Chapter 2 Discrete-Time Signals & Systems - 清華大學電 …cwlin/courses/dsp/notes/ch2_Mitr… · · 2008-03-172008/3/17 1 Discrete-Time Signals & Systems ... • Discrete-time

2008/3/17

1

Discrete-Time Signals & Systems

Chapter 2

The McGraw-Hill Companies, Inc., 2007Original PowerPoint slides prepared by S. K. Mitra 2-1-1

[email protected]

Discrete-Time Signals: Time-Domain Representation (1/10)

Signals represented as sequences of numbers, called samplessamples

Sample value of a typical signal or sequence denoted as x[n] with n being an integer in the range n

x[n] defined only for integer values of n and undefined for non-integer values of n

Discrete-time signal represented by {x[n]}


Discrete time signal represented by {x[n]}

Discrete-time signal may also be written as a sequence ofnumbers inside braces:

2008/3/17

2


Graphical representation of a discrete-time signal with real-valued samples is as shown below:real valued samples is as shown below:



In some applications, a discrete-time sequence {x[n]} may be generated by periodically sampling a continuous-may be generated by periodically sampling a continuoustime signal xa(t) at uniform intervals of time


Here, the n-th sample is given by

2008/3/17

3


The spacing T between two consecutive samples is called the sampling interval or sampling periodcalled the sampling interval or sampling period

Reciprocal of sampling interval T, denoted as FT, is called the sampling frequency (in Hz)

A complex sequence {x[n]} can be written as {x[n]} = {xre[n]}+ j{xim[n]} where xre[n] and xim[n] are the real and


{xre[n]} j{xim[n]} where xre[n] and xim[n] are the real and imaginary parts of x[n]

Often the braces are ignored to denote a sequence if there is no ambiguity


Example - {x[n]} = {cos0.25n} is a real sequence

{ [ ]} { j0 3n} i l {y[n]} = {ej0.3n} is a complex sequence

We can rewrite

where


where

2008/3/17

4


Two types of discrete-time signals:Sampled data signals in which samples are Sampled-data signals in which samples arecontinuous-valued

Digital signals in which samples are discrete-valued (by quantizing the sample values either byrounding or truncation)



A discrete-time signal may be a finite-length or an infinite-length sequenceinfinite length sequence

Example - x[n] = n2, 3 n 4 is a finite-length sequence of length 8

y[n] = cos0.4n is an infinite-length sequence

A length-N sequence is often referred to as an N-point sequence


sequence The length of a finite-length sequence can be

increased by zero-padding, i.e., by appending it with zeros

2008/3/17

5


A right-sided sequence x[n] has zero-valued samples for n < N1for n N1

If N1 0, a right-sided sequence is called a causal


1sequence

A left-sided sequence x[n] has zero-valued samples for n > N2 (called a anti-causal sequence if N2 0)


Size of a Signal - given by the norm of the signal

Lp-norm

where p is a positive integer

The value of p is typically 1 or 2 or

L i th t d ( ) l f { [ ]}


L2-norm is the root-mean-squared (rms) value of {x[n]}

L1-norm is the mean absolute value of {x[n]}

L-norm is the peak absolute value of {x[n]} (why?)1

x2

x

x

m axx x

=

2008/3/17

6


Example -L { [ ]} 0 N 1 b i i f { [ ]} 0 Let {y[n]}, 0 n N 1, be an approximation of {x[n]}, 0 n N 1

An estimate of the relative error is given by the ratio of the L2-norm of the difference signal and the L2-norm of {x[n]}:

1 / 212[ ] [ ]

N

y n x n


01

2

0

[ ] [ ]

[ ]

nrel N

n

y n x nE

x n

=

=

=

Elementary Operations on Sequences

Product (modulation) operation:


An application is in forming a finite-length sequence from an infinite-length sequence by multiplying the latter with a finite-length sequence called an window sequence (windowing)

2008/3/17

7


Addition operation:

Multiplication operation:


Time-shifting operation: y[n] = x[n N](Unit Delay)

(Unit Advance)


Time-reversal (folding) operation:

[ ] [ ]y[n] = x[n] Branching operation: used to provide multiple copies of

a sequence


Example:

2008/3/17

8

Elementary Operations on SequencesEnsemble Averaging An application of the addition operation in improving the

quality of measured data corrupted by an additive random noise

Let di denote the noise vector corrupting the i-thmeasurement of the uncorrupted data vector s

The average data vector called the ensemble average


The average data vector, called the ensemble average, obtained after K measurements is given by



2008/3/17

9

Combinations of Basic Operations

Example -


Sampling Rate Alteration

A process to generate a new sequence y[n] with a sampling rate higher or lower than that of the samplingTF sampling rate higher or lower than that of the sampling rate FT of a given sequence x[n]

Sampling rate alteration ratio:

if R > 1 the process called interpolation

T


if R > 1, the process called interpolation

if R < 1, the process called decimation

2008/3/17

10

Up-Sampling In up-sampling by an integer factor L > 1, L 1 equidistant

zero-valued samples are inserted between each two consecutive samples of the input sequence x[n]:


Down-Sampling In down-sampling by an integer factor M > 1, every M-th

samples of the input sequence are kept and M 1 in-between samples are removed:


2008/3/17

11

Classification of SequencesBased on Symmetry (1/4)

Conjugate-symmetric sequence:

If x[n] is real, then it is an even sequence for a conjugate-symmetric sequence {x[n]}, x[0]

must be a real number



Conjugate-antisymmetric sequence:

If x[n] is real, then it is an odd sequence for a conjugate anti-symmetric sequence {y[n]}, y[0]

must be an imaginary number


2008/3/17

12


Any complex sequence can be expressed as a sum of its conjugate-symmetric part and its conjugate-antisymmetricj g y p j g ypart:

where

Consider the length-7 sequence defined for 3 n 3



Any real sequence can be expressed as a sum of its even part and its odd part:even part and its odd part:

where


2008/3/17

13

Classification of SequencesBased on Periodicity

A sequence satisfying is is called a periodic sequence with a period N where N is a positiveperiodic sequence with a period N where N is a positive integer and k is any integer

Smallest value of N satisfying is calledthe fundamental period


A sequence not satisfying the periodicity condition is called an aperiodic sequence

Classification of SequencesEnergy & Power Signals (1/3)

Total energy of a sequence x[n] is defined by

An infinite length sequence with finite sample values may or may not have finite energy

The average power of an aperiodic sequence is defined by


Define the energy of a sequence x[n] over a finite interval K n K as

2008/3/17

14


The average power of a periodic sequence with a period N is given by

The average power of an infinite-length sequence may be finite or infinite

Example - Consider the causal sequence defined by


Note: x[n] has infinite energy, its average power is


An infinite energy signal with finite average power is called a power signalcalled a power signal Example - A periodic sequence which has a finite

average power but infinite energy

A finite energy signal with zero average power is called an energy signal Example - A finite-length sequence which has finite


Example A finite length sequence which has finiteenergy but zero average power

2008/3/17

15

Other Types of Classification (1/2) A sequence x[n] is said to be bounded if

Example - The sequence x[n] = cos0.3n is a bounded sequence as

A sequence x[n] is said to be absolutely summable if


Example - The following sequence is absolutely summable

Other Types of Classification (2/2)

A sequence x[n] is said to be square summable if

Example - The sequence

is square-summable but not absolutely summable


is square summable but not absolutely summable

2008/3/17

16

Basic Sequences (1/7)

Unit Sample Sequence -

Unit Step Sequence -



Real Sinusoidal Sequence

Example: A = 2, o = 0.1


2008/3/17

17


Exponential Sequence where A and are real or complex numberswhere A and are real or complex numbers

If we write


Basic Sequences (4/7) Sinusoidal sequence Acos(on + ) and complex

exponential sequence Bexp( jon ) are periodic sequences of period N if oN = 2r where N and r aresequences of period N if oN 2r where N and r are positive integers

Smallest value of N satisfying oN = 2r is thefundamental period of the sequence

To verify the above fact, consider


x1[n] = x2[n] if and only if oN = 2r or If 2/o is a noninteger rational number, then the period

will be a multiple of 2/o; otherwise, its aperiodic

2008/3/17

18


Here o = 0 Here o = 0.1


period period

Basic Sequences (6/7) Property 1 - Consider x[n] = exp(j1n) and y[n] =

exp(j2n) with 0 1 < and 2k 2 < 2(k +1)where k is any positive integer If 2 = 1 + 2k, then x[n] = y[n] then x[n] and y[n] are indistinguishable

Property 2 - The frequency of oscillation of Acos(on)increases as increases from 0 to , and then decreases as increases from to 2


as c eases o to frequencies in the neighborhood of = 0 (or 2k) are

called low frequencies, whereas, frequencies in the neighborhood of = (or (2k+1)) are called high frequencies

2008/3/17

19


An arbitrary sequence can be represented in the time-domain as a weighted sum of some basic sequence anddomain as a weighted sum of some basic sequence and its delayed (advanced) versions


The Sampling Process (1/3)

Sampling Process Convert x(t) to numbers x[n] Convert x(t) to numbers x[n] n is an integer; x[n] is a sequence of values Think of n as the storage address in memory

Uniform Sampling at t = nT IDEAL: x[n] = x(nT)


C-to-Dx(t) x[n]

2008/3/17

20

The Sampling Process (2/3)

Often, a discrete-time sequence x[n] is developed byuniformly sampling a continuous-time signal as followsuniformly sampling a continuous time signal as follows


Time variable t of xa(t) is related to the time variable n ofx[n] only at discrete-time instants given by

The Sampling Process (3/3) Consider the continuous-time signal

The corresponding discrete-time signal is


is the normalized digital angular frequency of x[n] The unit of normalized digital angular frequency o is

radians/sample ( o : radians/second)

2008/3/17

21

Sampling for Audio CD

x[n] is a sampled sinusoid A list of numbers stored in memoryA list of numbers stored in memory

Example: audio CD CD rate is 44,100 samples per second

16-bit samples Stereo uses 2 channels

Number of bytes for 1 minute is


2 X (16/8) X 60 X 44100 = 10.584 Mbytes So, a CD-ROM of 680-Mbyte can store up to about

one-hour music What about MP3?

Ambiguity in Sampling Sample the following three signals at 10 Hz

we obtain


g1[n] = g2[n] = g3[n]

2008/3/17

22

Aliasing (1/2) The phenomenon of a continuous-time signal of higher

frequency acquiring the identity of a sinusoidal sequenceof lower frequency after sampling is called aliasingof lower frequency after sampling is called aliasing

There are an infinite number of continuous-time signalsthat can lead to the same sequence when sampledperiodically

The family of continuous-time sinusoids leads to identicalsampled signals

( ) cos(( ) ) 0 1 2x t A t k t k= + + =


( ),

,

( ) cos(( ) ), 0, 1, 2,...

2( ) cos(( ) ) cos

2cos cos( ) [ ]

a k o T

o Ta k o T

T

oo

T

x t A t k t k

k nx nT A k nT A

nA a n x n

= + + =

+ = + + = +

= + = + =

Aliasing (2/2)Given the samples, draw a sinusoid through the values


)4.0cos(][ nnx = )4.2cos()4.0cos(integer an is When

nnn

=

2008/3/17

23

Sampling Theorem (1/2)

Recall

Thus if T > 2o, then the corresponding normalized digital angular frequency o of the discrete-time signal obtained by sampling the parent continuous-time sinusoidal signal will be in the range < <

No Aliasing


On the other hand, if T < 2o, the normalized digitalangular frequency will foldover into a lower digitalfrequency o = 2o / T 2 in the range < < because of aliasing

Sampling Theorem (2/2)

To prevent aliasing, the sampling frequency T should begreater than 2 times the frequency of the sinusoidalgreater than 2 times the frequency o of the sinusoidalsignal being sampled

Generalization: Consider an arbitrary continuous-timesignal x(t) composed of a weighted sum of a number ofsinusoidal signals


2008/3/17

24

Discrete-Time System

A discrete-time system processes a given input sequence x[n] to generates an output sequence y[n] with more [ ] g p q y[ ]desirable properties

In most applications, the discrete-time system is a single-input, single-output system:


2-input, 1-output discrete-time systems - Modulator, adder 1-input, 1-output discrete-time systems - Multiplier, unit

delay, unit advance

Discrete-Time System Examples (1/10) Accumulator -

The output y[n] is the sum of the input sample x[n] and the previous output y[n 1]

The system cumulatively adds, i.e., it accumulates all input sample values

Input-output relation can also be written in the form


Input-output relation can also be written in the form

The second form is used for a causal input sequence, in which case y[1] is called the initial condition

2008/3/17

25

Discrete-Time System Examples (2/10) M-point moving-average system -

Used in smoothing random variations in data A direct implementation of the M-point moving average

system requires M 1 additions, 1 division A more efficient implementation, which requires 2

additions and 1division


additions and 1division

Discrete-Time System Examples (3/10)

An application: Consider x[n] = s[n] + d[n] where s[n] isthe signal corrupted by a noise d[n]g p y [ ]


2008/3/17

26

Discrete-Time System Examples (4/10) Exponentially Weighted Running Average Filter

requires only 2 additions, 1 multiplication and storage of the previous running average

does not require storage of past input data samples the filter places more emphasis on current data samples

and less emphasis on past ones as illustrated below



Linear interpolation - employed to estimate sample values between pairs of adjacent sample values of avalues between pairs of adjacent sample values of a discrete-time sequence

Factor-of-4 interpolation


2008/3/17

27


Factor-of-2 interpolator






2008/3/17

28

Discrete-Time System Examples (8/10)Median Filter The median of a set of (2K+1) numbers is the number

h h K b f h h l hsuch that K numbers from the set have values greater than this number and the other K numbers have values smaller

Median can be determined by rank-ordering the numbers in the set by their values and choosing the number at the middle

Example: Consider the set of numbers


Rank-order set is given by

med{2, 3, 10, 5, 1} = 2


Median Filter Implemented by sliding a window of odd length over the

input sequence {x[n]} one sample at a time

Output y[n] at instant n is the median value of the samples inside the window centered at n

Useful in removing additive random noise, which shows up as sudden large errors in the corrupted signal


up as sudden large errors in the corrupted signal

Usually used for the smoothing of signals corrupted by impulse noise

2008/3/17

29

Discrete-Time System Examples (10/10) Median Filtering Example


Classifications of Discrete-Time Systems

Linear system Shift-invariant system) Causal system Stable system Passive and lossless system


2008/3/17

30

Linear Discrete-Time Systems (1/3) Definition - If is the output y1[n] due to an input x1[n] and

y2[n] is the output due to an input x2[n] then for an inputx[n] =x1[n] + x2[n]

the output is given byy[n] =y1[n] + y2[n]

for any arbitrary constants and Example: Accumulator

F i t


For an input

the output is

Linear Discrete-Time Systems (2/3) If the outputs y1[n] and y2[n] for inputs x1[n] and x2[n]

are given by +=n

lxyny 111 ][]1[][

The output y[n] for an input x1[n] + x2[n] is

=

=

+=n

l

l

lxyny

lxyny

0222

0111

][]1[][

][][][


Now

= =

= =

+++=

+++=

+

n

l

n

l

n

l

n

l

lxlxyy

lxylxy

nyny

0 02121

0 02211

21

])[][(])1[]1[(

][]1[(])[]1[(

][][

2008/3/17

31

Linear Discrete-Time Systems (3/3)

For the causal accumulator to be linear the condition y[-1] = y1 [-1] + y2 [-1] must hold for all initialy[ 1] y1 [ 1] y2 [ 1] must hold for all initial conditions y[1]. y1 [-1], y2 [-1] , and constants and

This condition cannot be satisfied unless the accumulator is initially at rest with zero initial condition

For nonzero initial condition, the system is nonlinear


Non-Linear Discrete-Time Systems The median filter described earlier is a nonlinear discrete-

time systemy To show this, consider a median filter with a window of

length 3 The output of the filter for an input {x1[n]}= {3, 4, 5} 0 n

2, is {y1[n]}= {3, 4, 4} 0 n 2 The output for an input {x2[n]}= {2, 1, 1}, 0 n 2 is

{y [n]}= {0 1 1} 0 n 2


{y2[n]}= {0, 1, 1}, 0 n 2 However, the output for an input {x[n]}= {x1[n] + x2[n]} is

{y[n]}= {3, 4, 3} {y1[n] + y2[n]}= {3, 3, 3} Hence, the median filter is a nonlinear discrete-time system

2008/3/17

32

Shift Invariant Systems (1/2)

For a shift-invariant system, if y1[n] is the response to an input x1[n], then the response to an input x[n] = x1[n n0]input x1[n], then the response to an input x[n] x1[n n0] is simply

y[n] = y1[n n0]where n0 is any positive or negative integer

In the case of sequences and systems with indices n related to discrete instants of time, the above property is called time invariance property


called time-invariance property Time-invariance property ensures that for a specified

input, the output is independent of the time the input is being applied

Shift Invariant Systems (2/2) Example Consider the following up-sampler

for input x1[n] = x[n no] the output x1,u[n] is


However, the upsampler is a time-varying system because 0 0 0 0

0

1,

[( ) / ], , , 2 ,...[ ]

0, otherwise[ ]

u

u

x n n L n n n L n Lx n n

x n

= =

2008/3/17

33

Linear Time-Invariant Systems

Linear Time-Invariant (LTI) System - A system satisfying both the linearity and the time-invariancesatisfying both the linearity and the time invariance property

LTI systems are mathematically easy to analyze and characterize, and consequently, easy to design

Highly useful signal processing algorithms have been developed utilizing this class of systems over the last


several decades

Causal Systems (1/2)

In a causal system, the no-th output sample y[no]depends only on input samples x[n] for n no and doesdepends only on input samples x[n] for n no and doesnot depend on input samples for n > no

Let y1[n] and y2[n] be the responses of a causal system to the inputs x1[n] and x2[n] , respectively. Then

x1[n] = x2[n] for n < N

Implied also that


Implied also that

y1[n] = y2[n] for n < N

For a causal system, changes in output samples do not precede changes in the input samples

2008/3/17

34

Causal Systems (2/2) Examples of causal systems:

Examples of noncausal systems:y[n] = xu[n] + 1/2 (xu[n 1] + xu[n +1])y[n] = xu[n] + 1/3 (xu[n 1] + xu[n +2]) + 2/3 (xu[n 2] + xu[n +1])

A noncausal system can be made causal by delaying the


A noncausal system can be made causal by delaying the output by an appropriate number of samples

A causal implementation of the factor-of-2 interpolator

Stable Systems Bounded-Input, Bounded Output (BIBO) stability If y[n] is the response to an input x[n] and if

Example the M-point moving average filter is BIBOstable

then


With a bounded input

2008/3/17

35

Passive and Lossless Systems A discrete-time system is defined to be passive if, for

every finite-energy input x[n], the output y[n] has, atmost the same energymost, the same energy

For a lossless system, the above inequality is satisfied with an equal sign for every input

Example - Consider the discrete-time system defined by y[n] = x[n N] with N a positive integer

=

=

nn

nxny 22 ][][


y[n] = x[n N] with N a positive integer Its output energy is given by

passive system if

2008/3/17

36

Impulse Response Example - The impulse response of the discrete-time

accumulator

is obtained by setting x[n] = [n] resulting in

Example - The impulse response {h[n]} of the factor-of-2 i l


interpolator

is

])1[]1[(21][][ +++= nxnxnxny uuu

])1[]1[(21][][ +++= nnnnh

Time-Domain Characterization of LTI Discrete-Time System (1/3)

Input-Output Relationship -A consequence of the linear time invariance property isA consequence of the linear, time-invariance property is that an LTI discrete-time system is completely characterized by its impulse response

Knowing the impulse response, one can computethe output of the system for any arbitrary input

Let h[n] denote the impulse response of an LTI discrete-time system we can compute y[n] for the input:


time system, we can compute y[n] for the input:

We can compute its outputs for each member of the input separately and add the individual outputs to determine y[n] (Linearity)

2008/3/17

37


Since the system is LTI

Because of the linearity property we get


Because of the linearity property we get


Now, any arbitrary input sequence x[n] can be expressed as a linear combination of delayed and advanced unitas a linear combination of delayed and advanced unit sample sequences in the form:

The response of the LTI system to an input x[k][n k]will be x[k]h[n k]

Hence, the response y[n] to an input


will beor

2008/3/17

38

Convolution Sum (1/3) The summation

is called the convolution sum of the sequences x[n] and h[n] and represented as

Properties of convolutionCommutative property:


Commutative property:

Associative property :

Distributive property :

Convolution Sum (2/3) Interpretation

Time-reverse h[k] to form h[-k][ ] [ ] Shift h[-k] to the right by n sampling periods if n > 0 (or

shift to the left by n sampling periods if n < 0) to formh[n-k]

Form the product v[k] = x[k]h[n-k] Sum all samples of v[k] to develop the n-th sample of

y[n] of the convolution sum


y[n] of the convolution sum

2008/3/17

39

Convolution Sum (3/3)

The computation of an output sample using the convolution sum is simply a sum of products p y p

Involves fairly simple operations such as additions, multiplications, and delays

We illustrate the convolution operation for the following two sequences:


The next several slides illustrate the convolution process

Convolution (1/12)


2008/3/17

40

Convolution (2/12)


Convolution (3/12)


2008/3/17

41

Convolution (4/12)


Convolution (5/12)


2008/3/17

42

Convolution (6/12)


Convolution (7/12)


2008/3/17

43

Convolution (8/12)


Convolution (9/12)


2008/3/17

44

Convolution (10/12)


Convolution (11/12)


2008/3/17

45

Convolution (12/12)


Computing Convolution Sum

In practice, if either the input or the impulse response is of finite length, the convolution sum can be used toof finite length, the convolution sum can be used to compute the output sample as it involves a finite sum of products

If both the input sequence and the impulse response sequence are of finite length, the output sequence is also of finite length

If both the input sequence and the impulse response


If both the input sequence and the impulse response sequence are of infinite length, convolution sum cannot be used to compute the output

2008/3/17

46

Convolution: Step by Step (1/4) Example - Develop the sequence y[n] generated by the

convolution of the sequences x[n] and h[n] shown below

As can be seen from the shifted time-reversed version {h[n k]} for n < 0 shown below for for any value of the


{h[n k]} for n < 0, shown below for , for any value of the sample index k, the k-th sample of either {x[k]} or {h[n k]} is zero

Convolution: Step by Step (2/4)

As a result, for n < 0, the product of the k-th samples of {x[k]} and {h[n k]} is always zero, and hence{x[k]} and {h[n k]} is always zero, and hence

y[n] = 0 for n < 0 Consider now the computation of y[0] The sequence {h[k]} is shown on the right The product sequence {x[k]h[k]} is plotted below which

has a single nonzero sample x[0]h[0] for k = 0


Thus y[0] = x[0]h[0] = 2

2008/3/17

47


For the computation of y[1], we shift {h[-k]} to the right byone sample period to form {h[1-k]} as shown belowone sample period to form {h[1 k]} as shown below

Hence, y[1] = x[0]h[1] + x[1]h[0] = 4 + 0 = 4


Similarly, y[2] = x[0]h[2] + x[1]h[1] + x[2]h[0] =1


Repeat the process, we obtain the following output


In general, if the lengths of the two sequences being convolved are M and N, then the sequence generated by the convolution is of length M + N 1

2008/3/17

48

Tabular Method of Convolution Sum Computation (1/2)

Can be used to convolve two finite-length sequences


Tabular Method of Convolution Sum Computation (2/2)

The method can also be applied to convolve two finite-length two-sided sequenceslength two sided sequences

In this case, a decimal point is first placed to the right of the sample with the time index n = 0 for each sequence

Next, convolution is computed ignoring the location of the decimal point

Finally, the decimal point is inserted according to the rules of conventional multiplication


of conventional multiplication The sample immediately to the left of the decimal point is

then located at the time index n = 0

2008/3/17

49

Simple Interconnection Schemes Two simple interconnection schemes are:

Cascade Connection

Parallel Connection


Cascade Connection (1/2) The ordering of the systems in the cascade has no effect on

the overall impulse responseA cascade connection of two stable systems is stable A cascade connection of two stable systems is stable

A cascade connection of two passive (lossless) systems is passive (lossless)

An application is in the development of an inverse system If the cascade connection satisfies the relation


then the LTI system h1[n] is said to be the inverse of h2[n] and vice-versa

2008/3/17

50

Cascade Connection (2/2) Example - Consider the discrete-time accumulator with

an impulse response [n] Its inverse system satisfy the condition

It follows from the above that h2[n] = 0 for n < 0 and


Thus the impulse response of the inverse system of the discrete-time accumulator is given by

(backward difference system)

Interconnection Schemes Consider the discrete-time system where


2008/3/17

51

Stability Condition of an LTI Discrete-Time System (1/3)

BIBO Stability Condition - A discrete-time system is BIBO stable if and only if the output sequence {y[n]} remains bo nded for all bo nded inp t seq ence { [n]}remains bounded for all bounded input sequence {x[n]}

An LTI discrete-time system is BIBO stable if and only if its impulse response sequence {h[n]} is absolutely summable, i.e.

Proof: Assume h[n] is a real sequence


Since the input sequence x[n] is bounded we have

therefore


Thus, S < implies y[n] By < , indicating that y[n] isalso boundedalso bounded

To prove the converse, assume y[n] is bounded, i.e., y[n] By

Consider the bounded input given byx[n] = sgn(h[n]) where sgn(c) = 1, for c 0, and

sgn(c) = 1 for c < 0


For this input, y[n] at n = 0 is

Therefore, if S = , then {y[n]} is not a bounded sequence

2008/3/17

52


Example - Consider a causal LTI discrete-time system with an impulse responsewith an impulse response

For this system


Therefore S < if || < 1 , for which the system is BIBO stable

If || = 1, the system is not BIBO stable

Causality Condition of an LTI Discrete-Time System (1/3)

Let x1 [n] and x2 [n] be two input sequences withx [n] = x [n] for n nx1[n] = x2[n] for n no

The corresponding output samples at of an LTI system with an impulse response {h[n]} are then given by


2008/3/17

53


If the LTI system is also causal, theny1[no] = y2[no]y1[no] y2[no]

As x1[n] = x2[n] for n no

This implies


As for x1[n] x2[n] the only way the condition

holds if h[k] = 0 for k < 0


An LTI discrete-time system is causal if and only if its impulse response {h[n]} is a causal sequenceimpulse response {h[n]} is a causal sequence

Example - The discrete-time accumulator defined by

is causal as it has a causal impulse response


Example - The factor-of-2 interpolator defined by

is noncausal as it has a noncausal impulse response

2008/3/17

54

Finite-Dimensional LTIDiscrete-Time Systems

An important subclass of LTI discrete-time systems is characterized by a linear constant coefficient difference yequation of the form

where {dk} and {pk} are constants characterizing the system The order of the system is given by max(N,M), which is the

order of the difference equation


order of the difference equation Suppose the system is causal, then the output y[n] can be

recursively computed using

provided d0 0

Classification of LTI Discrete-Time Systems (1/3)

Based on Impulse Response Length - If the impulse response h[n] is of finite length, i.e.,If the impulse response h[n] is of finite length, i.e.,

h[n] = 0 for n < N1 and n > N2, N1 < N2then it is known as a finite impulse response (FIR)

discrete-time system The convolution sum description here is


The output y[n] of an FIR LTI discrete-time system can be computed directly from the convolution sum as it is a finite sum of products (e.g., moving-average filter & interpolator)

2008/3/17

55


If the impulse response h[n] is of infinite length, then it is known as a infinite impulse response (IIR) discrete-time p p ( )system

Example - The discrete-time accumulator defined byy[n] = y[n 1] + x[n]

is an IIR system Example - The numerical integration formulas that are

used to numerically solve integrals of the form


used to numerically solve integrals of the form

can be characterized by the following 1st-order IIR system


Based on the Output Calculation Process - Nonrecursive System - Here the output can beNonrecursive System Here the output can be

calculated sequentially, knowing only the present and past input samples

Recursive System - Here the output computation involves past output samples in addition to the present and past input samples

Based on the Coefficients -


Based on the Coefficients - Real Discrete-Time System - The impulse response

samples are real valued Complex Discrete-Time System - The impulse response

samples are complex valued

2008/3/17

56

Correlation of Signals (1/4) There are applications where it is necessary to compare

one reference signal with one or more signalst d t i th i il it b t th i to determine the similarity between the pair

to determine additional information based on the similarity For example, in digital communications, the receiver has

to determine which particular sequence has been received by comparing the received signal with possible sequences that may be transmittedSi il l i d d li i h i d


Similarly, in radar and sonar applications, the received signal reflected from the target is a delayed (and even a distorted) version of the transmitted signal

The received signal is often corrupted by additive random noise, making signal detection more complicated

Correlation of Signals (2/4) A measure of similarity between a pair of energy signals,

x[n] and y[n], is given by the cross-correlation sequence

The parameter l called lag, indicates the time-shiftbetween the pair of signals

y[n] is said to be shifted by l samples to the right with respect to the reference sequence x[n] for positive


respect to the reference sequence x[n] for positive values of l

The ordering of the subscripts xy in rxy[l] specifies that x[n] is the reference sequence which remains fixed in time while y[n] is being shifted with respect to x[n]

2008/3/17

57

Correlation of Signals (3/4) If y[n] is made the reference signal and shift x[n] with

respect to y[n], then the corresponding cross-correlation seq ence is gi en bsequence is given by

The autocorrelation sequence of x[n] is given by


Note, , the energy of x[n] From the relation ryx[l] = rxy[l] it follows that rxx[l] = rxx[l],

implying that rxx[l] is an even function for real x[n]

Correlation of Signals (4/4) Rewrite the expression for the cross-correlation as

The cross-correlation of y[n] with the reference signal x[n] can be computed by processing x[n] with an LTI discrete-time system of impulse response y[n]


Likewise, the autocorrelation of x[n] can be computed by processing x[n] with an LTI discrete-time system of impulse response

2008/3/17

58

Properties of Autocorrelation andCross-correlation Sequences (1/3)

Consider two finite-energy sequences x[n] and y[n] The energy of the combined sequence ax[n]+y[n-l] isThe energy of the combined sequence ax[n] y[n l] is

also finite and nonnegative, i.e.,

Thus


where and rxx[0] = Ex > 0 and ryy[0] = Ey > 0 The above equation can be rewritten as

for any finite value of a


The matrix

is thus positive semidefinite

or, equivalently,

Th i lit id b d f th


The inequality provides an upper bound for the cross-correlation samples

If we set y[n] = x[n], then the inequality reduces to

[ ] [ ]0xx xx xr l r E =

2008/3/17

59


Thus, at zero lag (l = 0), the sample value of the autocorrelation sequence has its maximum valueq

Now consider the casey[n] = b x[n N]

where N is an integer and b > 0 is an arbitrary number In this case Therefore


Using the above result:

We get brxx[0] rxy[l] brxx[0]

Computation of Correlations (1/2)

Example - Consider the two finite-length sequences

x[n] = [1 3 2 1 2 1 4 4 2] y[n] = [2 1 4 1 2 3]x[n] = [1 3 2 1 2 1 4 4 2], y[n] = [2 1 4 1 2 3]


rxy[n] rxx[n]

2008/3/17

60

Computation of Correlations (2/2)

Example - The cross-correlation of x[n] and y[n] = x[n N] for N = 4]

Note: The peak of the cross-correlation is precisely the value of the delay N


rxy[n]

Normalized Forms of Correlation Normalized forms of autocorrelation and cross-

correlation are given by

Note: |xx[l]| 1 and |yy[l]| 1 independent of the range of values of x[n] and y[n]

The cross-correlation sequence for a pair of power signals, x[n] and y[n], is defined as


The autocorrelation sequence of a power signal x[n] is given by

2008/3/17

61

Normalized Forms of Correlation The cross-correlation sequence for a pair of periodic

signals of period N, and , is given by

The autocorrelation sequence of :

Both and are also periodic with a period N Let be a periodic signal corrupted by the random


noise d[n] resulting in the signal

which is observed for 0 n M 1 where M >> N

Normalized Forms of Correlation The autocorrelation of w[n] is given by

is a periodic sequence with a period N and hence


will have peaks at l = 0, N, 2N,... with the same amplitudes as l approaches M

As and d[n] are not correlated, samples of cross-correlation sequences and are likely to be very small relative to the amplitudes of

2008/3/17

62

Normalized Forms of Correlation The autocorrelation rdd [l] of d[n] will show a peak at l = 0

with other samples having rapidly decreasing amplitudes with increasing values of |l|

Hence, peaks of rww [l] for l > 0 are essentially due to the peaks of and can be used to determine whether is a periodic sequence and also its period N if the peaks occur at periodic intervals


Normalized Forms of Correlation Example - Determine the period of the sinusoidal

sequence x[n] = cos(0.25n), 0 n 95 corrupted by an additive uniformly distributed random noise of amplitude in the range [0.5,0.5]


rdd[l]rww[l]

Documents

Chapter 2 Discrete-Time Signals & Systems - 清華大學電 …cwlin/courses/dsp/notes/ch2_Mitr… · · 2008-03-172008/3/17 1 Discrete-Time Signals & Systems ... • Discrete-time