Signals and Systems

Lecture Notes
By
Mohamed-Yahia Dabbagh

2020

Note: All rights for these lecture notes are reserved by the author. The lecture notes are intended
for the exclusive use and learning purposes by teaching assistants and students who are enrolled
in the course ECE 207 at the University of Waterloo. Distribution of these lecture notes in any
form is not allowed.

© M.-Y. Dabbagh
 Chapter 3
Time-Domain Analysis of Discrete-Time Systems

3.1 Introduction

Systems whose inputs and outputs are discrete-time signals are called discrete-
time (DT) systems. For example, digital circuits and digital computers are DT
systems.

𝑥[𝑛] Discrete-Time (DT) 𝑦[𝑛]

System
DT DT

Figure 1

A discrete-time signal is a sequence of numbers. A DT system processes a

sequence of numbers or signal 𝑥[𝑛] at its input and produces another sequence
or signal 𝑦[𝑛] at its output, where 𝑛 is an integer.

Discrete-time signals can be naturally discrete such as census data and stock
prices. They are also obtained by sampling continuous-time (CT) signals such as
voice or image. There are several advantages in processing CT signals by DT
systems. This is achieved by converting the CT signal 𝑥(𝑡) to a DT signal 𝑥[𝑛]
which is processed by the DT system to produce the output signal 𝑦[𝑛] which is
converted to a CT signal 𝑦(𝑡). The conversion of a CT signal to a DT signal is
achieved by a system called a continuous-to-discrete (C/D) converter, also called
an analog-to-digital (A/D) converter. The conversion of a DT signal to a CT

© M.-Y. Dabbagh 3-1

 signal is achieved by a system called a discrete-to-continuous (D/C) converter,
also called a digital-to-analog (D/A) converter.

𝑥(𝑡) A/D 𝑥[𝑛] 𝑦[𝑛] D/A 𝑦(𝑡)

DT System
Converter Converter
C/D +D/C
Figure 2

3.2 Examples of Discrete-Time Systems

Several DT systems are discussed where their system models are derived and their
realizations are shown.

Example: (Bank Account)

Let: 𝑦[𝑛] = account balance at time n (beginning of month n)

𝑥[𝑛] = deposit (or withdrawal) during month n

𝑟 = monthly interest rate

The balance at the beginning of next month (𝑛 + 1) is:

𝑦[𝑛 + 1] = 𝑦[𝑛] + 𝑟𝑦[𝑛] + 𝑥[𝑛]

⇒ 𝑦[𝑛 + 1] − (1 + 𝑟)𝑦[𝑛] = 𝑥[𝑛]

This difference equation (DE) is the system model. It is in advance form since
the signals are in terms of the current or future time, here the time indices 𝑛 and
𝑛 + 1. The difference equation can be re-written differently by replacing 𝑛 by
(𝑛 − 1):

© M.-Y. Dabbagh 3-2

 𝑦[𝑛] − (1 + 𝑟)𝑦[𝑛 − 1] = 𝑥[𝑛 − 1]

This is the difference equation in delay form since the signals are in terms of the
current or past time, here the time indices 𝑛 and 𝑛 − 1. As we will see later, the
advance form is more convenient for mathematical analysis while the delay form
is more convenient for system realization (in software or hardware).

 Realization Block Diagram

Difference equations with constant coefficients can be realized using three basic
digital circuit elements: a time delay (D), a digital multiplier, and a digital adder.
The delay can be implemented using flip-flops.

For example, the last difference equation in delay form can be rewritten:

𝑦[𝑛] = 𝑎𝑦[𝑛 − 1] + 𝑥[𝑛 − 1]

where 𝑎 = 1 + 𝑟

This equation can be realized as shown in Figure 3 below.

𝑥[𝑛] 𝑥[𝑛 − 1] 𝑦[𝑛]

D +
+
+
𝑦[𝑛 − 1]
a D

Figure 3

© M.-Y. Dabbagh 3-3

 Example: (Digital Differentiator)
𝑑𝑥(𝑡)
The continuous-time differentiation system 𝑦(𝑡) = can be implemented
𝑑𝑡

approximately by a discrete-time differentiator. The CT signals 𝑥(𝑡) and 𝑦(𝑡) are

sampled every T seconds which gives 𝑥[𝑛] = 𝑥(𝑛𝑇) and 𝑦[𝑛] = 𝑦(𝑛𝑇).

The differentiation can be written as:

𝑥(𝑛𝑇) − 𝑥((𝑛 − 1)𝑇)

𝑦(𝑛𝑇) = lim
𝑇→0 𝑇

If 𝑥(𝑡) is slowly changing and T is relatively small, this can be approximated by:

1
𝑦[𝑛] ≅ [𝑥[𝑛] − 𝑥[𝑛 − 1]]
𝑇

This difference equation can be realized as shown in Figure 4.

𝑥[𝑛] 𝑦[𝑛]
+
- 1/𝑇

Figure 4

© M.-Y. Dabbagh 3-4

 Example: (Digital Integrator)
𝑡
The CT system described by 𝑦(𝑡) = ∫−∞ 𝑥(𝜏)𝑑𝜏 can be implemented by a digital
integrator. For 𝑡 = 𝑛𝑇, where 𝑇 is the sampling interval:
𝑛

𝑦(𝑛𝑇) = lim ∑ 𝑥(𝑘𝑇) 𝑇

𝑇→0
𝑘=−∞

For small 𝑇:
𝑛

𝑦[𝑛] = 𝑇 ∑ 𝑥[𝑘]
𝑘=−∞

This difference equation is in non-recursive form since it involve the current value
of 𝑦[𝑛] only. This form is not easy to implement since it involves the storage of
all current and past values of 𝑥[𝑛]. The last equation can be re-written as:

𝑛−1

𝑦[𝑛] = 𝑇 ∑ 𝑥[𝑘] + 𝑇𝑥[𝑛]

⏟𝑘=−∞
𝑦[𝑛−1]

⇒ 𝑦[𝑛] = 𝑦[𝑛 − 1] + 𝑇𝑥[𝑛]

This is the difference equation in recursive form, whose computation and

realization are more efficient, as shown in Figure 5.

𝑥[𝑛] 𝑦[𝑛]
T +

Figure 5

© M.-Y. Dabbagh 3-5

 Example: (Differential Equation)

𝑑𝑦
A CT system described by the differential equation + 𝑎𝑦(𝑡) = 𝑥(𝑡) an be
𝑑𝑡

approximately implemented by a DT system. For 𝑡 = 𝑛𝑇, the equation can be

written as:

𝑦[𝑛] − 𝑦[𝑛 − 1]
lim + 𝑎𝑦[𝑛] = 𝑥[𝑛]
𝑇→0 𝑇

For small 𝑇:

1 + 𝑎𝑇 1
𝑦[𝑛] − 𝑦[𝑛 − 1] = 𝑥[𝑛]
𝑇 𝑇

or

𝑦[𝑛] + 𝛼𝑦[𝑛 − 1] = 𝛽𝑥[𝑛]

−1 𝑇
where 𝛼 = and 𝛽 = . This is a first order difference equation with
1+𝑎𝑇 1+𝑎𝑇

constant coefficients.

Remark: From above examples, we notice that all of them are described by
difference equations with constant coefficients. Most practical systems are of this
class. Therefore, in this course, this class of systems is considered.

© M.-Y. Dabbagh 3-6

3.3 Linearity, Time-Invariance, and Causality

The properties of linearity, time-invariance, and causality for DT systems are the
same as for CT systems.

Examples:

𝑦[𝑛] = 𝑥[𝑛 − 2] LTI, Causal

𝑦[𝑛] = 𝑥[𝑛 + 2] LTI, Non-causal

𝑦[𝑛] = 𝑒 𝑛+1 𝑥[𝑛] Linear, Time-varying, Causal

𝑦[𝑛] = ∑𝑛𝑘=0 𝑥[𝑘] , 𝑛 ≥ 0 LTI, Causal

𝑦[𝑛 + 1] + ⏟ [𝑛 + 2] ⏟
𝑛 𝑦[𝑛] = 𝑥 ⏟ . 𝑥[𝑛]
time non non−
varying causal linear

Nonlinear, Time-varying, Non-causal

𝑦[𝑛 + 2] − 2𝑦[𝑛 + 1] + 𝑦[𝑛] = 3𝑥[𝑛 + 2] + 𝑥[𝑛 + 1] − 2𝑥[𝑛]

This last system is a 2nd order difference equation (DE) with constant coefficients
which is LTI and causal.

Remark: Systems described by difference equations with constant coefficients

are linear and time-invariant (LTI).

Remark: (Order of a DE) The order of a DE is the highest time difference 𝑁 of

the output 𝑦[𝑛] or the highest time difference 𝑀 of the input 𝑥[𝑛], whichever is
higher, when the DE is written in advance form. For example, the system:

𝑦[𝑛 + 2] − 𝑦[𝑛 + 1] + 𝑦[𝑛] = 𝑥[𝑛 + 3] + 𝑥[𝑛 + 2]

has 𝑁 = 2 and 𝑀 = 3. Therefore, the order of the system is 3.

© M.-Y. Dabbagh 3-7

 Remark: (Causality Condition) A system described by a DE with constant
coefficients is causal when 𝑀 ≤ 𝑁. Otherwise, it is non-causal. For example,
the system described by the last DE is non-causal since 𝑀 = 3 > 𝑁 = 2.

Example: (Order of a system)

𝑦[𝑛 + 1] + 𝑦[𝑛 − 1] = 𝑥[𝑛 + 1] + 𝑥[𝑛]

This is not of order 1 since it is not written in advance form. Replacing 𝑛 by 𝑛 +

1, we get the advance form:

𝑦[𝑛 + 2] + 𝑦[𝑛] = ⏟
⏟ 𝑥[𝑛 + 2] + 𝑥[𝑛 + 1]
order 𝑁=2 order 𝑀=2

So, the order of the system is 𝑀 = 𝑁 = 2. This system is LTI and causal.

© M.-Y. Dabbagh 3-8

3.4 Discrete-Time System Equations

A large class of practical LTI systems are described by difference equations (DEs)
with constant coefficients. The general difference equation in advance form is
given by:

𝑦[𝑛 + 𝑁] + 𝑎𝑁−1 𝑦[𝑛 + 𝑁 − 1] + ⋯ + 𝑎0 𝑦[𝑛]

= 𝑏𝑀 𝑥[𝑛 + 𝑀] + 𝑏𝑀−1 𝑥[𝑛 + 𝑀 − 1] + ⋯ + 𝑏0 𝑥[𝑛]

For a causal system, we must have 𝑀 ≤ 𝑁. For generality, we set 𝑀 = 𝑁.

𝑦[𝑛 + 𝑁] + 𝑎𝑁−1 𝑦[𝑛 + 𝑁 − 1] + ⋯ + 𝑎0 𝑦[𝑛]

= 𝑏𝑁 𝑥[𝑛 + 𝑁] + 𝑏𝑁−1 𝑥[𝑛 + 𝑁 − 1] + ⋯ + 𝑏0 𝑥[𝑛]

Remark: Above equation is normalized in the sense that 𝑎𝑁 = 1.

Remark: Some of the coefficients can be zero. However, if 𝑎0 and 𝑏0 are both
zero at the same time, then the time index 𝑛 can be replaced by 𝑛 − 1 which results
in a DE with order (𝑁 − 1) and at least one non-zero new coefficients 𝑎0 and 𝑏0 .
This means the terms of 𝑦[𝑛] or 𝑥[𝑛] or both must exist for the equation to be in
standard form.

 Advance Operator Notation (𝑬 − 𝑶𝒑𝒆𝒓𝒂𝒕𝒐𝒓):

𝐸𝑥[𝑛] = 𝑥[𝑛 + 1]

𝐸 2 𝑥[𝑛] = 𝑥[𝑛 + 2]

𝐸 𝑁 𝑥[𝑛] = 𝑥[𝑛 + 𝑁]

Using the E-operator, the general difference equation can be rewritten as:

© M.-Y. Dabbagh 3-9

 (𝐸 𝑁 + 𝑎𝑁−1 𝐸 𝑁−1 + ⋯ + 𝑎1 𝐸 + 𝑎0 )𝑦[𝑛]
= (𝑏𝑁 𝐸 𝑁 + 𝑏𝑁−1 𝐸 𝑁−1 + ⋯ + 𝑏1 𝐸 + 𝑏0 )𝑥[𝑛]

or

𝑄(𝐸)𝑦[𝑛] = 𝑃(𝐸)𝑥[𝑛]

where

𝑄(𝐸) = 𝐸 𝑁 + 𝑎𝑁−1 𝐸 𝑁−1 + ⋯ + 𝑎1 𝐸 + 𝑎0

𝑃(𝐸) = 𝑏𝑁 𝐸 𝑁 + 𝑏𝑁−1 𝐸 𝑁−1 + ⋯ + 𝑏1 𝐸 + 𝑏0

Remark: There are two causes for the system response 𝑦[𝑛], namely the
existence of initial conditions (ICs) and the application of the input 𝑥[𝑛]. Since
the system is linear, the total system response can be written as:

𝑦[𝑛] = 𝑦𝑧𝑖 [𝑛] + 𝑦𝑧𝑠 [𝑛] , 𝑛≥0

where

𝑦𝑧𝑖 [𝑛] = Zero-input response due to IC’s only (with 𝑥[𝑛] = 0, for all n)

𝑦𝑧𝑠 [𝑛] = Zero-state response due to input 𝑥[𝑛], 𝑛 ≥ 0 (with IC’s=0)

Remark: There are two methods for solving the DE (finding 𝑦[𝑛]):

– Recursive (Iterative) Solution

– Analytic Solution

© M.-Y. Dabbagh 3-10

 Recursive (Iterative) Solution

This method is based on rewriting the difference equation in delay form:

𝑦[𝑛] = −𝑎𝑁−1 𝑦[𝑛 − 1] − 𝑎𝑁−2 𝑦[𝑛 − 2] − ⋯ − 𝑎0 𝑦[𝑛 − 𝑁] + 𝑏𝑁 𝑥[𝑛]

+ 𝑏𝑁−1 𝑥[𝑛 − 1] + ⋯ + 𝑏0 𝑥[𝑛 − 𝑁]

This equation allows us to compute the system response point-by-point, i.e.

𝑦[0], 𝑦[1], 𝑦[2], ⋯ , etc. Also, it allows us to compute each of the zero-input
response 𝑦𝑧𝑖 [𝑛] and the zero-state response 𝑦𝑧𝑠 [𝑛] separately, or the total response
𝑦[𝑛] directly. Given the initial conditions {𝑦[−1], 𝑦[−2], ⋯ , 𝑦[−𝑁]} with
𝑥[𝑛] = 0 for 𝑛 ≥ 0, the zero-input response 𝑦𝑧𝑖 [𝑛] is computed. While, given the
input 𝑥[𝑛] for 𝑛 ≥ 0 with ICs=0, the zero-state response 𝑦𝑧𝑠 [𝑛] is computed.
The total response 𝑦[𝑛] can be computed when both the input and ICs are applied
at the same time.

Example: A system is described by (𝐸 + 0.5)𝑦[𝑛] = 𝑥[𝑛] with 𝑦[−1] = 1 and

𝑥[𝑛] = 𝑛. 𝑢[𝑛], 𝑛 ≥ 0. Find the total response 𝑦[𝑛]?

Solution:

Rewrite the difference equation in delay form:

1
𝑦[𝑛] = − 𝑦[𝑛 − 1] + 𝑥[𝑛 − 1]
2
Compute 𝑦[𝑛] point-by-point by iteration:
0
1 ⏞ 1 1
𝑛 = 0 ⇒ 𝑦[0] = − 𝑦[−1] + 𝑥[−1] =− −0=−
2 2 2

1 1 1 1
𝑛 = 1 ⇒ 𝑦[1] = − 𝑦[0] + 𝑥[0] = − (− ) + 0 =
2 2 2 4

© M.-Y. Dabbagh 3-11

 1 1 1 7
𝑛=2 ⇒ 𝑦[2] = − 𝑦[1] + 𝑥[1] = − ( ) + 1 =
2 2 4 8

1 1 7 25
𝑛=3 ⇒ 𝑦[3] = − 𝑦[2] + 𝑥[2] = − ( ) + 2 =
2 2 8 16

25/16
𝑦[𝑛]

1 7/8

1/4
…

-1 0 1 2 3 n

-1/2

Figure 6

Remark: Iterative solution is useful for computer computation or for finding few
initial values {𝑦[0], 𝑦[1], ⋯ }. It is difficult to apply for large values of 𝑛, say
finding 𝑦[1000] by hand computation without a computer. The other major
disadvantage is that it does not reveal the behavior of the system, such as the
stability of the system.

© M.-Y. Dabbagh 3-12

 Analytic Solution (Closed-Form Solution)

Given an LTI system described by the difference equation 𝑄(𝐸)𝑦[𝑛] = 𝑃(𝐸)𝑥[𝑛]

with initial conditions {𝑦[−1], 𝑦[−2], ⋯ , 𝑦[−𝑁]} and an input 𝑥[𝑛], 𝑛 ≥ 0, the
system response 𝑦[𝑛] can be decomposed into two components by linearity:

𝑦[𝑛] = 𝑦𝑧𝑖 [𝑛] + 𝑦𝑧𝑠 [𝑛], 𝑛≥0

where

𝑦𝑧𝑖 [𝑛] = Zero-input response due to ICs only (with 𝑥[𝑛] = 0, for all n)

𝑦𝑧𝑠 [𝑛] = Zero-state response due to the input 𝑥[𝑛], 𝑛 ≥ 0 (with ICs=0)

Remark: The analytic solution for DT systems is very similar to CT systems.

Each of the system response components are discussed in the following sections.

© M.-Y. Dabbagh 3-13

3.5 Zero-Input Response

Given the DT system 𝑄(𝐸)𝑦[𝑛] = 𝑃(𝐸)𝑥[𝑛], the zero-input response 𝑦𝑧𝑖 [𝑛] is
the system response due to non-zero initial conditions with 𝑥[𝑛] = 0 for all n, i.e.

𝑄(𝐸)𝑦𝑧𝑖 [𝑛] = 0, 𝑛≥0

or

(𝐸 𝑁 + 𝑎𝑁−1 𝐸 𝑁−1 + ⋯ + 𝑎1 𝐸 + 𝑎0 )𝑦𝑧𝑖 [𝑛] = 0

Assume the solution:

𝑦𝑧𝑖 [𝑛] = 𝐶𝑟 𝑛

where 𝐶 and 𝑟 are constants. Applying the E-operator:

𝐸𝑦𝑧𝑖 [𝑛] = 𝐶𝑟 𝑛+1 = 𝑟 𝐶𝑟 𝑛

𝐸 2 𝑦𝑧𝑖 [𝑛] = 𝐶𝑟 𝑛+2 = 𝑟 2 𝐶𝑟 𝑛

𝐸 𝑁 𝑦𝑧𝑖 [𝑛] = 𝐶𝑟 𝑛+𝑁 = 𝑟 𝑁 𝐶𝑟 𝑛

Substituting in above DE:

(𝑟 𝑁 + 𝑎𝑁−1 𝑟 𝑁−1 + ⋯ + 𝑎1 𝑟 + 𝑎0 ) 𝐶
⏟ 𝑟⏟𝑛 = 0
≠0 ≠0

⇒ 𝑟⏟𝑁 + 𝑎𝑁−1 𝑟 𝑁−1 + ⋯ + 𝑎1 𝑟 + 𝑎0 = 0

𝑄(𝑟)=charectristic polynomial

This is the characteristic equation that can be factored in the product form:

𝑄(𝑟) = (𝑟 − 𝑟1 )(𝑟 − 𝑟2 ) ⋯ (𝑟 − 𝑟𝑁 ) = 0

© M.-Y. Dabbagh 3-14

 where {𝑟1 , 𝑟2 , ⋯ , 𝑟𝑁 } are the characteristic values, characteristic roots,
eigenvalues, or natural frequencies of the system.

 Case 1 (Distinct Roots)

When the roots {𝑟1 , 𝑟2 , ⋯ , 𝑟𝑁 } are distinct, the characteristic modes of the system
are {𝑟1𝑛 , 𝑟2𝑛 , ⋯ , 𝑟𝑁𝑛 }. The system has 𝑁 possible solutions, and the sum is the
general solution:

𝑦𝑧𝑖 [𝑛] = 𝐶1 𝑟1𝑛 + 𝐶2 𝑟2𝑛 + ⋯ + 𝐶𝑁 𝑟𝑁𝑛 = ∑ 𝐶𝑖 𝑟𝑖𝑛

𝑖=1

 Case 2 (Repeated Roots)

When a root 𝑟𝑖 is repeated m times, the corresponding characteristic modes are

{𝑟𝑖𝑛 , 𝑛𝑟𝑖𝑛 , 𝑛2 𝑟𝑖𝑛 , ⋯ , 𝑛𝑚−1 𝑟𝑖𝑛 }, and the general solution for this root is:

𝑦𝑧𝑖 (𝑡) = ∑ 𝐶𝑘 𝑛𝑘−1 𝑟𝑖𝑛

𝑘=1

 Case 3 (Complex Roots)

For real systems (real coefficients of DE), complex roots occur in pairs of
conjugate; i.e. if 𝑟1 = 𝛼 + 𝑗𝛽 = |𝑟|𝑒 𝑗Ω is a root of the system, then 𝑟2 = 𝑟1∗ =
𝛼 − 𝑗𝛽 = |𝑟|𝑒 −𝑗Ω is also a root. For these two distinct roots, the solution is:

𝑦𝑧𝑖 (𝑡) = 𝐶1 𝑟1𝑛 + 𝐶2 𝑟2𝑛 = 𝐶1 |𝑟|𝑛 𝑒 𝑗Ωn + 𝐶2 |𝑟|𝑛 𝑒 −𝑗Ωn

Also, for real systems, 𝑦𝑧𝑖 [𝑛] must be real, for which the constants 𝐶1 and 𝐶2
𝐶 𝐶
must be complex conjugates, i.e. if 𝐶1 = 𝑒 𝑗𝜃 , then 𝐶2 = 𝐶1∗ = 𝑒 −𝑗𝜃 . And,
2 2

above complex form of the solution can be written in the real form:

© M.-Y. Dabbagh 3-15

 𝐶 𝑛 𝑗(Ωn+𝜃)
𝑦𝑧𝑖 [𝑛] = |𝑟| [𝑒 + 𝑒 −𝑗(Ωn+𝜃) ]
2

⇒ 𝑦𝑧𝑖 [𝑛] = 𝐶|𝑟|𝑛 cos(Ωn + 𝜃) , 𝑛 ≥ 0 (Real Form)

where 𝐶 and 𝜃 are constants found from ICs.

Example: A system is described by:

𝑦[𝑛 + 2] + 0.3𝑦[𝑛 + 1] − 0.1𝑦[𝑛] = 𝑥[𝑛 + 2] + 2𝑥[𝑛]

with initial conditions 𝑦[−1] = 1, 𝑦[−2] = −7. Find 𝑦𝑧𝑖 [𝑛]?

Solution:

Characteristic equation: 𝑄(𝑟) = 𝑟 2 + 0.3𝑟 − 0.1 = (𝑟 − 0.2)(𝑟 + 0.5) = 0

Roots: 𝑟1 = 0.2, 𝑟2 = −0.5

Modes: (0.2)𝑛 , (−0.5)𝑛

⇒ 𝑦𝑧𝑖 [𝑛] = [𝐶1 (0.2)𝑛 + 𝐶2 (−0.5)𝑛 ]𝑢[𝑛]

To find the constants 𝐶1 and 𝐶2 we compute 𝑦[0] and 𝑦[1] from the system
equation, iteratively:

𝑛 = −2 ⇒ 𝑦𝑧𝑖 [0] = −0.3 𝑦[−1]

⏟ + 0.1 𝑦[−2]
⏟ + 𝑥[0]
⏟ + 2 𝑥[−2]
⏟ = −1
1 −7 0 0

𝑛 = −1 ⇒ 𝑦𝑧𝑖 [1] = −0.3 𝑦[0]

⏟ + 0.1 𝑦[−1]
⏟ + 𝑥[1]
⏟ + 2 𝑥[−1]
⏟ = 0.4
−1 1 0 0

Substituting in 𝑦𝑧𝑖 [𝑛]:

𝑦𝑧𝑖 [0] = 𝐶1 + 𝐶2 = −1

𝑦𝑧𝑖 [1] = 𝐶1 (0.2) + 𝐶2 (−0.5) = 0.4

© M.-Y. Dabbagh 3-16

 1 6
The solution of these two linear equations is: 𝐶1 = − , 𝐶2 = −
7 7

1 6
⇒ 𝑦𝑧𝑖 [𝑛] = [− (0.2)𝑛 − (−0.5)𝑛 ] 𝑢[𝑛]
7 7

Remark: The initial conditions 𝒚[−𝟏] and 𝒚[−𝟐] are given on the total
response. However, the zero-state response 𝒚𝒛𝒔 [𝒏] = 𝟎 for 𝒏 < 𝟎 since 𝒙[𝒏] =
𝟎 for 𝒏 < 𝟎. Therefore, we have 𝒚[−𝟏] = 𝒚𝒛𝒊 [−𝟏] and 𝒚[−𝟐] = 𝒚𝒛𝒊 [−𝟐].
So, the constants 𝑪𝟏 and 𝑪𝟐 can be found from 𝒚𝒛𝒊 [𝒏] equation by setting
𝒏 = −𝟏 and 𝒏 = −𝟐, and ignoring 𝒖[𝒏]:

𝑦𝑧𝑖 [−1] = 𝐶1 (0.2)−1 + 𝐶2 (−0.5)−1 = 𝑦[−1] = 1

{ }
𝑦𝑧𝑖 [−2] = 𝐶1 (0.2)−2 + 𝐶2 (−0.5)−2 = 𝑦[−2] = −7

1 6
Solution of these equations gives {𝐶1 = − , 𝐶2 = − } which are the same as we
7 7

found before. This method of finding 𝐶1 and 𝐶2 is used in the textbook.

Example: A system is described by:

𝑦[𝑛 + 2] − 4𝑦[𝑛 + 1] + 4𝑦[𝑛] = 3𝑥[𝑛 + 1]

with ICs: 𝑦[−1] = 1, 𝑦[−2] = 0. Find 𝑦𝑧𝑖 [𝑛]?

Solution:

Characteristic equation: 𝑄(𝑟) = 𝑟 2 − 4𝑟 + 4 = (𝑟 − 2)2 = 0

Roots: 𝑟1 = 𝑟2 = 2 (repeated)

⇒ 𝑦𝑧𝑖 [𝑛] = [𝐶1 (2)𝑛 + 𝐶2 𝑛(2)𝑛 ]𝑢[𝑛]

Compute 𝐶1 and 𝐶2 :

© M.-Y. Dabbagh 3-17

 𝑦[−1] = 1 = 𝑦𝑧𝑖 [−1] = 𝐶1 (2)−1 + 𝐶2 (−1)(2)−1

𝑦[−2] = 0 = 𝑦𝑧𝑖 [−2] = 𝐶1 (2)−2 + 𝐶2 (−2)(2)−2

Solving these two equations gives: 𝐶1 = 4, 𝐶2 = 2

⇒ 𝑦𝑧𝑖 [𝑛] = [4(2)𝑛 + 2𝑛(2)𝑛 ]𝑢[𝑛]

Example: A system is described by:

𝑦[𝑛 + 2] + 4𝑦[𝑛] = 2𝑥[𝑛]

with ICs 𝑦[−1] = 0, 𝑦[−2] = 1. Find 𝑦𝑧𝑖 [𝑛]?

Solution:

Characteristic equation: 𝑄(𝑟) = 𝑟 2 + 4 = (𝑟 − 𝑗2)(𝑟 + 𝑗2) = 0

𝜋 𝜋
Roots: 𝑟1 = 𝑗2 = 2𝑒 𝑗 2 , 𝑟2 = −𝑗2 = 2𝑒 −𝑗 2

𝜋
⇒ 𝑦𝑧𝑖 [𝑛] = 𝐶(2)𝑛 cos ( 𝑛 + 𝜃) 𝑢[𝑛]
2

To find 𝐶 and 𝜃, compute 𝑦[0] and 𝑦[1] first from the system equation by
iteration:

𝑛 = −2 ⇒ 𝑦[0] = −4 𝑦[−2]
⏟ + 2 𝑥[−2]
⏟ = −4
1 0

𝑛 = −1 ⇒ 𝑦[1] = −4 𝑦[−1]
⏟ + 2 𝑥[−1]
⏟ =0
0 0

Substituting in 𝑦𝑧𝑖 [𝑛]:

𝜋
𝑦[0] = 𝑦𝑧𝑖 [0] = 𝐶(2)0 cos ( 0 + 𝜃) = 𝐶 cos(𝜃) = −4
2

© M.-Y. Dabbagh 3-18

 𝜋 𝜋
𝑦[1] = 𝑦𝑧𝑖 [1] = 𝐶(2)1 cos ( (1) + 𝜃) = 2𝐶 cos ( + 𝜃) = 0
2 ⏟ 2
− sin(𝜃)

Solving last two equations gives: 𝜃 = 0, 𝐶 = −4

𝜋
⇒ 𝑦𝑧𝑖 [𝑛] = [−4(2)𝑛 cos ( 𝑛)] 𝑢[𝑛]
2

Remark: Another possible solution for the constants: 𝜃 = 𝜋, 𝐶 = +4 .

Remark: It is possible to compute the constants 𝐶 and 𝜃 from the equations

𝑦[−1] = 𝑦𝑧𝑖 [−1] = 0 and 𝑦[−2] = 𝑦𝑧𝑖 [−2] = 1 directly without computing
𝑦[0] and 𝑦[1] by iteration.

© M.-Y. Dabbagh 3-19

3.6 The Unit Impulse Response

Consider an LTI and causal system described by the difference equation:

𝑄(𝐸)𝑦[𝑛] = 𝑃(𝐸)𝑥[𝑛]

(𝐸 𝑁 + 𝑎𝑁−1 𝐸 𝑁−1 + ⋯ + 𝑎1 𝐸 + 𝑎0 )𝑦[𝑛]

= (𝑏𝑁 𝐸 𝑁 + 𝑏𝑁−1 𝐸 𝑁−1 + ⋯ + 𝑏1 𝐸 + 𝑏0 )𝑥[𝑛]

The unit impulse response ℎ[𝑛] is the sytem response when the input is 𝑥[𝑛] =
𝛿[𝑛] with all initial conditions are zero (system at rest), that is 𝑦[𝑛] = ℎ[𝑛] when
𝑥[𝑛] = 𝛿[𝑛]:

𝑄(𝐸)ℎ[𝑛] = 𝑃(𝐸)𝛿[𝑛]

subject to ICs: ℎ[−1] = ℎ[−2] = ⋯ = ℎ[−𝑁] = 0

Remark: Since the system is causal, ℎ[𝑛] = 0 for 𝑛 < 0.

Remark: Remember 𝛿[𝑛] = 1 for 𝑛 = 0 and 𝛿[𝑛] = 0 for 𝑛 ≠ 0.

The solution for ℎ[𝑛] can be found by iteration or in closed-form.

Example: (Iterative Solution)

𝑦[𝑛 + 2] + 0.3𝑦[𝑛 + 1] − 0.1𝑦[𝑛] = 𝑥[𝑛 + 2] + 2𝑥[𝑛]

Find ℎ[𝑛]?

Solution:

Write the system equation in delay form and set 𝑥[𝑛] = 𝛿[𝑛] and 𝑦[𝑛] = ℎ[𝑛]:

ℎ[𝑛] = −0.3ℎ[𝑛 − 1] + 0.1ℎ[𝑛 − 2] + 𝛿[𝑛] + 2𝛿[𝑛 − 2]

© M.-Y. Dabbagh 3-20

 ℎ[0] = −0.3 ℎ[−1]
⏟ + 0.1 ℎ[−2]
⏟ + 𝛿[0]
⏟ + 2 𝛿[−2]
⏟ =1
0 0 1 0

ℎ[1] = −0.3 ℎ[0]

⏟ + 0.1 ℎ[−1]
⏟ + 𝛿[1]
⏟ + 2 𝛿[−1]
⏟ = −0.3
1 0 0 0

ℎ[2] = −0.3 ℎ[1]

⏟ + 0.1 ℎ[0]
⏟ + 𝛿[2]
⏟ + 2 𝛿[0]
⏟ = 2.19
−0.3 1 0 1

Remark: Iterative solution is tedious and does not reveal the behavior of ℎ[𝑛].

 Closed-Form Solution of 𝒉[𝒏]

We have 𝑥[𝑛] = 𝛿[𝑛] ⇒ 𝑦[𝑛] = ℎ[𝑛]. Since 𝑥[𝑛] = 0 for 𝑛 > 0, then ℎ[𝑛]
contains characteristic modes for 𝑛 > 0. At 𝑛 = 0, ℎ[𝑛] may have 𝛿[𝑛]
component, i.e.

ℎ[𝑛] = 𝐴0 𝛿[𝑛] + 𝑦𝑐 [𝑛]𝑢[𝑛]

where 𝑦𝑐 [𝑛] is the linear combination of the characteristic modes of the system
with N unknown coefficients {𝑐1 , 𝑐2 , ⋯ , 𝑐𝑁 } in 𝑦𝑐 [𝑛].

The unknowns {𝐴0 , 𝑐1 , 𝑐2 , ⋯ , 𝑐𝑁 } can be determined from ℎ[0], ℎ[1], ⋯ , ℎ[𝑁]

which can be computed from the ICs ℎ[−1] = ℎ[−2] = ⋯ = ℎ[−𝑁] = 0
iteratively.

𝑏0
Remark: It is shown in the textbook that 𝐴0 = where 𝑎0 is the coefficient of
𝑎0

𝑦[𝑛] and 𝑏0 is the coefficient of 𝑥[𝑛] in the system difference equation.

Therefore,

𝑏0
ℎ[𝑛] = 𝛿[𝑛] + 𝑦𝑐 [𝑛]𝑢[𝑛]
𝑎0

© M.-Y. Dabbagh 3-21

 Example: Find the impulse response ℎ[𝑛] for the system:

𝑦[𝑛 + 2] − 2𝑦[𝑛 + 1] + 𝑦[𝑛] = 3𝑥[𝑛 + 2] + 2𝑥[𝑛]

Solution:

Characteristic equation: 𝑄(𝑟) = 𝑟 2 − 2𝑟 + 1 = (𝑟 − 1)2 = 0

Roots: 𝑟1 = 𝑟2 = 1 (repeated)

Characteristic modes: (1)𝑛 = 1, 𝑛(1)𝑛 = 𝑛

⇒ ℎ[𝑛] = 𝐴0 𝛿[𝑛] + (𝑐1 + 𝑐2 𝑛)𝑢[𝑛], 𝑛 ≥ 0

Find new ICs iteratively using ℎ[−1] = ⋯ = ℎ[−𝑁] = 0. From the system
equation with 𝑥[𝑛] = 𝛿[𝑛] → 𝑦[𝑛] = ℎ[𝑛]:

ℎ[𝑛] = 2ℎ[𝑛 − 1] − ℎ[𝑛 − 2] + 3𝛿[𝑛] + 2𝛿[𝑛 − 2]

ℎ[0] = 2 ℎ[−1]
⏟ − ℎ[−2]
⏟ + 3 𝛿[0]
⏟ + 2 𝛿[−2]
⏟ =3
0 0 1 0

ℎ[1] = 2 ℎ[0]
⏟ − ℎ[−1]
⏟ + 3 𝛿[1]
⏟ + 2 𝛿[−1]
⏟ =6
3 0 0 0

ℎ[2] = 2 ℎ[1]
⏟ − ℎ[0]
⏟ + 3 𝛿[2]
⏟ + 2 𝛿[0]
⏟ = 11
6 3 0 1

Substituting in ℎ[𝑛] to solve for {𝐴0 , 𝑐1 , 𝑐2 },

𝑐1 = 1
ℎ[0] = 3 = 𝐴0 + 𝑐1
𝑐2 = 5
ℎ[1] = 6 = 𝑐1 + 𝑐2 } ⇒ 𝑏0
ℎ[2] = 11 = 𝑐1 + 2𝑐2 𝐴0 = 2 =
{ 𝑎0

⇒ ℎ[𝑛] = 2𝛿[𝑛] + [1 + 5𝑛]𝑢[𝑛]

© M.-Y. Dabbagh 3-22

 Remark: (Special Case)

𝑏0
If 𝑎0 = 0, then the coefficient 𝐴0 = in the relation for ℎ[𝑛] does not apply.
𝑎0

In this case, ℎ[𝑛] is given by:

ℎ[𝑛] = 𝐴0 𝛿[𝑛] + 𝐴1 𝛿[𝑛 − 1] + 𝑦𝑐 [𝑛]𝑢[𝑛]

Similarly, if 𝑎0 = 𝑎1 = 0, then:

ℎ[𝑛] = 𝐴0 𝛿[𝑛] + 𝐴1 𝛿[𝑛 − 1] + 𝐴2 𝛿[𝑛 − 2] + 𝑦𝑐 [𝑛]𝑢[𝑛]

and so on. (See textbook section 3.12)

© M.-Y. Dabbagh 3-23

3.7 Zero-State Response

Assume an LTI system 𝑆: 𝑦[𝑛] = 𝑆{𝑥[𝑛]}

The impulse response is: ℎ[𝑛] = 𝑆{𝛿[𝑛]}

By time-invariance: ℎ[𝑛 − 𝑚] = 𝑆{𝛿[𝑛 − 𝑚]}

By sifting/sampling property: 𝑥[𝑛] = ∑∞

𝑚=−∞ 𝑥[𝑚]𝛿[𝑛 − 𝑚]

Substituting in 𝑦[𝑛]:

𝑦[𝑛] = 𝑆 { ∑ 𝑥[𝑚]𝛿[𝑛 − 𝑚]}

𝑚=−∞

By linearity of 𝑆 and the summation operations:

𝑦[𝑛] = ∑ 𝑥[𝑚] ⏟
𝑆{𝛿[𝑛 − 𝑚]}
𝑚=−∞ ℎ[𝑛−𝑚]

This can be rewritten as:

𝑦[𝑛] = 𝑥[𝑛] ∗ ℎ[𝑛] = ∑ 𝑥[𝑚]ℎ[𝑛 − 𝑚]

𝑚=−∞

This is the convolution sum for DT systems, similar to the convolution integral
for CT systems. It is valid for all 𝑛 ( −∞ < 𝑛 < +∞).

Remark: Properties of the convolution sum are similar to the convolution integral.

Remark: For causal systems: ℎ[𝑛] = 0 for 𝑛 < 0

⇒ ℎ[𝑛 − 𝑚] = 0 for 𝑛 − 𝑚 < 0 or 𝑚 > 𝑛

© M.-Y. Dabbagh 3-24

 And, if the input 𝑥[𝑛] is a causal signal: 𝑥[𝑚] = 0 for 𝑚 < 0

Then, the convolution sum can be rewritten as:

𝑦[𝑛] = 𝑥[𝑛] ∗ ℎ[𝑛] = ∑ 𝑥[𝑚]ℎ[𝑛 − 𝑚]

𝑚=0

Therefore, the output 𝑦[𝑛] is a causal signal, too.

Example: Let 𝑥1 [𝑛] = 𝑟1𝑛 𝑢[𝑛] and 𝑥2 [𝑛] = 𝑟2𝑛 𝑢[𝑛], where 𝑟2 ≠ 𝑟1 . Find the
convolution 𝑥1 ∗ 𝑥2 ?

Solution:
∞ ∞

𝑥1 ∗ 𝑥2 = ∑ 𝑥1 [𝑚]𝑥2 [𝑛 − 𝑚] = ∑ 𝑟1𝑚 ⏟𝑢[𝑚] . 𝑟2𝑛−𝑚 ⏟

𝑢[𝑛 − 𝑚]
𝑚=−∞ 𝑚=−∞ 0, 𝑚<0 0, 𝑚>𝑛
={ 1, 𝑚≥0 ={ 1, 𝑚≤𝑛

For 𝑛 < 0: 𝑥1 ∗ 𝑥2 = 0

For 𝑛 ≥ 0:
𝑛 𝑛
𝑟1 𝑚
𝑥1 ∗ 𝑥2 = ∑ 𝑟1𝑚 . 𝑟2𝑛−𝑚 = 𝑟2𝑛 ∑( )
𝑟2
𝑚=0 ⏟
𝑚=0
geometric sum

𝑧 𝑛+1 −1
Using the geometric sum formula ∑𝑛𝑚=0 𝑧 𝑚 = , 𝑧 ≠ 1 gives:
𝑧−1

𝑟 𝑛+1
(𝑟1 ) −1 1
𝑛 2
𝑥1 ∗ 𝑥2 = 𝑟2 𝑟1 = (𝑟1 𝑛+1 − 𝑟2 𝑛+1 ), 𝑛≥0
(𝑟 ) − 1 𝑟1 − 𝑟2
2

1
⇒ 𝑥1 ∗ 𝑥2 = (𝑟 𝑛+1 − 𝑟2 𝑛+1 )𝑢[𝑛]
𝑟1 − 𝑟2 1

© M.-Y. Dabbagh 3-25

 Example: An LTI system is described by:

𝑦[𝑛 + 2] − 2𝑦[𝑛 + 1] + 𝑦[𝑛] = 3𝑥[𝑛 + 2] + 2𝑥[𝑛]

Find the output 𝑦[𝑛], 𝑛 ≥ 0 when 𝑦[−1] = −1, 𝑦[−2] = 1, and 𝑥[𝑛] = 𝑢[𝑛].

Solution:

(1) Find the zero-input response 𝑦𝑧𝑖 [𝑛] (due to ICs):

𝑄(𝑟) = 𝑟 2 − 2𝑟 + 1 = (𝑟 − 1)2 = 0

Roots: 𝑟1 = 𝑟2 = 1 (repeated twice)

Modes: (1)𝑛 = 1, 𝑛(1)𝑛 = 𝑛

⇒ 𝑦𝑧𝑖 [𝑛] = (𝐶1 + 𝐶2 𝑛) , 𝑛≥0

Use IC’s to find the constants 𝐶1 and 𝐶2 :

𝑦[−1] = 𝑦𝑧𝑖 [−1] = 𝐶1 − 𝐶2 = −1 (1)

𝑦[−2] = 𝑦𝑧𝑖 [−2] = 𝐶1 − 2𝐶2 = 1 (2)

Solving (1) and (2) gives: 𝐶1 = −3 and 𝐶2 = −2

⟹ 𝑦𝑧𝑖 [𝑛] = (−3 − 2𝑛)𝑢[𝑛]

(2) Find the impulse response ℎ[𝑛]:

This was found for the same system in a previous example:

ℎ[𝑛] = 2𝛿[𝑛] + (1 + 5𝑛)𝑢[𝑛]

© M.-Y. Dabbagh 3-26

 (3) Find the zero-state response 𝑦𝑧𝑠 [𝑛] (due to 𝑥[𝑛] = 𝑢[𝑛]):

For 𝑛 ≥ 0: 𝑦𝑧𝑠 [𝑛] = ℎ ∗ 𝑥 = ∑∞

𝑚=−∞ ℎ[𝑚]𝑥[𝑛 − 𝑚]

⟹ 𝑦𝑧𝑠 [𝑛] = ∑ {2𝛿[𝑚] + (1 + 5𝑚) ⏟

𝑢[𝑚] } ⏟
𝑢[𝑛 − 𝑚]
𝑚=−∞ 0, 𝑚<0 0, 𝑚>𝑛
={ 1, 𝑚≥0 ={ 1, 𝑚≤𝑛

𝑛 𝑛

=2+ ∑ 1+5 ∑ 𝑚
⏟
𝑚=0 ⏟
𝑚=0
𝑛+1 𝑛(𝑛+1)/2

5
⟹ 𝑦𝑧𝑠 [𝑛] = {2 + (𝑛 + 1) + 𝑛(𝑛 + 1)} 𝑢[𝑛]
2

(4) Total Response: 𝑦[𝑛] = 𝑦𝑧𝑖 [𝑛] + 𝑦𝑧𝑠 [𝑛], 𝑛 ≥ 0

5
𝑦[𝑛] = (−3 − 2𝑛)𝑢[𝑛] + {2 + (𝑛 + 1) + 𝑛(𝑛 + 1)} 𝑢[𝑛]
2

3 5
⟹ 𝑦[𝑛] = ( 𝑛 + 𝑛2 ) 𝑢[𝑛]
2 2

Remark: Similar to the convolution integral, the convolution sum can be

performed in several methods to help in its computation.

(See example 3.21 in textbook for using the convolution Table 3.1).

(See example 3.23 in textbook for graphical method).

(See example 3.24 in textbook for sliding-tape method).

© M.-Y. Dabbagh 3-27

 Unit Step Response

Assume an LTI system.

Impulse response: ℎ[𝑛] = 𝑆{𝛿[𝑛]}

Step response: 𝑔[𝑛] = 𝑆{𝑢[𝑛]}

then

𝑔[𝑛] = ∑ ℎ[𝑚]
𝑚=−∞

ℎ[𝑛] = 𝑔[𝑛] − 𝑔[𝑛 − 1]

Proof: From

𝑦[𝑛] = ∑ ℎ[𝑚] 𝑥[𝑛 − 𝑚]

𝑚=−∞

∞ 𝑛

⇒ 𝑔[𝑛] = ∑ ℎ[𝑚] ⏟
𝑢[𝑛 − 𝑚] = ∑ ℎ[𝑚]
𝑚=−∞ 1, 𝑚≤𝑛 𝑚=−∞
={0, 𝑚>𝑛

Also, 𝑔[𝑛] − 𝑔[𝑛 − 1] = ∑𝑛𝑚=−∞ ℎ[𝑚] − ∑𝑛−1

𝑚=−∞ ℎ[𝑚]

= ∑𝑛−1 𝑛−1
𝑚=−∞ ℎ[𝑚] + ℎ[𝑛] − ∑𝑚=−∞ ℎ[𝑚] = ℎ[𝑛]

© M.-Y. Dabbagh 3-28

 System Response to the Everlasting Exponential

Assume an LTI system with an impulse response ℎ[𝑛].

Let the input be the everlasting exponential 𝑥[𝑛] = 𝑧 𝑛 , −∞ < 𝑛 < ∞ where 𝑧
is a complex constant. The output is given by:

𝑦[𝑛] = ℎ[𝑛] ∗ 𝑥[𝑛] = ∑ ℎ[𝑚] 𝑥[𝑛 − 𝑚]

𝑚=−∞

∞ ∞

= ∑ ℎ[𝑚] 𝑧 𝑛−𝑚 = 𝑧 𝑛 ∑ ℎ[𝑚] 𝑧 −𝑚

𝑚=−∞ ⏟
𝑚=−∞
𝐻(𝑧) function of 𝑧

⇒ 𝑦[𝑛] = 𝐻(𝑧)𝑧 𝑛

The function 𝐻(𝑧) is called the transfer function of the system. For a given value
of 𝑧, 𝐻(𝑧) is a constant, not a function of time 𝑛. Therefore, the output 𝑦[𝑛]
is also an everlasting exponential, i.e.

Everlasting Exponential input ⟹ Everlasting Exponential output

– For the case of an LTI system described by a difference equation:

𝑄(𝐸)𝑦[𝑛] = 𝑃(𝐸)𝑥[𝑛]

The transfer function is:

𝑃(𝑧)
𝐻(𝑧) =
𝑄(𝑧)

The proof is similar to the case of CT systems (See textbook).

© M.-Y. Dabbagh 3-29

3.8 Interconnection of Systems

Similar to CT systems.

 Parallel Connection:

ℎ1 [𝑛]
𝑥[𝑛] 𝑦[𝑛]
+
ℎ2 [𝑛]

Figure 7

Equivalent impulse response: ℎ[𝑛] = ℎ1 [𝑛] + ℎ2 [𝑛]

 Cascade Connection:

𝑥[𝑛] 𝑦[𝑛]
ℎ1 [𝑛] ℎ2 [𝑛]

Figure 8

Equivalent impulse response: ℎ[𝑛] = ℎ1 [𝑛] ∗ ℎ2 [𝑛]

© M.-Y. Dabbagh 3-30

3.9 DT System Stability

Almost similar to CT systems.

 Internal (Asymptotic) Stability

Assume an LTI system is in zero-state or rest state [ICs = 0, 𝑥[𝑛] = 0, 𝑦[𝑛] = 0]

which is a stable state. Due to the application of small disturbances (non-zero
ICs), we say:

(1) The system is asymptotically stable (AS) if all system modes → 0 as 𝑡 → ∞,

i.e. system goes back to its rest state.

(2) The system is unstable if at least one system mode → ∞ as 𝑡 → ∞.

(3) The system is marginally stable (MS) if some modes are bounded, while the
remaining modes → 0 as 𝑡 → ∞.

For an LTI system described by 𝑄(𝐸)𝑦[𝑛] = 𝑃(𝐸)𝑥[𝑛] with N roots:

𝑟1 , 𝑟2 , ⋯ , 𝑟𝑁 (Complex in general)

The system modes are of the form:

𝑟𝑖𝑛 (Distinct) or 𝑛𝑘 𝑟𝑖𝑛 (Repeated), 𝑘 = 1,2, …

Let 𝑟𝑖 = |𝑟𝑖 |𝑒 𝑗𝛽 , and for the distinct roots:

0 if |𝑟𝑖 | < 1
lim |𝑟𝑖𝑛 | = lim |𝑟𝑖 |𝑛 |𝑒
⏟
𝑗𝛽𝑛
| = lim |𝑟𝑖 |𝑛 = {
𝑛→∞ 𝑛→∞ 𝑛→∞ ∞ if |𝑟𝑖 | > 1
=1

This is also valid for the repeated modes 𝑛𝑘 𝑟𝑖𝑛 . Therefore, we conclude:

System is asymptotically stable (AS) if |𝑟𝑖 | < 1 for all roots

System is unstable if |𝑟𝑖 | > 1 for at least one root

© M.-Y. Dabbagh 3-31

 For the case |𝑟𝑖 | = 1 and unrepeated root:

lim |𝑟𝑖𝑛 | = lim |𝑟𝑖 |𝑛 = lim (1)𝑛 = 1

𝑛→∞ 𝑛→∞ 𝑛→∞

For the case |𝑟𝑖 | = 1 and repeated root:

lim |𝑛𝑘 𝑟𝑖𝑛 | = lim |𝑛|𝑘 |𝑟𝑖 |𝑛 = lim |𝑛|𝑘 (1)𝑛 = lim |𝑛|𝑘 → ∞
𝑛→∞ 𝑛→∞ 𝑛→∞ 𝑛→∞

Therefore, we conclude:

System is marginally stable (MS) if for some roots |𝑟𝑖 | = 1 and unrepeated,
while the remaining roots are inside the unit circle, i.e. |𝑟𝑖 | < 1.

System is unstable if for at least one root |𝑟𝑖 | = 1 and repeated.

Summary: (Stability and location of roots of 𝑸(𝒓) in the complex plane)

𝐼𝑚(𝑟𝑖 )

j Unstable
xx
x
x xx
x x xx
x x x x
-1 xx 1 𝑅𝑒(𝑟𝑖 )
x x xx
AS x
Unstable x
xx
-j Unstable
MS

Figure 9

© M.-Y. Dabbagh 3-32

 Example:

𝑦[𝑛 + 2] − 0.6𝑦[𝑛 + 1] − 0.16𝑦[𝑛] = 5𝑥[𝑛]

⟹ 𝑄(𝑟) = 𝑟 2 − 0.6𝑟 − 0.16 = (𝑟 + 0.2)(𝑟 − 0.8)

Roots: 𝑟1 = −0.2, 𝑟2 = 0.8

⟹ System is AS since |𝑟1 | < 1 and |𝑟2 | < 1 (inside the unit circle)

Example:

𝑦[𝑛 + 2] + 1.5𝑦[𝑛 + 1] − 𝑦[𝑛] = 2𝑥[𝑛]

⟹ 𝑄(𝑟) = 𝑟 2 + 1.5𝑟 − 1 = (𝑟 − 0.5)(𝑟 + 2)

Roots: 𝑟1 = 0.5, 𝑟2 = −2

⟹ System is unstable due to 𝑟2 (outside the unit circle)

Example: (Oscillator)

𝑦[𝑛 + 2] − 𝑦[𝑛 + 1] + 𝑦[𝑛] = 3𝑥[𝑛 + 1]

1 √3 1 √3
𝑄(𝑟) = 𝑟 2 − 𝑟 + 1 = (𝑟 − − 𝑗 ) (𝑟 − + 𝑗 )
2 2 2 2

𝜋
1 √3 ±𝑗
Roots: 𝑟1 , 𝑟2 = ± 𝑗 = 1𝑒 3 (complex on the unit circle)
2 2

⟹ System is MS since |𝑟1 | = |𝑟2 | = 1 (unrepeated)

© M.-Y. Dabbagh 3-33

 External Bounded-Input Bounded-Output (BIBO) Stability

BIBO stability for DT systems is similar to CT systems.

Definition: A system is BIBO stable if every bounded input 𝑥[𝑛] produces

bounded output 𝑦[𝑛]. If even one bounded input produces an unbounded output,
the system is BIBO unstable.

For LTI system, and from the convolution sum:

∞

𝑦[𝑛] = ℎ[𝑛] ∗ 𝑥[𝑛] = ∑ ℎ[𝑚] 𝑥[𝑛 − 𝑚]

𝑚=−∞

∞ ∞

|𝑦[𝑛]| = | ∑ ℎ[𝑚] 𝑥[𝑛 − 𝑚]| ≤ ∑ |ℎ[𝑚]| |𝑥[𝑛 − 𝑚]|

𝑚=−∞ 𝑚=−∞

If 𝑥[𝑛] is bounded, i.e. |𝑥[𝑛 − 𝑚]| ≤ 𝐾1 < ∞, where 𝐾1 is a constant, then:

∞

|𝑦[𝑛]| ≤ 𝐾1 ∑ |ℎ[𝑚]|
𝑚=−∞

The output is bounded iff ℎ[𝑛] is absolutely summable, i.e.:

∞

∑ |ℎ[𝑚]| ≤ 𝐾2 < ∞
𝑚=−∞

where 𝐾2 is a constant. We conclude:

An LTI system is BIBO stable iff its impulse response ℎ[𝑛] is

absolutely summable.

Remark: The relationships between asymptotic stability and BIBO stability for
DT systems are similar to CT systems:

© M.-Y. Dabbagh 3-34

 AS system ⇒ BIBO stable

MS (with no hidden MS modes) ⇒ BIBO unstable

BIBO stable ⇏ ? (cannot say, system can be AS, MS or even unstable).

Example:

ℎ[𝑛] = {(−0.2)𝑛 + (0.8)𝑛 }𝑢[𝑛]

∞ ∞

∑ |ℎ[𝑛]| = ∑ |{(−0.2)𝑛 + (0.8)𝑛 }𝑢[𝑛]|

𝑛=−∞ 𝑛=−∞

∞ ∞ ∞

= ∑ |(−0.2)𝑛 + (0.8)𝑛 | ≤ ∑ |(−0.2)𝑛 | + ∑ |(0.8)𝑛 |

𝑛=0 𝑛=0 𝑛=0

∞ ∞
1 1
= ∑ (0.2)𝑛 + ∑ (0.8)𝑛 = + = 6.25
1 − 0.2 1 − 0.8
𝑛=0 𝑛=0

Since ℎ[𝑛] is absolutely summable ⇒ System is BIBO stable.

Example: (Oscillator)
∞
𝜋
ℎ[𝑛] = 6 cos ( 𝑛) 𝑢[𝑛] ⇒ ∑ |ℎ[𝑛]| → ∞
3
𝑛=−∞

⇒ System is BIBO unstable.

𝜋
Bad input: 𝑥[𝑛] = 𝑐 cos ( 𝑛) 𝑢[𝑛]
3

(External resonance: The output is unbounded even though the input is bounded)

© M.-Y. Dabbagh 3-35