ECE4330

Lecture 21
Linear Discrete-Time Systems:
Solution of Difference Equations
by Iteration, by the Z-transform and by Convolution
Prof. Mohamad Hassoun

Linear Time-Invariant Discrete-Time (LTID) System Analysis

Consider a linear discrete-time system. We are interested in solving for the
complete response 𝑦[𝑘] given the difference equation governing the
system, its associated initial conditions and the input 𝑓[𝑘].

It will be shown that 𝑦[𝑘] is a sequence of numbers that can be obtained

explicitly by simple recursion (iteration). Alternatively, an analytic
solution of the linear difference equation can be obtained based on the Z-
transform method. The analytic solution can also be obtained based on a
systematic time-domain method, as covered in the next lecture.

There are two equivalent formulations for a difference equation: The

advance operator formulation and the delay operator formulation. The
following analysis illustrates how to convert one formulation to the other.
Delay Operator Formulation
The following is an example of a delay formulation, second-order linear
difference equation. Here, the signal 𝑦[𝑘] appears right shifted,

𝑦[𝑘] − 5𝑦[𝑘 − 1] + 6𝑦[𝑘 − 2] = 𝑓 [𝑘 − 1] − 5𝑓[𝑘 − 2] (1)

with the initial conditions 𝑦[−1] = 2 and 𝑦[−2] = 1. Let the input signal
be the causal signal 𝑓 [𝑘] = (3𝑘 + 5)𝑢[𝑘]. Solve for 𝑦[𝑘], 𝑘 ≥ 0

Advance Operator Formulation

The advance operator formulation for the above difference equation can be
obtained by simply shifting the difference equation so that the largest
difference term 𝑦[𝑘 − 2] becomes 𝑦[𝑘]. Which requires replacing 𝑘 by
𝑘 + 2 throughout the equation, to obtain:

𝑦[𝑘 + 2] − 5𝑦[𝑘 + 1] + 6𝑦[𝑘] = 𝑓 [𝑘 + 1] − 5𝑓[𝑘] (2)

Next, we need to establish a set of appropriate initial conditions, 𝑦[0] and

𝑦[1]. This can be done by setting 𝑘 to 0 and then to 1 in Equation (1) and
solving for 𝑦[0] and 𝑦[1], respectively, as follows.
Set 𝑘 = 0 in Eqn. (1) to obtain 𝑦[0],

𝑦[0] = 5𝑦[−1] − 6𝑦[−2] + 𝑓[−1] − 5𝑓 [−2]

𝑦[0] = 5(2) − 6(1) + 0 − 0 = 10 − 6 → 𝑦[0] = 4
Set 𝑘 = 1 in Eqn. (1) to obtain 𝑦[1],
𝑦[1] = 5𝑦[0] − 6𝑦[−1] + 𝑓[0] − 5𝑓[−1]
𝑦[1] = 5(4) − 6(2) + 5 − 0 = 13 → 𝑦[1] = 13
It is important to note that when we derived the initial conditions for this
system, one of them (𝑦[1]) depended on the input value 𝑓[0]. So we say
that the initial conditions are tainted by the input. When that occurs, the
advance formulation should be avoided if we are looking for decoupled
solutions of the form 𝑦[𝑘] = 𝑦𝑧𝑠 [𝑘] + 𝑦𝑧𝑖 [𝑘]. This will be illustrated
through an example later in this lecture.

Analysis Methods for Computing 𝑦[𝑘]

 Iterative method (numerical solution; also works for non-linear systems)
 Z-transform method; allows for finding 𝑦𝑧𝑠 [𝑘] and 𝑦𝑧𝑖 [𝑘].
 Time-domain convolution for finding 𝑦𝑧𝑠 [𝑘]. (Refer to the last section.)
 Time-domain analysis. (Refer to the classical method of solution of
difference equations covered in Lecture 22.)
Solve the above advance-formulation, second-order linear difference
equation using the iterative method. First, we solve for the highest delay
term in the difference equation to arrive at the recursion formula
𝑦[𝑘 + 2] = 5𝑦[𝑘 + 1] − 6𝑦[𝑘] + [3(𝑘 + 1) + 5]𝑢[𝑘 + 1]
−5(3𝑘 + 5)𝑢[𝑘]
and then start iterating with 𝑦[0] = 4 and 𝑦[1] = 13, as is shown below.
𝑘 = 0: 𝑦[2] = 5𝑦[1] − 6𝑦[0] + (3 + 5)(1) − 5(0 + 5)(1)
𝑦[2] = 5(13) − 6(4) + 8 − 25 = 24
𝑘 = 1: 𝑦[3] = 5𝑦[2] − 6𝑦[1] + (6 + 5)(1) − 5(3 + 5)(1)
𝑦[3] = 5(24) − 6(13) + 11 − 40 = 13
Repeat for 𝑘 = 2, 3, 4, … to generate the solution sequence (𝑘 ≥ 0)
𝑦[𝑘] = 4, 13, 24, 13, −120, …

Next, we illustrate the iterative solution applied to the delay formulation

equation, Eqn. (1). First, we obtain the recursion formula for 𝑦[𝑘] as,

𝑦[𝑘] = 5𝑦[𝑘 − 1] − 6𝑦[𝑘 − 2] + (3𝑘 + 2)𝑢[𝑘 − 1] − 5(3𝑘 − 1)𝑢[𝑘 − 2]

and start iterating with 𝑦[−1] = 2 and 𝑦[−2] = 1 as follows,

𝑘 = 0: 𝑦[0] = 5𝑦[−1] − 6𝑦[−2] + 2𝑢[−1] − 5(−1)𝑢[−2]

𝑦[0] = 5(2) − 6(1) + 0 + 0
𝑦[0] = 4
𝑘 = 1: 𝑦[1] = 5𝑦[0] − 6𝑦[−1] + 5𝑢[0] − 5(2)𝑢[−1]
𝑦[1] = 5(4) − 6(2) + 5 − 0
𝑦[1] = 13
Repeat for 𝑘 = 2, 3, 4, … to generate the solution sequence (𝑘 ≥ 0)
𝑦[𝑘] = 4, 13, 24, 13, −120, …
This sequence is identical to the one obtained earlier using the advance
formulation.
Mathcad numerical solution for the advance operator difference equation
(note: press ‘[‘ to input the subscript):

Mathcad numerical solution for the delay operator difference equation

Note how the above Mathcad solution automatically performs the
required iterations to compute the sequence (vector) 𝑦.
Mathcad allows the user to redefine the array origin. For the above
delay operator recursion we had to set the origin at −2 in the Math
Options menu, as illustrated below. (For the advance formulation
solution the array origin has to be set to 0.)
 Solving for the Complete Response 𝑦[𝑘] = 𝑦𝑧𝑠 [𝑘] + 𝑦𝑧𝑖 [𝑘]
using the Z-Transform Method

The Z-transform method can be applied to the difference equation to

(analytically) generate the complete response. However, if we intend to
use it to generate the decoupled response 𝑦[𝑘] = 𝑦𝑧𝑠 [𝑘] + 𝑦𝑧𝑖 [𝑘], we have
to make sure that the initial conditions we use are not dependent on the
input signal.
As pointed out earlier, one of the advance formulation initial conditions
(for our example system) is dependent on the input. Therefore, for that
advance formulation system the Z-transform method can be applied to
generate 𝑦[𝑘], but not 𝑦𝑧𝑠 [𝑘] and 𝑦𝑧𝑖 [𝑘]. On the other hand, the Z-
transform method can be used to generate 𝑦[𝑘] = 𝑦𝑧𝑠 [𝑘] + 𝑦𝑧𝑖 [𝑘] for the
delay formulation system.
Example. Z-transform analysis for the delay formulation system:
𝑦[𝑘] − 5𝑦[𝑘 − 1] + 6𝑦[𝑘 − 2] = 𝑓[𝑘 − 1] − 5𝑓[𝑘 − 2]
𝑦[−1] = 2 , 𝑦[−2] = 1
𝑓 [𝑘] = (3𝑘 + 5)𝑢[𝑘]
We will find 𝑦𝑧𝑠 [𝑘] first by Z-transforming the difference equation,
assuming zero initial conditions, to obtain 𝐻(𝑧) and then employ the
formula,
𝑦𝑧𝑠 [𝑘] = 𝑍 −1 {𝐻(𝑧)𝐹(𝑧)}
But first, we must be clear about the meaning of the shifted signals
𝑦[𝑘 − 1] and 𝑦[𝑘 − 2]. Do they mean 𝑦[𝑘 − 1]𝑢[𝑘] and 𝑦[𝑘 − 2]𝑢[𝑘] or
𝑦[𝑘 − 1]𝑢[𝑘 − 1] and 𝑦[𝑘 − 2]𝑢[𝑘 − 2]?
Well, since our Z-transform is one-sided it means that we are considering
the situation for 𝑘 ≥ 0. So, every signal is assumed to be causal and is
counted from 𝑘 = 0. Therefore, the terms 𝑦[𝑘 − 𝑚] mean 𝑦[𝑘 − 𝑚]𝑢[𝑘],
for integer 𝑚. Transforming (and using the right-shift property) we obtain
1 1 1
𝑌(𝑧) − 5 ( 𝑌(𝑧) + 𝑦[−1]) + 6 ( 2 𝑌(𝑧) + 𝑦[−1] + 𝑦[−2])
𝑧 𝑧 𝑧
1 1 1
= ( 𝐹(𝑧) + 𝑓[−1]) − 5 ( 2 𝐹(𝑧) + 𝑓 [−1] + 𝑓[−2])
𝑧 𝑧 𝑧
which, with the requirement that the initial conditions are zero (zero-state)
and by noting that 𝑓 [𝑘] is a causal signal {𝑓 [−1] = 𝑓[−2] = 0}, reduces
to
5 6 1 5
(1 − + 2 ) 𝑌𝑧𝑠 (𝑧) = ( − 2 ) 𝐹(𝑧)
𝑧 𝑧 𝑧 𝑧
Next, multiplying the above equation by 𝑧 2 , leads to
(𝑧 2 − 5𝑧 + 6)𝑌𝑧𝑠 (𝑧) = (𝑧 − 5)𝐹(𝑧)
which has the form 𝑄 (𝑧)𝑌𝑧𝑠 (𝑧) = 𝑃(𝑧)𝐹(𝑧). The transfer function is then,
𝑌𝑧𝑠 (𝑧) 𝑃(𝑧)
𝐻(𝑧) ≜ =
𝐹(𝑧) 𝑄(𝑧)
which results in
𝑧−5
𝐻(𝑧) = → The poles are at 𝑧1 = 2 , 𝑧2 = 3
(𝑧−2)(𝑧−3)

The system is unstable since |𝑧𝑖 | > 1.

Before we proceed to solve for 𝑦𝑧𝑠 [𝑘], let us determine the unit-impulse
response,
ℎ[𝑘] = 𝑍 −1 {𝐻(𝑧)}
 𝐻(𝑧)
First, perform partial fraction expansion on (say, using Mathcad):
𝑧

z5 5 3 2
convert  parfrac  z   
z ( z  2) ( z  3) 6 z 2 ( z  2) 3 ( z  3)

Next, multiply by z and use the table of Z-transforms to find the inverse Z-
transform:
3 2
5 𝑧 𝑧
𝐻(𝑧) = − + 2 − 3
6 𝑧−2 𝑧−3
5 3 2
ℎ [𝑘 ] = − 𝛿 [𝑘] + (2)𝑘 𝑢[𝑘] − (3)𝑘 𝑢[𝑘]
6 2 3

The zero-state response 𝑦𝑧𝑠 [𝑘] may now be obtained using convolution
sums,
𝑦𝑧𝑠 [𝑘] = ℎ[𝑘] ∗ 𝑓[𝑘] , where 𝑓 [𝑘] = 3𝑘𝑢[𝑘] + 5𝑢[𝑘]

[Your turn: Employ the Convolution Table for discrete-time signals,

Page 590, to solve for 𝑦𝑧𝑠 [𝑘]. Refer to the last section (last three slides of
this lecture) for an example.]

Alternatively, we may use the inverse Z-transform to obtain 𝑦𝑧𝑠 [𝑘] as

follows,
𝑦𝑧𝑠 [𝑘] = 𝑍 −1 {𝑌𝑧𝑠 (𝑧)} = 𝑍 −1 {𝐻(𝑧)𝐹(𝑧)}

−1
𝑧−5 𝑧(5𝑧 − 2)
𝑦𝑧𝑠 [𝑘] = 𝑍 { }
(𝑧 − 2)(𝑧 − 3) (𝑧 − 1)2

where the following transform pair was used to obtain 𝐹(𝑧),

 3𝑧 5𝑧 𝑧(5𝑧 − 2)
𝑓[𝑘] = (3𝑘 + 5)𝑢[𝑘] ↔ 𝐹 (𝑧) = + =
(𝑧 − 1)2 𝑧 − 1 (𝑧 − 1)2

𝑌𝑧𝑠 (𝑧)
Your turn: Apply partial fraction expansion to , multiply the result
𝑧
by 𝑧, then employ the Z-transform Table to obtain the following result,
35 13
𝑦𝑧𝑠 [𝑘] = (− − 6𝑘 − (3)𝑘 + 24(2)𝑘 ) 𝑢[𝑘]
2 2

Next, we solve for the zero-input response, 𝑦𝑧𝑖 [𝑘].

Set 𝑓 [𝑘] = 0 in the difference equation,

𝑦[𝑘] − 5𝑦[𝑘 − 1] + 6𝑦[𝑘 − 2] = 0
Apply the Z-transform (include the contribution due to initial conditions),
1 1 1
𝑌𝑧𝑖 (𝑧) − 5 ( 𝑌𝑧𝑖 (𝑧) + 𝑦[−1]) + 6 ( 2 𝑌𝑧𝑖 (𝑧) + 𝑦[−1] + 𝑦[−2]) = 0
𝑧 𝑧 𝑧
Rearrange terms so as to keep the equation in the form
𝑄(𝑧)𝑌𝑧𝑖 (𝑧) + 𝐼(𝑧) = 0
5 6 12
(1 − + 2 ) 𝑌𝑧𝑖 (𝑧) + (−10 + + 6) = 0
𝑧 𝑧 𝑧
Multiply the equation by 𝑧 2 ,
 (⏟𝑧 2 − 5𝑧 + 6) 𝑌𝑧𝑖 (𝑧) + (−4𝑧
⏟ 2 + 12𝑧) = 0

𝑄(𝑧) 𝐼(𝑧)

Now we solve for 𝑌𝑧𝑖 (𝑧) in the form of a rational function to obtain,

−𝐼(𝑧) 4𝑧 2 − 12𝑧 4𝑧(𝑧 − 3) 4𝑧

𝑌𝑧𝑖 (𝑧) = = 2 = =
𝑄(𝑧) 𝑧 − 5𝑧 + 6 (𝑧 − 2)(𝑧 − 3) 𝑧 − 2

and readily observe that the solution is 𝑦𝑧𝑖 [𝑘] = 4(2)𝑘 𝑢[𝑘]. It is
interesting to note that the specific initial conditions that we used have
only excited one of the natural modes of the system [due to the
cancellation of the (𝑧 − 3) term]. The complete (decoupled) response can
now be expressed as,

𝑦[𝑘] = 𝑦𝑧𝑠 [𝑘] + 𝑦𝑧𝑖 [𝑘]

35 13
𝑦[𝑘] = (− − 6𝑘 − (3)𝑘 + 24(2)𝑘 ) 𝑢[𝑘] + 4
⏟(2)𝑘 𝑢[𝑘]
⏟2 2

𝑦𝑧𝑠 𝑦𝑧𝑖
 Mathcad Verification

Do not forget to multiply the above Mathcad answers by the discrete unit-
step function, 𝑢[𝑘].
Suppose that we are only interested in the (coupled) complete response.
Then, the solution is obtained as
Next, we apply the Z-transform method to the advance formulation
system,
𝑦[𝑘 + 2] − 5𝑦[𝑘 + 1] + 6𝑦[𝑘] = 𝑓 [𝑘 + 1] − 5𝑓 [𝑘]
where, 𝑦[0] = 4, 𝑦[1] = 13 and 𝑓 [𝑘] = (3𝑘 + 5)𝑢[𝑘]. Remember that
the signals 𝑦[𝑘 + 𝑚] and 𝑓 [𝑘 + 𝑚] in the advance formulation stand for
𝑦[𝑘 + 𝑚]𝑢[𝑘] and 𝑓[𝑘 + 𝑚]𝑢[𝑘], respectively.

Because, in this case, one of the initial conditions depends on the input to
the system, we can only obtain the coupled solution as follows,
𝑦[𝑘 + 2]𝑢[𝑘] ↔ 𝑧 2 𝑌(𝑧) − 𝑧 2 𝑦[0] − 𝑧𝑦[1] = 𝑧 2 𝑌(𝑧) − 4𝑧 2 − 13𝑧
−5𝑦[𝑘 + 1]𝑢[𝑘] ↔ −5(𝑧𝑌(𝑧) − 𝑧𝑦[0]) = −5(𝑧𝑌(𝑧) − 4𝑧)
6𝑦[𝑘]𝑢[𝑘] ↔ 6𝑌(𝑧)
𝑧(5𝑧 − 2)
𝑓[𝑘]𝑢[𝑘] = (3𝑘 + 5)𝑢[𝑘] ↔ 𝐹 (𝑧) =
(𝑧 − 1)2
𝑧(5𝑧 − 2)
𝑓 [𝑘 + 1]𝑢[𝑘] ↔ 𝑧𝐹(𝑧) − 𝑧𝑓[0] = 𝑧 − 5𝑧
(𝑧 − 1)2
Substituting the above expressions in the difference equation we get

2 2
𝑧(5𝑧 − 2)
(𝑧 − 5𝑧 + 6)𝑌(𝑧) + (−4𝑧 − 13𝑧 + 20𝑧) = (𝑧 − 5) − 5𝑧
(𝑧 − 1)2

Solve (Your turn) for 𝑌(𝑧) as one lumped rational function to obtain
𝑁(𝑧) 𝑧(4𝑧 3 − 15𝑧 2 + 𝑧 − 2)
𝑌(𝑧) = =
𝐷 (𝑧) (𝑧 − 1)2 (𝑧 − 2)(𝑧 − 3)
 𝑌(𝑧)
which, after applying partial fraction expansion to and then
𝑧
multiplying the result by 𝑧, leads to
35 13
6𝑧 𝑧 28𝑧 𝑧
2 2
𝑌(𝑧) = − − + −
(𝑧 − 1)2 𝑧 − 1 𝑧 − 2 𝑧 − 3
Using the Z-transform Table leads to the following final result
35 13
𝑦[𝑘] = 𝑍 −1 {𝑌(𝑧)} = −6𝑘𝑢[𝑘] − 𝑢[𝑘] + 28(2)𝑘 𝑢[𝑘] − 3𝑘 𝑢[𝑘]
2 2
By inspecting the above solution, we can identify the natural and forced
responses (but not the zero-state and zero-input. Why?) as
13 𝑘
𝑦𝑛 [𝑘] = 28(2)𝑘 𝑢[𝑘] − 3 𝑢[𝑘]
2
35
𝑦𝑓 [𝑘] = − 𝑢[𝑘] − 6𝑘𝑢[𝑘]
2
Your turn: Show that if you are to find 𝑦𝑧𝑠 [𝑘] and 𝑦𝑧𝑖 [𝑘] using the
advance formulation for this system it would result in the following wrong
decomposition (it does not agree with the decoupled delay formulation
solution that we obtained earlier):
𝑦𝑧𝑖 [𝑘] = 5(3)𝑘 𝑢[𝑘] − (2)𝑘 𝑢[𝑘]
35 23
𝑦𝑧𝑠 [𝑘] = − 𝑢[𝑘] − 6𝑘𝑢[𝑘] − 3𝑘 𝑢[𝑘] + 29(2)𝑘 𝑢[𝑘]
2 2
However, the total (coupled) response obtained by adding these two
signals would still be correct:
𝑦[𝑘] = 𝑦𝑧𝑖 [𝑘] + 𝑦𝑧𝑠 [𝑘] =
35 13
= −6𝑘𝑢[𝑘] − 𝑢[𝑘] + 28(2)𝑘 𝑢[𝑘] − 3𝑘 𝑢[𝑘]
2 2
Example. Consider the three-point moving average LTID system
1
𝑦[𝑘] = (𝑓[𝑘] + 𝑓 [𝑘 − 1] + 𝑓[𝑘 − 2])
3
and the discrete-time input signal shown below.

a. Find the impulse-response ℎ[𝑘]. Is the system an FIR or IIR system?

b. Find the zero-state response 𝑦[𝑘].
c. Verify and plot 𝑦[𝑘] employing Mathcad.

Solution:
Applying the 𝑧-transform to the given recursion formula and using Pair 1
in the 𝑧-Transform Table we obtain (note that 𝑓 [−1] = 𝑓 [−2] = 0)
1 1
𝑌(𝑧) = [𝐹 (𝑧) + 𝑧 −1 𝐹 (𝑧) + 𝑧 −2 𝐹 (𝑧)] = (1 + 𝑧 −1 + 𝑧 −2 )𝐹(𝑧)
3 3
The transfer function is then
𝑌 (𝑧) 1
𝐻 (𝑧) = = (1 + 𝑧 −1 + 𝑧 −2 )
𝐹 (𝑧) 3
The impulse response is obtained by applying the inverse 𝑧-transform to
the above expression to obtain
 1
ℎ 𝑘 = ([𝛿 ] + 𝛿 [𝑘 − 1] + 𝛿 [𝑘 − 2])
[ ]
3
1
So, ℎ[𝑘] = 3 for 𝑘 = 0,1,2 and zero otherwise. Therefore, the system has
a finite impulse response (FIR).
The input is given by
𝑓[𝑘] = 𝛿[𝑘] + 2𝛿[𝑘 − 1] − 𝛿[𝑘 − 2] + 𝛿[𝑘 − 3] + 2𝛿[𝑘 − 4]
which has a 𝑧-transform given by
𝐹 (𝑧) = 1 + 2𝑧 −1 − 𝑧 −2 + 𝑧 −3 + 2𝑧 −4
The zero-state response in the 𝑧-domain is
1
𝑌(𝑧) = 𝐻 (𝑧)𝐹 (𝑧) = (1 + 𝑧 −1 + 𝑧 −2 )(1 + 2𝑧 −1 − 𝑧 −2 + 𝑧 −3 + 2𝑧 −4 )
3
1
= (1 + 3𝑧 −1 + 2𝑧 −2 + 2𝑧 −3 + 2𝑧 −4 + 3𝑧 −5 + 2𝑧 −6 )
3
Applying the inverse 𝑧-tarnsform gives the zero-state response
1 2 2 2
[ ]
𝑦 𝑘 = 𝛿[𝑘] + 𝛿[𝑘 − 1] + 𝛿[𝑘 − 2] + 𝛿[𝑘 − 3] + 𝛿[𝑘 − 4] + 𝛿[𝑘 − 5]
3 3 3 3
2
+ 𝛿[𝑘 − 6]
3
From the definition of the discrete-time unit-impulse function, we can
express the response as the sequence (starting at 𝑘 = 0)
1 2 2 2 2
𝑦[𝑘] = { , 1, , , , 1, , 0,0, … }
3 3 3 3 3

Your turn: Show that the above sequence is identical to the one obtained
using manual averaging of the input signal.
Mathcad verification session for the three-point moving average system:

Since averaging is a form of low-pass filtering we see how the system

supresses the sharp variations in the input signal.
Example. Determine the unit-impulse response for the following
difference equation,
𝑦[𝑘] − 2𝑦[𝑘 − 1] + 2𝑦[𝑘 − 2] = 𝑓 [𝑘]
We will first obtain the transfer function employing the formula
𝑌(𝑧)
𝐻(𝑧) = |
𝐹(𝑧) 𝐼.𝐶.=0
Z-transforming the equation (while setting initial conditions to zero) gives
1 1
𝑌(𝑧) − 2 ( 𝑌 𝑧 ) + 2 ( 2 𝑌(𝑧)) = 𝐹(𝑧)
( )
𝑧 𝑧
2 2
( 2 − + 1) 𝑌(𝑧) = 𝐹(𝑧)
𝑧 𝑧
𝑌 (𝑧) 1 𝑧2
( )
𝐻 𝑧 = = 2 2 = 2
𝐹 (𝑧) − + 1 𝑧 − 2𝑧 + 2
𝑧2 𝑧

The poles are at 𝑧1,2 = 1 ± 𝑗. This system is unstable since |𝑧𝑖 | = √2 > 1.
𝐻(𝑧) has complex conjugate poles. Therefore, we employ the Z-transform
Pair 12c to find the unit-impulse response ℎ[𝑘],
𝐴𝑧 2 + 𝐵𝑧
2 2
↔ 𝑟𝑏𝑘 cos(𝛽𝑘 + 𝜃) 𝑢[𝑘], 𝑏>0
𝑧 + 2𝑎𝑧 + 𝑏
where,

𝐴2 𝑏2 + 𝐵2 − 2𝐴𝑎𝐵 −𝑎
𝑟= √ −1
, 𝛽 = cos ( )
𝑏2 − 𝑎2 𝑏

−1
𝐴𝑎 − 𝐵
𝜃 = tan ( )
𝐴√𝑏2 − 𝑎2
By matching coefficients in
𝐴𝑧 2 + 𝐵𝑧 𝑧2
=
𝑧 2 + 2𝑎𝑧 + 𝑏2 𝑧 2 − 2𝑧 + 2
we obtain,

𝐴 = 1, 𝐵 = 0, 𝑏2 = 2 → 𝑏 = +√2 and 𝑎 = −1. Then we use the above

formulas to compute 𝑟, 𝛽 and 𝜃, as follows,
1 𝜋
𝑟 = √2, 𝛽 = cos −1 ( ) =
√2 4
−1 − 0 𝜋
𝜃 = tan−1 ( ) = tan−1 (−1) = −
1√2 − 1 4
Therefore,
𝑘 𝜋 𝜋 𝜋 𝜋
ℎ[𝑘] = √2(√2) cos ( 𝑘 − ) 𝑢[𝑘] = (√2)𝑘+1 cos ( 𝑘 − ) 𝑢[𝑘], 𝑘 ≥ 0
4 4 4 4

The system has an infinite impulse response (IIR). The first six output
values are

We can employ Simulink to solve the system (numerically) as shown

below.
Alternative time-domain Simulink solution:

Your turn: Solve numerically for the first 10 terms of the sequence
generated by the following recursion formula, 𝑢𝑛+2 = 4𝑢𝑛+1 − 8𝑢𝑛 .
Assume that 𝑢0 = 𝑢1 = 1. Then, employ the Z-transform method to find
the expression for the 𝑛th term, 𝑢𝑛 . Note: The above recursion formula
can also be represented as the second-order homogeneous LTI difference
equation 𝑦[𝑛 + 2] − 4𝑦[𝑛 + 1] + 8𝑦[𝑛] = 0, with 𝑦[0] = 𝑦[1] = 1.
3𝑛
𝑛𝜋 𝑛𝜋
Ans. 𝑢𝑛 = 2 2 −1 [2cos ( 4 ) − sin ( 4 )] , 𝑛 ≥ 0
The Amortization Problem. If you purchase a car, or a house, and take a
loan of 𝑑 dollars, with fixed annual interest rate, 𝑅 (i.e., a rate of 𝑟 =
𝑅/12 per month), then the loan is paid back through the process known in
economics as amortization. Let us say you will be making fixed payments
of 𝑝 dollars every month. What should your monthly payment be if you
want to pay the entire loan within 𝑁 months? (Typically, 𝑁 = 48 for a car
and 𝑁 = 180, 240 or 360 for a house).

Well, the outstanding principal at the end of a given month, 𝑦[𝑘 + 1], is
equal to the outstanding principal from the previous month, 𝑦[𝑘], plus the
monthly interest accrued on that principal, 𝑟𝑦[𝑘], minus the monthly
payment, 𝑓[𝑘]. We may express the above statement analytically as a
recursive formula (difference equation):
𝑦[𝑘 + 1] = 𝑦[𝑘] + 𝑟𝑦[𝑘] − 𝑓 [𝑘]
which amounts to the difference equation
𝑦[𝑘 + 1] − (1 + 𝑟)𝑦[𝑘] = −𝑓[𝑘]
With monthly payments (system input) 𝑓 [𝑘] = 𝑝𝑢[𝑘] and initial principal,
𝑦[0] = 𝑑.
Applying the Z-transform to the difference equation we get
𝑧𝑌(𝑧) − 𝑧𝑦[0] − (1 + 𝑟)𝑌(𝑧) = −𝐹 (𝑧)

𝑝𝑧
Solving for 𝑌(𝑧) and substituting 𝐹 (𝑧) = 𝑧−1 and 𝑦[0] = 𝑑, we obtain

−𝑝𝑧 𝑑𝑧
𝑌(𝑧) = +
[𝑧 − (1 + 𝑟)](𝑧 − 1) 𝑧 − (1 + 𝑟)
 𝑌 (𝑧 )
Performing pfe on , we get
𝑧

𝑌(𝑧) 𝑝 1 𝑝 1 𝑑
= (− ) +( ) +
𝑧 𝑟 𝑧 − (1 + 𝑟) 𝑟 𝑧 − 1 𝑧 − (1 + 𝑟)
𝑝 1 𝑝 1
= (𝑑 − ) ( )+( )
𝑟 𝑧 − (1 + 𝑟) 𝑟 𝑧−1

Next, applying the inverse Z-transform

𝑝 𝑧 𝑝 𝑧
𝑦[𝑘] = 𝑍 −1 {(𝑑 − ) ( )+( ) }
𝑟 𝑧 − (1 + 𝑟) 𝑟 𝑧−1
leads to the desired solution for the owed principal after 𝑘 months,
𝑝 𝑝
𝑦[𝑘] = ((𝑑 − ) (1 + 𝑟)𝑘 + ) 𝑢[𝑘]
𝑟 𝑟
If the loan is supposed to be paid back in full after 𝑁 months, then we can
solve for the required monthly payment by setting 𝑦[𝑁] = 0 in the above
equation to obtain,
𝑝 𝑝
((𝑑 − ) (1 + 𝑟)𝑁 + ) = 0
𝑟 𝑟
which has the following solution for the monthly payment
𝑟 (1 + 𝑟)𝑁
𝑝= 𝑑
(1 + 𝑟)𝑁 − 1
For example, assume you get approved for a 30 year (360 months)
$300,000 mortgage at an annual rate, 𝑅 = 4%. Then, the monthly
payment would be
0.04 0.04 360
( 12 ) (1 + )
12
𝑝= (300,000) = $1,432.25
0.04 360
(1 + ) −1
12

The principal after 20 years is

1,432.25 1,432.25
𝑦[240] = (300,000 − ) (1 + 0.04/12)240 + = $141,461.67
0.04/12 0.04/12

The total cost of the mortgage is

360𝑝 = 360($1432.25) = $515,610
which means that in this scenario your $300,000 house will coast you an
additional $515,610 − $300,000 = $215,610 in interest!

Online calculators are available to solve the amortization problem. They

use the same equations that were derived above. Here is one example:
Your turn. National income is the total value a country’s output of all
new goods and services produced in one year. The national income for
year 𝑘, 𝑦[𝑘], is governed by the following (Samuelson model) set of
difference equations [Paul Samuelson (1915-2009) a Nobel Prize-winning
economist]:
𝑦 [𝑘 ] = 𝑐 [𝑘 ] + 𝑖 [𝑘 ] + 𝑓 [𝑘 ]
𝑐 [𝑘] = 𝑎𝑦[𝑘 − 1]
𝑖[𝑘] = 𝑏(𝑐 [𝑘] − 𝑐 [𝑘 − 1])

where 𝑎 and 𝑏 are positive constants, 𝑐[𝑘] represents consumer

expenditures, 𝑖[𝑘] is the induced private investment (new investment
stimulated by an increase in demand), and 𝑓 [𝑘] represents government
expenditures.

Apply the Z-transform to the above system of equations and show that the
transfer function is given by
𝑌(𝑧) 𝑧2
𝐻 (𝑧) = =
𝐹 (𝑧) 𝑧 2 − 𝑎(1 + 𝑏)𝑧 + 𝑎𝑏
Assume that the government makes a sustained expenditure of 1 (trillion
dollars), 𝑓[𝑘] = 𝑢[𝑘]. Also, assume that the consumer’s “propensity to
consume coefficient” is 𝑎 = 0.8 (i.e., consumer expenditures are 80% of
the national income). Solve for and plot 𝑦𝑧𝑠 [𝑘] for 𝑏 = 0, 0.25, 0.5 and
0.9. Discuss the effects of the induced private investment “acceleration
coefficient” 𝑏 on the national income. What will the national income be
after many years? How does 𝑏 affect that value?
 Numerical Integration: Trapezoidal Rule as a LTID System
Consider the integral (for convenience, we set the lower limit to zero)
𝑡
𝑦𝑡𝑟𝑢𝑒 (𝑡) = ∫ 𝑓(𝜏)𝑑𝜏
0

The trapezoidal rule for numerical integration leads to a discrete-time

approximation 𝑦(𝑡𝑘 ) ≅ 𝑦𝑡𝑟𝑢𝑒 (𝑡)|𝑡𝑘 according to the recursion
ℎ
𝑦(𝑡𝑘+1 ) = 𝑦(𝑡𝑘 ) + [𝑓 (𝑡𝑘+1 ) + 𝑓 (𝑡𝑘 )]
2
where 𝑘 is an integer representing the iteration number, and 𝑦(0) = 0. The
idea is to approximate the incremental area under the curve using a
ℎ
trapezoid of area 2 [𝑓(𝑡𝑘+1 ) + 𝑓 (𝑡𝑘 )]. The trapezoid’s base, ℎ = 𝑡𝑘+1 −
𝑡𝑘 , is a small positive constant (sampling period, ℎ = 𝑇𝑠 ) defined by the
user, as shown in the figure below.
An analysis based on Taylor series expansion leads to the total relative
truncation error after 𝑛 iterations, at time 𝑡𝑘 = 𝑛ℎ, for (refer to Lecture 19
of Prof. Hassoun’s Numerical Methods course for detailed derivation)
We may compute an expression for the absolute relative error at 𝑡𝑘 as,

𝑦𝑡𝑟𝑢𝑒 (𝑡𝑘 ) − 𝑦(𝑡𝑘 ) 1 2 𝑓 ′ (𝑡𝑘 ) − 𝑓 ′ (0)

| |≅ ℎ | |
𝑦𝑡𝑟𝑢𝑒 (𝑡𝑘 ) 12 𝑦𝑡𝑟𝑢𝑒 (𝑡𝑘)

This result holds for any differentiable function 𝑓(𝑡). The important thing
to realize is that the trapezoidal rule has an error that is proportional to ℎ2 ;
i.e., the error is 𝑂(ℎ2 ). It is interesting to note that if 𝑓(𝑡) is linear (its
derivative is a constant for all 𝑡) then 𝑓 ′ (𝑡) − 𝑓 ′ (0) = 0. Thus, the
trapezoidal rule is exact when integrating linear functions.

If we assume the sinusoid 𝑓(𝑡) = sin(𝜔𝑜 𝑡), then the true value of the
integral is
𝑡
1
𝑦𝑡𝑟𝑢𝑒 (𝑡) = ∫ sin(𝜔𝑜 𝜏) 𝑑𝜏 = (1 − cos(𝜔𝑜 𝑡))
0 𝜔𝑜
and it can be easily shown that

𝑦𝑡𝑟𝑢𝑒 (𝑡𝑘 ) − 𝑦(𝑡𝑘 ) 1 2 2

| |≅ ℎ 𝜔𝑜
𝑦𝑡𝑟𝑢𝑒 (𝑡𝑘 ) 12

This last result makes qualitative sense: as 𝜔𝑜 increases it causes the

signal to have sharper variations and, therefore, the sampling interval, ℎ
must be reduced in order for the error not to increase. This result is
consistent with sampling theory (refer to Lecture 13).
Next, we treat the trapezoidal rule as a difference equation and solve it for
one specific function, 𝑓 (𝑡) = sin(𝜔𝑜 𝑡) 𝑢(𝑡). In order to arrive at the
difference equation, we transform the discrete-time signals 𝑦(𝑡𝑘 ) and
𝑓(𝑡𝑘 ) into the sequences 𝑦[𝑘] and 𝑓[𝑘].

The iterative version of the trapezoidal rule is given by,

ℎ
𝑦[𝑘 + 1] = 𝑦[𝑘] + {𝑓[𝑘 + 1] + 𝑓[𝑘]}
2
Now, we are ready to solve for 𝑦[𝑘]. Employing the Z-transform we
obtain
ℎ
𝑧𝑌(𝑧) − 𝑧𝑦[0] − 𝑌(𝑧) = [𝑧𝐹(𝑧) − 𝑧𝑓[0] + 𝐹 (𝑧)]
2
or, since we had assumed 𝑦[0] = 𝑦(0) = 0 and 𝑓 [0] = sin(0) = 0,
ℎ
(𝑧 − 1)𝑌(𝑧) = (𝑧 + 1)𝐹(𝑧)
2
which leads to
ℎ 𝑧+1
𝑌(𝑧) = 𝐹(𝑧)
2 𝑧−1
The transfer function of the discrete-time system is given by
𝑌(𝑧) ℎ 𝑧 + 1
𝐻 (𝑧) = =
𝐹 (𝑧) 2 𝑧 − 1
Letting 𝑓(𝑡) = sin(𝜔𝑜 𝑡)𝑢(𝑡) leads to the sampled input sequence 𝑓 [𝑘] =
sin(𝜔𝑜 ℎ𝑘)𝑢[ℎ𝑘] = sin(Ω𝑜 𝑘) 𝑢[𝑘], with Ω𝑜 ≜ 𝜔𝑜 ℎ. The Z-transform of
𝑓[𝑘] is (using Pair 11b with 𝛾 = 1)
𝑧 sin(Ω𝑜 )
𝐹 (𝑧) =
𝑧 2 − 2𝑧 cos(Ω𝑜 ) + 1
Then,
ℎ 𝑧+1 𝑧 sin(Ω𝑜 )
𝑌 (𝑧) = 𝐻(𝑧)𝐹 (𝑧) = ( ) 2
2 𝑧 − 1 𝑧 − 2𝑧 cos(Ω𝑜 ) + 1
𝑌 (𝑧 )
Performing partial fraction expansion on we obtain (try it)
𝑧

𝑌(𝑧) ℎ − sin(Ω𝑜 ) 1 ℎ sin(Ω𝑜 ) 𝑧 − cos(Ω𝑜 )

= ( ) + ( )
𝑧 2 cos(Ω𝑜 ) − 1 𝑧 − 1 2 cos(Ω𝑜 ) − 1 𝑧 2 − 2𝑧 cos(Ω𝑜 ) + 1

Finally, we employ the Z-transform Table to perform the inverse Z-

transform of 𝑌(𝑧) (Pairs 2 and 11a), and obtain the solution

−1 { ( )}
−ℎ sin(Ω𝑜 ) −1
𝑧 𝑧(𝑧 − cos(Ω𝑜 ))
[ ]
𝑦 𝑘 =𝑍 𝑌 𝑧 =( )𝑍 { − }
2(cos(Ω𝑜 ) − 1) 𝑧 − 1 𝑧 2 − 2𝑧 cos(Ω𝑜 ) + 1

−ℎ sin(Ω𝑜 )
𝑦[𝑘] = [1 − cos(Ω𝑜 𝑘)]𝑢[𝑘] (1)
2[cos(Ω𝑜 ) − 1]

By employing the earlier definition Ω𝑜 ≜ 𝜔𝑜 ℎ, we may write the (exact)

sampled solution of the integral as
1
𝑦𝑡𝑟𝑢𝑒 [𝑘] = (1 − cos(Ω𝑜 𝑘)) (2)
𝜔𝑜
For demonstration purposes, let us choose 𝜔𝑜 = 10 and ℎ = 0.08 (leading
to Ω𝑜 = ℎ𝜔𝑜 = 0.8. The following plot compares the exact solution in (2)
𝜋 𝜋
to the approximate one in (1), for 𝑡 ∈ [0 ]. Here, Ω represents half of
Ω𝑜 𝑜
the full period, 2𝜋/Ω𝑜 . The following Mathcad worksheet uses 𝑎 for Ω𝑜 .
We chose to plot the sampled exact solution as a continuous (red) curve
for clarity. Obviously, if we reduce the value of ℎ we improve the
approximation [recall that the error for the trapezoidal rule is proportional
to ℎ2 . The following figure compares the results for ℎ = 0.02, 0.08 and
0.2 over two periods of the input.
Comparison between the Transfer Function of an Integrator and the
Trapezoidal Rule-Based 𝐻(𝑧)
We saw earlier that the trapezoidal integration rule can be viewed as a
discrete-time LTI system with transfer function
𝑇 𝑧+1
𝐻 (𝑧) = (1)
2 𝑧−1
The analog (exact) integrator transfer function is determined by applying
the Laplace transform to the integration
𝑡
𝑦(𝑡) = ∫ 𝑓(𝜏)𝑑𝜏
0−

which by means of the integration property of the Laplace transform gives

1
𝑌(𝑠) = 𝐹(𝑠)
𝑠
Therefore, the integrator transfer function is
𝑌 (𝑠) 1
𝐻(𝑠) = = (2)
𝐹 (𝑠) 𝑠
Comparing (1) and (2) we conclude that we can arrive at 𝐻(𝑧) by using
2 𝑧−1
the mapping, 𝑠 = 𝑇 𝑧+1. This is called the bilinear transformation and will
be used in Lecture 24 in connection with digital filter design. We can also
2+𝑠𝑇
express the inverse of this transformation (your turn) as 𝑧 = 2−𝑠𝑇 and
show that the left half 𝑠-plane is mapped inside the unit circle in the 𝑧-
plane.
Your Turn: Write a Matlab (or Mathcad) script that generates 1,000
random points in the left half 𝑠-plane (including the 𝑗𝜔 axis) and maps and
plots these points in the 𝑧-plane. Assume 𝑇 = 1 𝑠𝑒𝑐. According to your
plot, where does the whole left half 𝑠-plane map into the 𝑧-plane?

We may also map a whole analog system to a discrete-time system using

the bilinear transformation as demonstrated in the following example.

2
Example. Let 𝐻 (𝑠) = 𝑠+1 and 𝑇 = 0.1. Then, we can arrive at the
corresponding discrete-time transfer function using the substitution
2𝑧−1
𝑠=
𝑇𝑧 +1
to get
2 2
2 𝑧 + 21
21
𝐻1 (𝑧) = 𝐻(𝑠)|𝑠= 2 𝑧−1 = 𝑧−1
= 19
0.1 𝑧+1 20 (𝑧+1) + 1 𝑧−
21

In Lecture 20, we employed the mapping 𝑧 = 𝑒 𝑠𝑇 and arrived at the

following discrete-time system transfer function approximation for the
2
analog system 𝐻(𝑠) = 𝑠+1 (in the limit of very small 𝑇 and the use of
Taylor series approximation)
2𝑇𝑧 2𝑇
𝐻2 (𝑧) = =
𝑧 − 𝑒 −𝑇 1 − 𝑒 −𝑇 𝑧 −1

The following Mathcad simulation compares the step-response of the

analog system, 𝐻(𝑠), and its two discrete-time system representations,
𝐻1 (𝑧) and 𝐻2 (𝑧). It is interesting to note that the bilinear approximation is
more accurate than the Taylor expansion-based approximation.
Your turn: Repeat the above simulations for the transfer function 𝐻 (𝑠) =
1
with 𝑇 = 0.1.
𝑠 2 +1

Your turn:
a. Show that Euler’s integration rule
𝑦[𝑘 + 1] = 𝑦[𝑘] + 𝑇𝑓[𝑘 + 1]
1
Leads to the mapping 𝑧 = 1−𝑠𝑇.

b. Write a Matlab (or Mathcad) script that generates 1,000 random

points in the left half 𝑠-plane (including the 𝑗𝜔 axis) and maps and
1
plots these points in the 𝑧-plane according to 𝑧 = 1−𝑠𝑇. Assume 𝑇 =
1 𝑠𝑒𝑐. According to your plot, where does the whole left half 𝑠-plane
map into the 𝑧-plane?
 Convolution Sums and LTI System Analysis

Note: For Pair 1 in the above table, 𝑗 is a positive integer.

Just as we employed the convolution integral in the continuous-time

domain to obtain the zero-state response of a LTI system, we can employ
convolution sums to obtain the zero-state response of a discrete-time
system, 𝑦𝑧𝑠 [𝑘] = 𝑓 [𝑘] ∗ ℎ[𝑘]. The convolution sum generates a discrete-
time signal according to the expression (refer to the section before last for
derivation.)
∞

𝑦𝑧𝑠 [𝑘] = 𝑓[𝑘] ∗ ℎ[𝑘] = ∑ 𝑓[𝑛]ℎ[𝑘 − 𝑛]

𝑛=−∞
It should be emphasized that the convolution operation is linear, which
means that the commutative property, distributive (over addition and
subtraction) property, associative property, scaling property, shift property
and superposition property apply. The following example illustrates the
convolution analysis method.

Example. Employ convolution sums from the above Table to solve for the
zero-state response of a LTID system to the input 𝑓 [𝑘] = 2𝑘𝑢[𝑘]. Assume
1 𝑘
the system’s unit-impulse response is given by ℎ[𝑘] = 𝛿 [𝑘] + (2) 𝑢[𝑘].

Solution. In the following we employ the distributive property of

convolution, Convolution Pair 1 (with 𝑗 = 0) and Pair 6 (with 𝛾1 = 1)

1 𝑘
𝑦𝑧𝑠 [𝑘] = 𝑓[𝑘] ∗ ℎ[𝑘] = 2𝑘𝑢[𝑘] ∗ [𝛿 [𝑘] + ( ) 𝑢[𝑘]]
2

1 𝑘
= 2𝑘𝑢[𝑘] ∗ 𝛿 [𝑘] + 2𝑘𝑢[𝑘] ∗ ( ) 𝑢[𝑘]
2
1 1
(1) ( ) 1 𝑘 1−2
2 𝑘
= 2𝑘𝑢[𝑘] + 2 2 ((2) − 1 + 1 𝑘 (1)𝑘 ) 𝑢(𝑘)
1
(1 − 2) 2

1 𝑘
= 2𝑘𝑢[𝑘] + 4 (( ) − 1 + 𝑘) 𝑢[𝑘]
2

1 𝑘
𝑦𝑧𝑠 [𝑘] = (4 ( ) + 6𝑘 − 4) 𝑢[𝑘]
2

Mathcad verification employing the 𝑧-transform method:

Example: Employ the definition of the convolution sum to solve for the
zero-state sequence for the three-point moving average system.
Solution: The input signal and the impulse response are (as determined
earlier) given by
1
ℎ[𝑘] = ([𝛿 ] + 𝛿 [𝑘 − 1] + 𝛿 [𝑘 − 2])
3
𝑓[𝑘] = 𝛿[𝑘] + 2𝛿[𝑘 − 1] − 𝛿[𝑘 − 2] + 𝛿[𝑘 − 3] + 2𝛿[𝑘 − 4]
The zero-state response is given by
∞

𝑦[𝑘] = 𝑓[𝑘] ∗ ℎ[𝑘] = ∑ 𝑓 [𝑛]ℎ[𝑘 − 𝑛]

𝑛=−∞

Which can be written as (due to the causal and finite nature of the input),
4

𝑦[𝑘] = ∑ 𝑓 [𝑛]ℎ[𝑘 − 𝑛]
𝑛=0

For 𝑘 = 0, we obtain
4

𝑦[0] = ∑ 𝑓[𝑛]ℎ[−𝑛] = 𝑓[0]ℎ[0] + 𝑓 [1]ℎ[−1] + 𝑓[2]ℎ[−2] + ⋯ + 𝑓[4]ℎ[−4]

𝑛=0
 1 1
[ ] ( ) ( )( ) ( )( ) ( )( ) ( )(
𝑦 0 = 1 ( ) + 2 0 + −1 0 + 1 0 + 2 0 = )
3 3
For 𝑘 = 1:
4

𝑦[1] = ∑ 𝑓[𝑛]ℎ[1 − 𝑛] = 𝑓[0]ℎ[1] + 𝑓[1]ℎ[0] + 𝑓[2]ℎ[−1] + ⋯ + 𝑓[4]ℎ[−3]

𝑛=0
1 1
= (1) ( ) + (2) ( ) + (−1)(0) + (1)(0) + (2)(0) = 1
3 3
For 𝑘 = 2:
4

𝑦[2] = ∑ 𝑓[𝑛]ℎ[2 − 𝑛]
𝑛=0
= 𝑓 [0]ℎ[2] + 𝑓[1]ℎ[1] + 𝑓[2]ℎ[0] + 𝑓 [3]ℎ[−1] + 𝑓[4]ℎ[−2]
1 1 1 2
( ) ( ) ( ) ( )(
= 1 ( ) + 2 ( ) + −1 ( ) + 1 0 + 2 0 = ) ( )( )
3 3 3 3

Your turn: Finish the above example by determining 𝑦[3], 𝑦[4] and 𝑦[5].

Your turn: Employ the Table of Convolution Sums to determine the zero-
1 𝑘
state response of the system ℎ[𝑘] = (2) 𝑢[𝑘] to the following input 𝑓[𝑘]:

a. 𝑢[𝑘]

1 𝑘
b. (3) 𝑢[𝑘]

1 𝑘−2 1 𝑘−4
c. 5 (3) 𝑢[𝑘 − 2] − 4 (3) 𝑢[𝑘 − 4]

Hint: employ the result from Part b, shift and superposition.

 +1 𝑖𝑓 𝑘 = 0
−1 𝑖𝑓 𝑘 = 1
d. {
−1 𝑖𝑓 𝑘 = 2
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
Hint: express the function as the sum of unit-steps and use the result
form Part a, shift and superposition.

Your turn: Verify your solutions to the above problem employing the 𝑧-
tarnsform and Mathcad. Plot the responses as stem plots.
 Derivation of the Convolution Sum for the Zero-State Response
We start by assuming a LTID system with unit-impulse response ℎ[𝑘].
That is, for the input 𝛿[𝑘] the system responds with ℎ[𝑘],
𝛿 [𝑘] → 𝑠𝑦𝑠𝑡𝑒𝑚 → ℎ[𝑘]
We are interested in showing that
∞

𝑦𝑧𝑠 [𝑘] = 𝑓[𝑘] ∗ ℎ[𝑘] = ∑ 𝑓[𝑛]ℎ[𝑘 − 𝑛]

𝑛=−∞

Let us express the signal 𝑓[𝑘] as the sum

𝑓[𝑘] = ⋯ + 𝑓[−1]𝛿 [𝑘 + 1] + 𝑓 [0]𝛿 [𝑘] + 𝑓[1]𝛿 [𝑘 − 1] + ⋯
∞

= ∑ 𝑓[𝑛]𝛿[𝑘 − 𝑛]
𝑛=−∞

Next, we employ the scaling and shift properties of linear systems and find
the response to single sample input 𝑓 [𝑛]𝛿[𝑘 − 𝑛] to be 𝑓[𝑛]ℎ[𝑘 − 𝑛].

𝑓[𝑛]𝛿[𝑘 − 𝑛] → 𝑠𝑦𝑠𝑡𝑒𝑚 → 𝑓[𝑛]ℎ[𝑘 − 𝑛]

Finally, applying the superposition property we obtain the zero-state

response as the convolution sum
∞

𝑦𝑧𝑠 [𝑘] = ∑ 𝑓 [𝑛]ℎ[𝑘 − 𝑛]

𝑛=−∞
 Derivation of the Convolution Property of the Z-transform
In a similar fashion to the way we had derived the convolution property of
the Laplace transform we shall derive the convolution property for the 𝑧-
transform.
Assuming causal signals and Z-transforming the convolution sum that we
have just derived, we obtain
∞ ∞

𝑌𝑧𝑠 (𝑧) = ∑ ∑ 𝑓[𝑛]ℎ[𝑘 − 𝑛] 𝑧 −𝑘

𝑘=0 𝑛=0

Interchanging the order of the summations in the above expression gives

∞ ∞

𝑌𝑧𝑠 (𝑧) = ∑ 𝑓[𝑛] ∑ ℎ[𝑘 − 𝑛] 𝑧 −𝑘

𝑛=0 𝑘=0

The right hand side sum is the 𝑧-transform of a right-shifted signal by 𝑛

units, and can be represented as (employing the shift property of the 𝑧-
transform)
∞

∑ ℎ[𝑘 − 𝑛] 𝑧 −𝑘 = 𝑧 −𝑛 𝐻(𝑧)
𝑘=0

Substituting the above sum in the expression for 𝑌𝑧𝑠 (𝑧) gives
∞ ∞

𝑌𝑧𝑠 (𝑧) = ∑ 𝑓[𝑛]𝐻(𝑧) 𝑧 −𝑛 = 𝐻(𝑧) ∑ 𝑓 [𝑛] 𝑧 −𝑛 = 𝐻(𝑧)𝐹(𝑧)

𝑛=0 𝑛=0

We have just proved that

𝑓[𝑛] ∗ ℎ[𝑛] ↔ 𝐹(𝑧)𝐻(𝑧)