog Filter Design 40

 Summary of filter characteristics
 Butterworth: maximally flat in the pass-band and monotonic
everywhere
 Chebyshev: More selective than Butterworth (sharper transition),
but ripple in the pass-band (monotonic in the stop-band)
 Inverse Chebyshev: Same selectivity than Chebyshev, but ripple in
the stop-band (flat in the pass-band). Magnitude do not
decrease asymptotically in the stop-band
 Elliptic: Best selectivity, but ripple in both the pass-band and stop-
band. Magnitude do not decrease asymptotically in the
stop-band
 Bessel: The least selective of all other filters, but the best in terms
of phase linearity (constant group-delay in the pass-band)

P. Bruschi - Analog Filter Design 41

 Other continuous time filters
 Optimum L-filter (Papoulis)
Obtains the best selectivity with a monotonic response. Compared with a
Butterworth of the same order filter it is sharper in the transition band, but
less flat (but still monotonic) in the pass-band.
 All pass filters (phase equalizers). Their common characteristic is that for
each pole they have a zero with opposite real part. As a result, they have
RHP (right half plane) zeros and their step response is generally preceded
by a glitch in the opposite direction with respect to the final value.
For step-like signals, low-pass phase equalizers (e.g. Bessel filters) are to
be preferred.
 Filters based on Padè approximations: The Padè approximation is the
best n-order rational function that approximate an arbitrary function. It is
used for the approximation of the ideal delay: exp(-jwtD). The all-pass
functions are a particular case of Padè approximation.

P. Bruschi - Analog Filter Design 42

 Frequency transformations
The aim of frequency transformations are:
 Change the characteristic frequencies with respect to the normalized case

s
sn  All characteristic frequencies are multiplied by wN
wN

 Change the low pass response into an high pass, band-pass etc.
1
wN w0  s w0   w0  s w0 
sn  sn     sn    

B  w0 s   w  s 
s  B  0  

High pass Band-Pass Band Stop

P. Bruschi - Analog Filter Design 43

 Pass Band transformation: meaning of B, w0
w0  s w0  w 0  jw w 0  w0  w w0 
sn     jw n     j   

B  w0 s  B  w 0 jw  B  w 0 w 

w0  w w0 
wn    

B  w0 w 
w
for small variations around w0, such that:  1
w0
 
 
w 0  w 1  w
w 0  w  1  2
B  w0 w  B
 1 
 w0 
P. Bruschi - Analog Filter Design 44
 Pass Band transformation: meaning of B, w0

2 2
 B  B  B  B
wn  1  w  w2  w0 1    
 w n  1  w  w1  w 0 1    

 2w 0  2  2w 0  2

w 2  w1  B
2
w 2  w1  B 
 w 0 1   
  w0
2  2w 0 
for : B  w 0

P. Bruschi - Analog Filter Design 45

 Pass Band transformation: meaning of B, w0
 When w=w0, wn =0. Then, the response of the
pass-band filter at w0 is the D.C. value (wn=0)
of the prototype low pass filter.
 For w variations from w0, wn moves away from
the origin. When w<w0, wn is negative, so that
H(w) is the complex conjugate of the values
at w>w0 (see the phase diagram in the figure)
 The bandwidth B is the difference between
the frequencies w1 and w2, for which the
absolute value of the normalized frequency is
unity.
 If the bandwidth B is much smaller than
frequency w0 (selective filter), than w1 and w2
are symmetrical with respect to w0. 0

P. Bruschi - Analog Filter Design 46