Basics of Digital Filters

Elena Punskaya
www-sigproc.eng.cam.ac.uk/~op205
Some material adapted from courses by Prof. Simon Godsill, Dr. Arnaud Doucet, Dr. Malcolm Macleod and Prof. Peter Rayner

What is a Digital Filter?

Digital Filter: numerical procedure or algorithm that transforms a given sequence of numbers into a second sequence that has some more desirable properties.

xn

Input sequence

Digital Filter

Output sequence

yn

Desired features

Desired features depend on the application, for example Input Signal generated by sensing device (microphone) speech Output having less noise or interferences with reduced redundancy for more efficient transmission

Examples of filtering operations Noise suppression received radio signals signals received by image sensors (TV, infrared imaging devices) electrical signals measured from human body (brain heart, neurological signals) signals recorded on analog media such as analog magnetic tapes
4

Examples of filtering operations Enhancement of selected frequency ranges

treble and bass control or graphic equalizers
increase sound level and high and low level frequencies to compensate for the lower sensitivity of the ear at those frequencies or for special sound effects

enhancement of edges in images

improve recognition of object (by human or computer) edge a sharp transition in the image brightness, sharp transitions in a signal (from Fourier theory) appear as high-frequency components which can be amplified
5

Examples of filtering operations Bandwidth limiting

means of aliasing prevention in sampling

communication radio or TV signal transmitted over specific channel has to have a limited bandwidth to prevent interference with neighbouring channels frequency components outside the permitted band are attenuated below a specific power level
6

Examples of filtering operations Specific operations

differentiation integration Hilbert transform

These operations can be approximated by digital filters operating on the sampled input signal
7

Linear time-invariant (LTI) digital filters.

We limit ourselves to LTI digital filters only. Time-invariant: xn LTI yn xn-k LTI yn-k

Linear: defined by the principle of linear superposition xn1 LTI yn1 xn2 LTI LTI yn2
scaling & additivity properties

a1xn1+ a2xn2

a1yn1+ a2yn2

If linear system's parameters are constant it is Linear Time Invariant

Analysis We analyse DSP algorithms by determining: their time-domain characteristics linear difference equations filters unit-sample (impulse) response their frequency-domain characteristics more general, Z-transform domain system transfer function poles and zeros diagram in the z-plane Fourier domain frequency response spectrum of the signal
9

First method in time domain: Linear difference equations

The linear time-invariant digital filter can then be described by the linear difference equation:

where {ak} and {bk} real The order of the filter is the larger of M or N
10

Elements of a Digital Filter Adders and Multipliers Adders:

x1 n
Multipliers:

yn=x1n+x2n x2 n

xn

yn=axn

Simple components implemented in the arithmetic logic unit of the computer

11

Elements of a Digital Filter Delays Positive delay (delay):

stores the current value for one sample interval

xn

Z-1

yn= xn-1
allows to look ahead, e.g. image processing

Negative delay (advance):

xn

yn= xn+1

Components that allow access to future and past values in the sequence
12

Example: three-sample averager Three-sample averager: yn=(xn+1+ xn + xn-1)/3

Z

1/3

xn
Order: number of delays & advances needed to implement Z-1

yn
second-order filter

Nonrecursive filter output is a function of only the input sequence

13

Example: first-order recursive filter First-order recursive filter: yn=ayn-1+ xn

xn

+
Z-1

yn a
one delay first-order

Example of feedback implemented in a digital filter

Recursive filter output is also a function of the previous output

14

Full set of possible linear operations

The operations shown in the slides above are the full set of possible linear operations: constant delays (by any number of samples) addition or subtraction of signal paths multiplication (scaling) of signal paths by constants - (including -1) Any other operations make the system nonlinear.
15

Linear difference equations and digital filter structure Useful for implementing digital filter structures

xn b0 bM

= unit delay yn

a1

aN

slightly alternative representation

16

Second method in time domain: unit-sample response

Input signal is resolved into a weighted sum of elementary signal components, i.e. sum of unit samples or impulses

xn= is determined

k=-

xk n-k

The response of the system to the unit sample sequence

n
LTI hn

Taking into account properties of the LTI system, the response of the system to xn is the corresponding sum of weighted outputs

xn=

k=-

xk n-k

LTI

yn=
k=-

xk hn-k

17

Linear convolution

Linear convolution

yn=
k=-

xk hn-k

gives the response of the LTI system as a function of the input signal and the unit sample (impulse) response

LTI is completely characterized by hn

18

Causal LTI system Causal system: output at time n depends only on

present and past inputs but not on future

Impulse response: hn = 0
Thus

for n<0

yn=
k=-

xk hn-k =

k=0

hk xn-k

19

Essentials of the z-transform

For a discrete time signal

where

is a complex variable

For causal signals

is well defined for

20

Essentials of the z-transform - convolution

Convolution

admits a z-transform satisfying

where

Indeed,

21

Transfer function of LTI Linear difference equation

Now, take z-transforms term by term

Rearranging, transfer function of the filter is

22

FIR and IIR filters

Transfer function of the filter is

Finite Impulse Response (FIR) Filters: N = 0, no feedback Infinite Impulse Response (IIR) Filters

23

Poles and Zeros

The roots of the numerator polynomial in H(z) are known as the zeros, and the roots of the denominator polynomial as poles. In particular, factorize H(z) top and bottom:

24

Linear Digital Filter in Matlab

Matlab has a filter command for implementation of linear digital filters. Matlab transfer function: The format is
y = filter( b, a, x);

where
b = [b0 b1 b2 ... bM ]; a = [ 1 a1 a2 a3 ... aN ];

So to compute the first P+1 samples of the filters impulse response,

y = filter( b, a, [1 zeros(1,P)]);

Or step response,
y = filter( b, a, [ones(1,P)]);
25

z-transform and DTFT

DTFT

Similarly, to z-transforms a convolution theorem holds:

26

Frequency Response of LTI LTI system with impulse response hk

xn

LTI

yn=
k=0

hk xn-k convolution

Fourier Transform

Fourier transform of hk

- frequency response of LTI

27

Frequency Response of LTI

Indeed, let us excite the system with the complex exponential
where

The response of the system to complex exponential

Fourier transform of hk

is also in the form of complex exponential but altered by the multiplicative factor

28

Example Assume
where for 0 otherwise

Frequency response

29

Frequency Response of LTI

By knowing we can determine the response of the system to any sinusoidal input signal, hence it specifies the response of the system in the frequency domain

=
is called magnitude response of a system

is called phase response of a system

30

Frequency response of LTI

Transfer function: Frequency response:

The complex modulus:

and argument:

linear phase term

sum of the angles from the zeros/poles to 31 circle unit

Frequency response of LTI

Im(z) unit circle X

Suppose we have a system with 2 poles (o) and 2 zeros (x) We are going around the unit circle with
O 1

-1

32

Frequency response of LTI

Im(z) unit circle X
D1

Transfer function:

-1

O
C1

D2

C2

O 1

Frequency response:
C1C2 D1D2
33

Frequency response of LTI

Im(z) unit circle X
D1 C1C2 D1D2

-1

O
C1

D2

C2

O 1

The magnitude of the frequency response is given by times the product of the distances from the zeros to divided by the product of the distances from the poles to

The phase response is given by the sum of the angles from the zeros to minus the sum of the angles from the poles to plus a linear phase term (M-N) 34

Frequency response of LTI

Im(z) unit circle X
D1 Thus when 'is close to' a pole, the magnitude of the response rises (resonance). When 'is close to' a zero, the magnitude falls (a null).

-1

O
C1

D2

C2

O 1

The phase response more difficult to get intuition.

35

Frequency response of filter in Matlab

To evaluate the frequency response at n points equally spaced in the normalised frequency range =0 to = , Matlab's function freqz is used: freqz(b,a,n);
b=[1 -0.1 -0.56]; a=[1 -0.9 0.81]; freqz(b,a) Peak close to pole frequency

Troughs at zero frequencies

36

36

Filtering example
Generate a Gaussian random noise sequence:
x=randn(100000,1); figure(1), plot(x) figure(2), plot(abs(fft(x))) a = [1 -0.99 0.9801]; b = [1, -0.1, -0.56]; y=filter(b,a,x); figure(3), plot(y) Figure(4), plot(abs(fft(y)))

freqz(b,a);

Selective amplification of one frequency

Use soundsc(x,44100) and soundsc(y,44100) and hear the difference!
37 37

Filter specification
Before a filter is designed and implemented we need to specify its performance requirements. There are four basic filter types: Low-pass High-pass Band-pass Band-stop

Frequency band where signal is passed is passband Frequency band where signal is removed is stopband
38

Ideal Low-pass Filter

Low-pass: designed to pass low frequencies from zero to a certain cut-off frequency and to block high frequencies

Ideal Frequency Response

39

Ideal High-pass Filter

High-pass: designed to pass high frequencies from a certain cut-off frequency to and to block low frequencies

Ideal Frequency Response

40

Ideal Band-pass Filter

Band-pass: designed to pass a certain frequency range which does not include zero and to block other frequencies

Ideal Frequency Response

41

Ideal Band-stop Filter

Band-stop: designed to block a certain frequency range which does not include zero and to pass other frequencies

Ideal Frequency Response

42

Ideal Filters Magnitude Response

Ideal Filters are usually such that they admit a gain of 1 in a given passband (where signal is passed) and 0 in their stopband (where signal is removed).

43

Ideal Filters Phase Response

Another important characteristics is related to the phase Ideal filter: admits a linear phase response

Indeed,

Fourier Transform of delay

44

Ideal Filters Linear Phase Response

Important property: all frequencies suffer from the same delays

In some applications it is critical for this property to hold (at least approximately)

45

Linear / non-linear phase response

phase distortion
46

Ideal Filters Linear Phase Response

The derivative of the phase/signal respect with respect to is known as group delay of the filter - effectively time delay of the frequency

when the phase admits a linear phase response

group delay is constant

47

Designing Ideal Filters is Impossible

48

Ideal Low- pass Filter Example

49

Filter Design

Lets have a go

50

Frequency response of LTI

Remember?
unit circle

Im(z)
Thus when 'is close to' a pole, the magnitude of the response rises (resonance). When 'is close to' a zero, the magnitude falls (a null).

X
D1

-1

O
C1

D2

C2

O 1

The phase response more difficult to get intuition.

51

Approximate filter design rough guidelines

A few rough guidelines to start with:

One needs to put the poles within the unit circle to ensure stability

52

Approximate Low-pass filter

Rule 1: The closer to a pole, the higher the magnitude of the response

Rule 2: The closer to a zero, the lower the magnitude of the response

53

Approximate Low-pass filter

one pole

with added zero

54

Approximate High-pass filter

We had for Low-pass

one can simply reflect the poles-zeros locations of the lowpass filter about the imaginary axis
55

Approximate High-pass filter

56

Alternative way: Low-pass to High-pass

in time domain

57

Bandpass and Resonator

a filter which has its centre of the passband at 0

58

Resonator

This filter is actually more a digital resonator than an bandpass filter; see its frequency response for r = 0.9 and 0 = /4 - it has a large magnitude response around 0
59

Band-pass Filter

60

Band-pass Filter

frequency response for r = 0.9 and 0 = /4

61

Notch Filter
a filter that contains one or several deeps/ notches in its frequency response

to eliminate 0
Frequencies around the desired null are also seriously attenuated

62

Notch Filter
Frequencies around seriously attenuated

with added pair of poles

pair of zeros only

63

All pass Filter

Used as phase equalizer

64

Inverse Filter

The zeros become poles and the poles become zeros

65

Digital Filter Design Considerations

Four steps: 1. 2. Specification of the filters response (very important! senior engineers) Design the transfer function of the filter (main goal: meet spec with minimum complexity, often = minimum order) Verification of the filters performance analytic means simulations testing with real data if possible Implementation by hardware / software (or both)
66

3.

4.

Approximate filter design

Given precise specification what would you do??

67

Thank you!

68