Part 9
Discrete-Time Linear Time-Invariant (LTI) Systems
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 395
Why Linear Time-Invariant (LTI) Systems?
In engineering, linear time-invariant (LTI) systems play a very important
role.
Very powerful mathematical tools have been developed for analyzing LTI
systems.
LTI systems are much easier to analyze than systems that are not LTI.
In practice, systems that are not LTI can be well approximated using LTI
models.
So, even when dealing with systems that are not LTI, LTI systems still play
an important role.
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 396
Section 9.1
Convolution
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 397
DT Convolution
The (DT) convolution of the sequences x and h, denoted x ∗ h, is defined
as the sequence
∞
x ∗ h(n) = ∑ x(k)h(n − k)
k=−∞
The convolution x ∗ h evaluated at the point n is simply a weighted sum of
elements of x, where the weighting is given by h time reversed and shifted
by n.
Herein, the asterisk symbol (i.e., “∗”) will always be used to denote
convolution, not multiplication.
As we shall see, convolution is used extensively in the theory of (DT)
systems.
In particular, convolution has a special significance in the context of (DT)
LTI systems.
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 398
Practical Convolution Computation
To compute the convolution
∞
x ∗ h(n) = ∑ x(k)h(n − k),
k=−∞
we proceed as follows:
1 Plot x(k) and h(n − k) as a function of k .
2 Initially, consider an arbitrarily large negative value for n. This will result in
h(n − k) being shifted very far to the left on the time axis.
3 Write the mathematical expression for x ∗ h(n).
4 Increase n gradually until the expression for x ∗ h(n) changes form. Record
the interval over which the expression for x ∗ h(n) was valid.
5 Repeat steps 3 and 4 until n is an arbitrarily large positive value. This
corresponds to h(n − k) being shifted very far to the right on the time axis.
6 The results for the various intervals can be combined in order to obtain an
expression for x ∗ h(n) for all n.
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 399
Properties of Convolution
The convolution operation is commutative. That is, for any two sequences
x and h,
x ∗ h = h ∗ x
The convolution operation is associative. That is, for any sequences x, h1 ,
and h2 ,
(x ∗ h1 ) ∗ h2 = x ∗ (h1 ∗ h2 )
The convolution operation is distributive with respect to addition. That is,
for any sequences x, h1 , and h2 ,
x ∗ (h1 + h2 ) = x ∗ h1 + x ∗ h2
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 400
Representation of Sequences Using Impulses
For any sequence x,
∞
x(n) = ∑ x(k)δ(n − k) = x ∗ δ(n)
k=−∞
Thus, any sequence x can be written in terms of an expression involving δ.
Moreover, δ is the convolutional identity. That is, for any sequence x,
x ∗ δ = x
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 401
Circular Convolution
The convolution of two periodic sequences is usually not well defined.
This motivates an alternative notion of convolution for periodic sequences
known as circular convolution.
The circular convolution (also known as the DT periodic convolution) of
the N -periodic sequences x and h, denoted x ~ h, is defined as
N−1
x ~ h(n) = ∑ x(k)h(n − k) = ∑ x(k)h(mod(n − k, N)),
k=〈N〉 k=0
where mod(a, b) is the remainder after division when a is divided by b.
The circular convolution and (linear) convolution of the N -periodic
sequences x and h are related as follows:
∞
x ~ h(n) = x0 ∗ h(n) where x(n) = ∑ x0 (n − kN)
k=−∞
(i.e., x0 (n) equals x(n) over a single period of x and is zero elsewhere).
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 402
Section 9.2
Convolution and LTI Systems
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 403
Impulse Response
The response h of a system H to the input δ is called the impulse
response of the system (i.e., h = Hδ).
For any LTI system with input x, output y, and impulse response h, the
following relationship holds:
y = x ∗ h
In other words, a LTI system simply computes a convolution.
Furthermore, a LTI system is completely characterized by its impulse
response.
That is, if the impulse response of a LTI system is known, we can
determine the response of the system to any input.
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 404
Step Response
The response s of a system H to the input u is called the step response of
the system (i.e., s = Hu).
The impulse response h and step response s of a system are related as
h(n) = s(n) − s(n − 1)
Therefore, the impulse response of a system can be determined from its
step response by (first-order) differencing.
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 405
Block Diagram of LTI Systems
Often, it is convenient to represent a (DT) LTI system in block diagram
form.
Since such systems are completely characterized by their impulse
response, we often label a system with its impulse response.
That is, we represent a system with input x, output y, and impulse
response h, as shown below.
x y
h
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 406
Interconnection of LTI Systems
The series interconnection of the LTI systems with impulse responses h1
and h2 is the LTI system with impulse response h = h1 ∗ h2 . That is, we
have the equivalences shown below.
x y x y
h1 h2 ≡ h1 ∗ h2
x y x y
h1 h2 ≡ h2 h1
The parallel interconnection of the LTI systems with impulse responses
h1 and h2 is a LTI system with the impulse response h = h1 + h2 . That is,
we have the equivalence shown below.
x y
h1 +
x y
≡ h1 + h2
h2
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 407
Section 9.3
Properties of LTI Systems
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 408
Memory
A LTI system with impulse response h is memoryless if and only if
h(n) = 0 for all n 6= 0
That is, a LTI system is memoryless if and only if its impulse response h is
of the form
h(n) = Kδ(n),
where K is a complex constant.
Consequently, every memoryless LTI system with input x and output y is
characterized by an equation of the form
y = x ∗ (Kδ) = Kx
(i.e., the system is an ideal amplifier).
For a LTI system, the memoryless constraint is extremely restrictive (as
every memoryless LTI system is an ideal amplifier).
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 409
Causality
A LTI system with impulse response h is causal if and only if
h(n) = 0 for all n < 0
(i.e., h is a causal sequence).
It is due to the above relationship that we call a sequence x, satisfying
x(n) = 0 for all n < 0,
a causal sequence.
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 410
Invertibility
The inverse of a LTI system, if such a system exists, is a LTI system.
Let h and hinv denote the impulse responses of a LTI system and its (LTI)
inverse, respectively. Then,
h ∗ hinv = δ
Consequently, a LTI system with impulse response h is invertible if and
only if there exists a sequence hinv such that
h ∗ hinv = δ
Except in simple cases, the above condition is often quite difficult to test.
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 411
BIBO Stability
A LTI system with impulse response h is BIBO stable if and only if
∞
∑ h(n) < ∞
n=−∞
(i.e., h is absolutely summable).
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 412
Eigensequences of LTI Systems
As it turns out, every complex exponential is an eigensequence of all LTI
systems.
For a LTI system H with impulse response h,
Hzn (n) = H(z)zn ,
where z is a complex constant and
∞
H(z) = ∑ h(n)z−n
n=−∞
zn
That is, is an eigensequence of a LTI system and H(z) is the
corresponding eigenvalue.
We refer to H as the system function (or transfer function) of the
system H.
From above, we can see that the response of a LTI system to a complex
exponential is the same complex exponential multiplied by the complex
factor H(z).
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 413
Representation of Sequences Using Eigensequences
Consider a LTI system with input x, output y, and system function H .
Suppose that the input x can be expressed as the linear combination of
complex exponentials
x(n) = ∑ ak znk ,
k
where the ak and zk are complex constants.
Using the fact that complex exponentials are eigenfunctions of LTI
systems, we can conclude
y(n) = ∑ ak H(zk )znk
k
Thus, if an input to a LTI system can be expressed as a linear combination
of complex exponentials, the output can also be expressed as linear
combination of the same complex exponentials.
The above formula can be used to determine the output of a LTI system
from its input in a way that does not require convolution.
Copyright © 2013–2020 Michael D. Adams Signals and Systems Edition 3.0 414