Introduction to Signal Processing

Lecture 5: The z-transform

Filip Elvander

Dept. Information and Communications Engineering

Aalto University
0.
Today’s lecture

Discrete-time signals • We continue our journey through the

(Ch. 2) non-time-domain world.
Discrete-time systems • Last week was frequency domain,
(Ch. 2) now we go to the so-called z-domain.
• Main concept: the z-transform.
Frequency analysis of signals
(Ch. 4 and 6.1 (Ch. 7 later)) • Allows us to compute the output from
an LTI system in an easy way.
z-transform
• Convenient analysis of system
(Ch. 3)
properties (next lecture).
Freq. analysis of systems and filters • Later in the course: tool for designing
(Ch. 5 and 7) our own systems.
Implementation of systems • Reading: Chapter 3.
(Ch. 9) You can now find the first hand-in
assignment on MyCourses. Note that
Filter design
parts of it is mandatory for course
(Ch. 10)
completion.
March 10, 2025
0.
The Laplace transform
In your course in signals and systems, you have seen the Laplace transform:
ˆ ∞
L ( f ) (s) = f (t)e−st dt,
−∞

where s = σ + jω, with σ ∈ R, ω ∈ R.

• Transforms functions of time to functions in complex plane.
• Time-domain → s-domain.
• The LT is not finite everywhere (for all s): we talk about the region of
convergence (ROC) as the set of s where the integral converges.
• The LT is both the algebraic expression L ( f ) (s) and the ROC: f1 (t) and
f2 (t) may have the same L ( f ) (s), but then have different ROC.
• Uses: gives very easy way of solving ordinary differential equations.
Today, we will see the discrete-time analog: the z-transform.

March 10, 2025

0.
The z-transform
The z-transform X(z) of a discrete-time signal x(n) is defined as
∞
X(z) = ∑ x(n)z−n = . . . + x(−2)z2 + x(−1)z + x(0) + x(1)z−1 + x(2)z−2 + . . .
n=−∞

where z ∈ C
• The z-transform maps sequences (signals) x(n) to complex-valued
functions X(z) ≡ Z (x(n)).
• X(z) is a power series and may not converge for all z ∈ C.
• The values of z for which X(z) exists (for which the series converges) is
called the region of convergence (ROC):
( )
∞
ROC = z∈C| ∑ x(n)z−n < ∞
n=−∞

March 10, 2025

0.
x(n) and X(z)
Note: X(z) is a complex-valued function. Often, we plot the magnitude |X(z)|.

-1

-2
0 1 2 3 4 5 6 7
n

March 10, 2025

0.
A simple signal
Let’s take the signal x(n) = an u(n), where a ∈ C. Then,
∞ ∞ ∞
X(z) = Z (x(n)) = ∑ an u(n)z−n = ∑ an z−n = ∑ (az−1 )n
n=−∞ n=0 n=0
1
= , az−1 < 1 ⇐⇒ |z| > |a| .
1 − az−1
We get the ROC {z ∈ C | |z| > |a|}.

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 5 10 15 20
n
March 10, 2025
0.
A simple signal
Let’s take the signal x(n) = an u(n), where a ∈ C. Then,
∞ ∞ ∞
X(z) = Z (x(n)) = ∑ an u(n)z−n = ∑ an z−n = ∑ (az−1 )n
n=−∞ n=0 n=0
1
= , az−1 < 1 ⇐⇒ |z| > |a| .
1 − az−1
We get the ROC {z ∈ C | |z| > |a|}.

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 5 10 15 20
n
March 10, 2025
0.
A simple signal (2)
Let’s try another signal: x(n) = −an u(−n − 1), where a ∈ C. Then,
∞ −1 ∞
X(z) = Z (x(n)) = ∑ −an u(−n − 1)z−n = ∑ −an z−n = − ∑ a−k zk
n=−∞ n=−∞ k=1
∞
1 1 1
= −(a−1 z) ∑ (a−k z)k = −(a−1 z) −1 z
= −1 −1
k=0 1 − a az a z−1
1
= , a−1 z < 1 ⇐⇒ |z| < |a| .
1 − az−1
We get the ROC {z ∈ C | |z| < |a|}.
10

-2

-4

-6

-8

-10
-20 -15 -10 -5 0
n

March 10, 2025

0.
A simple signal (2)
Let’s try another signal: x(n) = −an u(−n − 1), where a ∈ C. Then,
∞ −1 ∞
X(z) = Z (x(n)) = ∑ −an u(−n − 1)z−n = ∑ −an z−n = − ∑ a−k zk
n=−∞ n=−∞ k=1
∞
1 1 1
= −(a−1 z) ∑ (a−k z)k = −(a−1 z) −1 z
= −1 −1
k=0 1 − a az a z−1
1
= , a−1 z < 1 ⇐⇒ |z| < |a| .
1 − az−1
We get the ROC {z ∈ C | |z| < |a|}.
10

-2

-4

-6

-8

-10
-20 -15 -10 -5 0
n

March 10, 2025

0.
The ROC is part of the z-transform

• Two completely different signals had the same z-transform in the sense
that X(z) ”looked” the same.
• In order to connect X(z) to a signal x(n), the ROC needs to be specified.
• z-transform = X(z) + ROC.
Let’s remember this particular z-transform (will be useful for us later):
z 1
x(n) = an u(n) ←→ X(z) = , ROC = {z | |z| > |a|}
1 − az−1

1 7
1

0.9 0.8
6
0.8 0.6

0.7
5
0.4

0.6 0.2
4
0.5 0
3
0.4 -0.2

0.3 -0.4
2
0.2 -0.6
1
0.1 -0.8

0 0 -1
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
n n n
0<a<1 a>1 −1 < a < 0

March 10, 2025

0.
Finite length signals
Let x(n) = 0 for n < −N1 and n > N2 . Then,
∞ N2 −1 N2
X(z) = ∑ x(n)z−n = ∑ x(n)z−n = ∑ x(n)z−n + ∑ x(n)z−n
n=−∞ n=−N1 n=−N1 n=0
N1 N2
= ∑ x(−n)zn + ∑ x(n)z−n .
n=1 n=0

Both terms are finite for all z ∈ C except potentially at z = 0 and ”z = ∞”.
• The anti-causal part converges for all z except at infinity.
• The causal part converges for all z except at zero.

March 10, 2025

0.
Causal signals
Note that if z ∈ C, then z = re jϕ , where r > 0 and ϕ ∈ R. Also, |z| = r. Consider
a causal signal x(n). Then,
∞
X(z) = Z (x(n)) = ∑ x(n)z−n .
n=0
We have
∞ ∞ ∞
|X(z)| = ∑ x(n)z−n ≤ ∑ x(n)z−n = ∑ |x(n)| z−n
n=0 n=0 n=0
∞ ∞ |x(n)|
= ∑ |x(n)| r−n = ∑ n
.
n=0 n=0 r Im(z)
⇒ if we choose r big enough, the terms
decay fast enough for summation.
R
ROC = {z | |z| > R}

Note: R could be infinity. Special case: Re(z)

x(n) = an u(n) ⇒ R = |a|

March 10, 2025

0.
Anti-causal signals
Let x(n) be anti-causal, i.e., x(n) = 0 for n ≥ 0. Then,
−1 ∞
X(z) = Z (x(n)) = ∑ x(n)z−n = ∑ x(−n)zn
n=−∞ n=1
In the same way as for the causal signal,
∞
|X(z)| ≤ ∑ |x(−n)| rn
n=1
Im(z)
⇒ if we choose r small enough, the terms
decay fast enough for summation.

ROC = {z | |z| < R} for some R > 0.

R
Special case: Re(z)

x(n) = −an u(−n − 1) ⇒ R = |a|

March 10, 2025

0.
General infinite length signals
For a general x(n), we have
∞ −1 ∞
X(z) = Z (x(n)) = ∑ x(n)z−n = ∑ x(n)z−n + ∑ x(n)z−n
n=−∞ n=−∞ n=0
∞ ∞
= ∑ x(−n)zn + ∑ x(n)z−n
n=1 n=0
| {z } | {z }
r = |z| has to be small enough r = |z| has to be big enough

• Causal ROC: |z| > R1 for some R1 > 0.

• Anti-causal ROC: |z| < R2 for some R2 > 0. Im(z)

R2
R1
⇒ the total sum converges for |z| ∈ (R1 , R2 ).
Note: if R1 ≥ R2 , then the ROC is empty. Re(z)

March 10, 2025

0.
Region of convergence

• Finite-length signals:

C \ {0}
 if causal
ROC = C \ {∞} if anti-causal

C \ {0, ∞} otherwise


• Infinite length signals:


{z ∈ C | |z| > R1 }
 if causal
ROC = {z ∈ C | |z| < R2 } if anti-causal

{z ∈ C | R1 < |z| < R2 } otherwise


for some R1 > 0, R2 > 0.

March 10, 2025

0.
ROC, infinite length signals


{z ∈ C | |z| > R1 }
 if causal
ROC = {z ∈ C | |z| < R2 } if anti-causal

{z ∈ C | R1 < |z| < R2 } otherwise


Im(z)
Im(z) Im(z)

R2
R1 R1
R2
Re(z) Re(z) Re(z)

March 10, 2025

0.
Inversion of the z-transform
Recall from last week’s lecture that a transform is only useful in practice if it is
invertible:
x(n) → X(z) , and X(z) → x(n).
We have X(z) = ∑∞ −n
by definition. Can we get back x(n) from X(z)?
n=−∞ x(n)z
Let C be a closed curve going counter-clockwise around the origin and
entirely contained in the ROC. Then,
‰ ‰ ∞
! ‰ ∞
n−1 −k
C
X(z)z dz =
C
∑ x(k)z zn−1 dz = ∑ x(k)zn−k−1 dz C k=−∞
k=−∞
∞ ‰
C in ROC
= ∑ x(k) zn−k−1 dz = j2πx(n).
k=−∞ C
| {z }
j2π if n = k

The last equality is the Cauchy integral theorem. Thus, the inverse
z-transform is
‰
1
x(n) = Z −1 (X(z)) = X(z)zn−1 dz , C closed curve in ROC.
j2π C
In practice, we will compute inverse z-transforms in easier ways.
March 10, 2025
0.
From Lecture 2: the delay operation
Recall that

x(n) x(n − 1)
z−1

Why do we use this strange notation? In fact,

∞ ∞
Z (x(n − k)) = ∑ x(n − k)z−n = z−k ∑ x(n − k)z−(n−k)
n=−∞ n=−∞
∞
= z−k ∑ x(n)z−n = z−k X(z)
n=−∞

So, this is a short-hand for the following sequence of operations:

x(n) X(z) z−1 X(z) x(n − 1)

Z () z−1 Z −1 ()

March 10, 2025

0.
Linearity of the z-transform and its inverse
Let X1 (z) = Z (x1 (n)) and X2 (z) = Z (x(n)). Then, for any α, β ∈ C,
∞
Z (αx1 (n) + β x2 (n)) = ∑ (αx1 (n) + β x2 (n)) z−n
n=−∞
∞ ∞
=α ∑ x1 (n)z−n + β ∑ x2 (n)z−n
n=−∞ n=−∞
= αX1 (n) + β X2 (n).

The ROC is the intersection of the ROCs of X1 (z) and X2 (z). The same holds
for the inverse:

Z −1 (αX1 (n) + β X2 (n)) = αZ −1 (X1 (z)) + β Z −1 (X2 (z)) = αx1 (n) + β x2 (n).

This is very useful: if we have a complicated z-transform X(z) that can be

written as
K K
X(z) = ∑ αk Xk (z) , then x(n) = Z −1 (X(z)) = ∑ αk Z −1 (Xk (z))
k=1 k=1

where Z −1 (Xk (z)) might be easy to find.

March 10, 2025
0.
Easy to invert
We have just seen that

z 1
x(n) = an u(n) ←→ X(z) = with ROC |z| > |a|
1 − az−1
and
z 1
x(n) = −an u(−n − 1) ←→ X(z) = with ROC |z| < |a|.
1 − az−1
⇒ if we are able to decompose an X(z) into these types of components,
inversion is easy! By linearity of Z −1 (), we can simply invert each component
and sum them up.
It turns out that an important class of X(z) lends itself to this: rational
transforms.

March 10, 2025

0.
Important class: rational z-transforms
A common type of z-transform is the rational function:

B(z) ∑M bℓ z−ℓ z−M ∑M bℓ zM−ℓ

X(z) = = Nℓ=0 −k = −N ℓ=0 .
A(z) ∑k=0 ak z z ∑N
k=0 ak z
N−k

By the fundamental theorem of algebra, ∃p1 , . . . , pN , and q1 , . . . , qM , such that

N N M M
∑ ak zN−k = a0 ∏ (z − pk ) , ∑ bℓ zM−ℓ = b0 ∏ (z − qℓ ),
k=0 k=1 ℓ=0 ℓ=1

yielding

b0 ∏M
ℓ=1 (z − qℓ )
X(z) = zN−M N
.
a0 ∏k=1 (z − pk )

• Typically, we assume M < N: otherwise we can factor out a polynomial

from the rational function.
• q1 , . . . , qM are called the zeros and p1 , . . . , qN are called the poles (more
on this later).
March 10, 2025
0.
Partial fraction decomposition
For simplicity, let a0 = 1. If M < N, we can write
b0 zN−1 + b1 zN−2 + . . . + bM zN−M−1
z−1 X(z) = . (1)
∏N k=1 (z − pk )
If the poles p1 , . . . , pN are distinct, then we can do a partial fractional
decomposition (PFD):
A1 A2 AN
z−1 X(z) = + +...+
z − p1 z − p2 z − pN
for some numbers A1 , . . . , AN . You can find these numbers by
cross-multiplication:
N−1
A1 AN A1 ∏N N
k=2 (z − pk ) + A2 ∏k=1,k̸=2 (z − pk ) + . . . + AN ∏k=1 (z − pk )
+...+ =
z − p1 z − pN ∏N
k=1 (z − pk )
and compare the coefficients with the right-hand side of (1).
• If you don’t recall this, have a look at your old calculus course. Suomi:
Osamurtokehitelmä. Svenska: partialbråksuppdelning.
• If the poles are not distinct (have multiplicity) the PFD will have to look
slightly different. Again, see your calculus book.
March 10, 2025
0.
Simple inversion
Now that we have
A1 AN A1 AN
z−1 X(z) = +...+ ⇒ X(z) = +...+ ,
z − p1 z − pN 1 − p1 z−1 1 − pN z−1
we can use the linearity of the inverse transform to get
   
1 1
x(n) = Z −1 (X(z)) = A1Z −1 + . . . + AN Z −1
.
1 − p1 z−1 1 − pN z−1

Also, we have to decide on a ROC. Recall

  ( n
1 pk u(n) if ROC |z| > |pk | (causal)
Z −1 =
1 − pk z−1 −pnk u(−n − 1) if ROC |z| < |pk | (anti-causal).

If we decide on a causal signal, the ROC is |z| = maxk |pk | and

x(n) = A1 pn1 u(n) + . . . + AN pnN u(n).

Example later!

March 10, 2025

0.
Connection to the DTFT
Recall from last week that the discrete-time Fourier transform (”analysis
equation”) was

XDT FT (ω) = ∑ x(n)e− jωn , ω ∈ (−π, π].

n=−∞

The z-transform evaluated at z = e jω is

 −n
X(z) jω
= ∑ x(n) e jω = ∑ x(n)e− jωn = XDT FT (ω).
z=e n=−∞ n=−∞

This makes sense if {z | |z| = 1} is in the ROC! For inversion, we can pick
C = {z | |z| = 1}, i.e., the unit circle. This yields
‰ ˆ π
e jω = z
 
1 1
Z −1 (X(z)) = X(z)zn−1 dz = = X(e jω )e jωn dω
j2π C je jω dω = dz 2π −π
ˆ π
1
= XDT FT (ω)e jωn dω = x(n).
2π −π
| {z }
synthesis eq.

The DTFT is the z-transform on the unit circle.

March 10, 2025
0.
So why are we interested in the z-transform?
Because it is perfect for analyzing LTI systems! First, let’s check some
properties.
Time-delay
∞ ∞
Z (x(n − k)) = ∑ x(n − k)z−n = z−k ∑ x(n − k)z−(n−k)
n=−∞ n=−∞
∞
= z−k ∑ x(n)z−n = z−k X(z)
n=−∞

Convolution Let X1 (z) = Z (x1 (n)) and X2 (z) = Z (x2 (n)). Then,
!
∞ ∞ ∞ ∞
Z (x1 (n) ∗ x2 (n)) = ∑ ∑ x1 (k)x2 (n − k) z−n = ∑ x1 (k) ∑ x2 (n − k)z−n
n=−∞ k=−∞ k=−∞ n=−∞
| {z }
∞ ∞
−k z−k X2 (z)
= ∑ x1 (k)z X2 (z) = X2 (z) ∑ x1 (k)z−k
k=−∞ k=−∞
= X2 (z)X1 (z).
⇒ Convolution transforms into simple multiplication!
March 10, 2025
0.
Polynomial multiplication and convolution
It is no surprise that convolution becomes multiplication with the z-transform:
everything is polynomials! Let’s say that you have two polynomials in the real
variable x:

p1 (x) = a0 + a1 x + x2 x2 , p2 (x) = b0 + b1 x + b2 x2

Then, p(x) = p1 (x)p2 (x) is also a polynomial:

p(x) = c0 + c1 x + c2 x2 + c3 x3 + c4 x4 .

How do we get the coefficients? By convolving the sequences a0 , a1 , a2

and b0 , b1 , b2 :
∞
cn = ∑ ak bn−k .
k=−∞

For the z-transform,


X(z) ↔ p(x) , x(n) ↔ c0 , c1 , c2 , c3 , c4

March 10, 2025

0.
LTI systems
Recall, we can get the output of an LTI system by convolving the input with
the impulse response:
∞
y(n) = x(n) ∗ h(n) = ∑ x(k)h(n − k).
k=−∞

We have

Y (z) = Z (y(n)) = Z (x(n) ∗ h(n)) = Z (x(n)) Z (h(n)) = X(z)H(z).

This means that we can get the output y(n) as follows:

1. Compute X(z) = Z (x(n)) and H(z) = Z (h(n)) (time domain → z-domain)
2. Compute Y (z) = X(z)H(z) (z-domain)
3. Invert z-transform: y(n) = Z −1 (Y (z)) (z-domain → time-domain)
Super easy (kind of)!

March 10, 2025

0.
Stability and causality
Let H(z) = ∑∞ −n
n=−∞ h(n)z .
Stability
Recall that a system is stable if and only if ∑∞
n=−∞ |h(n)| < ∞.
∞ ∞
|H(z)| = ∑ h(n)z−n ≤ ∑ |h(n)| z−n
n=−∞ n=−∞
∞
⇒ |H(z)| ≤ ∑ |h(n)| . (stability implies unit circle in ROC)
|z|=1 n=−∞
In fact, the converse is also true. Conclusion: a system is stable if and only if
the unit circle {z | |z| = 1} is contained in the ROC:
system stable ⇐⇒ {z | |z| = 1} ⊂ ROC
Causality
Causal system ⇐⇒ h(n) = 0 for n < 0. We then have that a system is causal
if and only if ROC = {z | |z| > R} for some R.
Causal and stable
An LTI system is causal and stable if and only if the region of convergence is
ROC = {z | |z| > R} , R < 1
March 10, 2025
0.
Example
Let y(n) = ay(n − 1) + x(n) , y(−1) = 0, where x(n) = cu(n)

x(n) y(n)
+
∞
1
Z (u(n)) = ∑ z−n = 1 − z−1
n=0
a z−1

How to compute y(n)?

1. Take z-transform of both sides:
Y (z) = az−1Y (z) + X(z) ⇐⇒ (1 − az−1 )Y (z) = X(z).
1 1 c
2. Solve for Y (z) : Y (z) = X(z) = 1−az−1 1−z−1 .
1 − az−1
| {z }
H(z)
 
3. Invert: y(n) = Z −1 1 b
1−az−1 1−z−1
.
But how do we do the inversion?
z 1
• We saw before that y(n) = an u(n) ←→ Y (z) = 1−az−1
with ROC |z| > |a|
• Can we get it on this form in some way?
March 10, 2025
0.
Partial fraction decomposition
Recall from calculus that if a ̸= b, then ∃A, B such that
1 A B
= + . (2)
(1 − az−1 )(1 − bz−1 ) 1 − az−1 1 − bz−1

Let’s try:

A B A(1 − bz−1 ) + B(1 − az−1 ) (A + B) − (Ab + Ba)z−1

−1
+ −1
= = .
1 − az 1 − bz (1 − az−1 )(1 − bz−1 ) (1 − az−1 )(1 − bz−1 )

The only way this matches the left-hand side of (2) for all z is if the
coefficients match:
(
a
A = a−b
A + B = 1 , −(Ab + Ba) = 0 ⇒ b
B = 1 − A = − a−b

We get
1 a 1 b 1
= − .
(1 − az−1 )(1 − bz−1 ) a − b 1 − az−1 a − b 1 − bz−1

March 10, 2025

0.
Example – continued
In our case a ↔ a and b ↔ 1, yielding
1 c 1
Y (z) = −1 −1
=c
1 − az 1 − z (1 − az )(1 − z−1 )
−1
 
a 1 1 1
=c − .
a − 1 1 − az−1 a − 1 1 − z−1

The inverse z-transform is linear, and thus

   
ca 1 c 1
y(n) = Z −1 (Y (z)) = Z −1 − Z −1
a−1 1 − az−1 a−1 1 − z−1
| {z } | {z }
an u(n) u(n)
ca n c
= a u(n) − u(n).
a−1 a−1
Note:
• We have implicitly used ROC = {z | |z| > |a|} as to get a causal solution.
• Recall from before, this type of system is stable if and only if |a| < 1.

March 10, 2025

0.
Systems defined by difference equations
Recall that we could define an LTI system through the difference equation
N M
∑ ak y(n − k) = ∑ bℓ x(n − ℓ) , a0 = 1.
k=0 ℓ=0

• These systems can be implemented easily – practically useful.

• Recall that we could compute the IR h(n) by solving the equation with
x(n) = δ (n).

x(n) y(n)
b0 + +

z−1 b1 + + −a1 z−1

z−1 b2 + + −a2 z−1

z−1 b3 +

March 10, 2025

0.
Difference equation ↔ rational z-transform
Let’s compute the z-transform of both sides:
N M N M
z
∑ ak y(n − k) = ∑ bℓ x(n − ℓ) ←→ ∑ ak z−kY (z) = ∑ bℓ z−ℓ X(z),
k=0 ℓ=0 k=0 ℓ=0

This gives
−ℓ
N M
∑M
ℓ=0 bℓ z
Y (z) ∑ ak z−k = X(z) ∑ bℓ z−ℓ ⇐⇒ Y (z) = N a z−k
X(z).
k=0 ℓ=0 ∑k=0 k

That is, the system function H(z) is the rational function

−ℓ
∑M
ℓ=0 bℓ z z−M ∑M bℓ zM−ℓ ∑M bℓ zM−ℓ
H(z) = N a z−k
= −N ℓ=0
N
= zN−M ℓ=0
∑k=0 k z ∑k=0 ak z N−k ∑N
k=0 ak z
N−k

• The numerator is a M-order polynomial in z.

• The denominator is a N-order polynomial in z.

March 10, 2025

0.
Factoring H(z): poles and zeros
The fundamental theorem of algebra says that we can factor
N N M M
∑ ak zN−k = ∏ (z − pk ) , ∑ bℓ zM−ℓ = b0 ∏ (z − qℓ ),
k=0 k=1 ℓ=0 ℓ=1
yielding
b0 ∏M
ℓ=1 (z − qℓ )
H(z) = zN−M .
∏Nk=1 (z − pk )

• The roots q1 , . . . , qM of the numerator are called the zeros: H(qℓ ) = 0.

• The roots p1 , . . . , pN of the denominator are called the poles: ”H(pk ) = ∞”
Stability?
• A system is stable if |z| = 1 is in the ROC.
• As we want causality, we need ROC = {z | |z| > R} for some R < 1.
• But H(z) explodes for z = pk , k = 1, . . . , N.
• For this to fit together all poles must be within the unit disk:
causal system stable ⇐⇒ |pk | < 1 , k = 1, . . . , N.

March 10, 2025

0.
Poles and stability

Im(z)

|p|
Re(z)

March 10, 2025

0.
Poles and stability

Im(z)

|p|
Re(z)

March 10, 2025

0.
Computing the output for rational H(z)
Let both the input x(n) and IR h(n) have rational z-transforms:

R(z) B(z) R(z)B(z)

X(z) = , H(z) = ⇒ Y (z) = .
Q(z) A(z) A(z)Q(z)

Let p1 , . . . , pN and q1 , . . . , qL be the zeros of A(z) and Q(z), i.e., the poles of
Y (z), and assume that they are distinct. Then,
N L
A Q
Y (z) = ∑ 1 − pkk z−1 + ∑ 1 − qℓℓz−1 , for some Ak , Qℓ .
k=1 ℓ=1

Assuming a causal system, we get

N L
y(n) = ∑ A1 pnk u(n) + ∑ Qℓ qnℓ u(n)
k=1 ℓ=1
| {z } | {z }
natural response forced response
• If |pk | < 1, the natural response dies out eventually (transient). Note that
this is the requirement for stability.
• If |qℓ | < 1, the forced response dies out as well (as the input is transient).

March 10, 2025

0.
Back to our example
We had y(n) = ay(n − 1) + x(n) , y(−1) = 0, where x(n) = cu(n). We also
concluded that
1
H(z) = ⇒ pole p1 = a.
1 − az−1
1
X(z) = ⇒ pole q1 = 1.
1 − z−1

• If |p1 | = |a| < 1, the natural response is transient and the system is
stable.
• As |q1 | = 1, the input is persistent, and we expect the forced response to
be persistent.
Indeed, what we got was
ca n c
y(n) = a u(n) − u(n)
a−1 a−1
| {z } | {z }
natural response forced response

March 10, 2025

0.
Summary

• z-transform – a way of representing discrete-time signals.

• Advantage: convolution in time-domain becomes multiplication in
z-domain. Easy to compute system output
• LTI system analysis: the stability of a system is determined by the
behavior of H(z) on the unit circle.
• Rational systems: if a system is defined by a difference equation, then
the z-transform of the system function is
−ℓ
∑M
ℓ=0 bℓ z
H(z) = N a z−k
∑k=0 k
• The roots of the denominator determines the stability of the system: a
causal system is stable if and only if all poles are (strictly) inside the unit
disc.
• Reading related to this lecture: Chapter 3. Read thoroughly as this is a
very important topic necessary for the rest of the course.
• Reading for next lecture: Chapter 5 – Frequency-Domain Analysis of LTI
Systems.
March 10, 2025