Content

1. Introduction
2. Characteristics of signals
3. Analog signals in frequency domain
4. Analog LTI systems
5. Sampling theorem and reconstruction
6. Discrete time signals in frequency domain
7. Discrete time LTI systems
8. Digital filters
9. Correlation
10. Advanced topics

269
7 Discrete time LTI systems
7.1 Description in time domain
§ Linearity and time-invariancy
§ Impulse response
§ Causality and stability
§ Discrete convolution
§ Difference equation
§ Block diagram
7.2 Description in frequency domain
§ Frequency response
§ z-transform
§ Transfer function
§ Poles and zero points

270
7.1 Discrete time LTI systems - Introduction
§ Discrete time system: A system, which processes a discrete time input signal <[']
to a discrete time output signal |['].
§ Discrete time LTI system: A discrete time system, which fulfills the conditions of the linearity and time-invariance
(LTI system… Linear time invariant system).

<['] LTI |[']

system

§ Examples of discrete time LTI systems

§ Digital filter
§ Digital controller
§ Linear equalizer in digital receivers
§ Digital signal generators

271
7.1 Discrete time LTI systems - Linearity and time-invariancy

§ A discrete time system will be called linear if each linear combination of input signals .5 ' , f = 1, 2, 3
leads to the respective linear combination of output signals à5 ' .
For arbitrary constants &5 , the superposition is valid:

Tr V &5 ⋅ .5 ' = V &5 ⋅ Tr .5 ' = V &5 ⋅ à5 '

5 5 5

§ A discrete time system will be called time-invariant if the form of the output signal is independent
of a temporal shift of the input signals, i.e.,
Tr .[' − f] = à[' − f]

272
7.1 Discrete time LTI systems - Impulse response
§ Impulse response ℎ['] of a LTI systems:
The output signal | ' = ℎ['] of the system, if the unit pulse <['] = Q['] is applied at the input.

discrete
time system

@[B] = D[B] ⇒ F[B] = ℎ[B]

273
7.1 Discrete time LTI systems - Impulse response
impulse response ℎ['] step response |step[']
§ The output signal of a
discrete LTI system is
calculated from the
input signal convolved
with the impulse
response of the system
arbitrary input signal <['] corresponding output signal |[']

274
7.1 Discrete time LTI systems - Causality and stability
§ Causality
§ A signal <['] is called causal, if <['] = 0 for ' < 0.
§ A discrete LTI system will be called causal, if to each causal input signal <['] belongs a causal
output signal |[']. One can show that a discrete LTI system will be exactly then causal if
ℎ['] = 0 for ' < 0.

§ Stability
§ A discrete LTI system will be called stable, if to each limited input signal <['], a limited output signal |[']
belongs. One can show that a discrete LTI system will be exactly then stable if the sum over the absolute
values of the impulse response is finite, i.e.,
+

V ℎ['] < ∞
"-*+

275
7.1 Discrete time LTI systems - Discrete convolution
§ Input-output relation of discrete LTI systems

<['] LTI |[']

system
ℎ[']
with
+

|['] = V <[f]ℎ[' − f]
5-*+

§ This relation is called discrete convolution.

§ The output signal of a discrete LTI system, thus, is given by the input signal convolved
with the impulse response of the system.
§ Consequently, the impulse response completely characterizes the LTI system.

276
7.1 Discrete time LTI systems - Discrete convolution

§ Discrete convolution:
. Y =! Y ∗ℎ Y

§ Commutative law:
.[Y] = ![Y] ∗ ℎ[Y] = ℎ[Y] ∗ ![Y]

277
7.1 Discrete time LTI systems - Discrete convolution
Derivation of the formula
§ We analyze a discrete time LTI system with the impulse response ℎ['], to which the input sequence <['] is feed in.
§ We have already seen that each discrete time signal <['] can be written as sum of delayed, weighted unit pulses,
i.e., +

<['] = V <[f]Q[' − f]
5-*+

§ According to the definition, the unit pulse Q['] at the input of the LTI system causes the impulse response ℎ[']
at it’s output:
Q['] ⇒ ℎ[']
§ Because of the time-invariance, the following holds: Q[' − f] ⇒ ℎ[' − f]
§ Because of the linearity, the following holds: <[f]Q[' − f] ⇒ <[f]ℎ[' − f]
§ Also because of the linearity, we obtain: ∑+
5-*+ <[f]Q[' − f] ⇒ ∑+ 5-*+ <[f]ℎ[' − f]

0["] h["]

278
7.1 Discrete time LTI systems - Discrete convolution
6[*] ℎ[*]
Step by step of the discrete convolution
* *
1. Substitute the discrete time variable ' to the summation variable f.
2. Flip ℎ[f] around the |-axis. This results in ℎ[−f]. ℎ[−M] ℎ[*1 − M]

M M
3. Shift ℎ[−f] with '2 sample points to the right. This results in ℎ['2 − f].
4. Multiply <[f] with ℎ['% − f]. This results in <[f]ℎ['% − f]. 6[M]ℎ[*1 − M]

5. Sum up <[f]ℎ['% − f] over all f. This results in |['% ].

M

6. Repeat steps 3 to 5 for all points ' on the discrete time axis.
9[*]
This results in the output signal |['].
9[*1 ]
*
Ø The length of the output signal is eh = e0 + eb − 1. 9[*]

* 279
7.1 Discrete time LTI systems - Difference equation
§ The most important subclass of discrete time LTI systems consists of those causal systems,
which fulfill the linear difference equation
i 7

|['] = − V &5 |[' − f] + V k5 <[' − f]

5-9 5-#

§ The actual output sample |['] is calculated from a linear combination of past output and input values
plus the actual input sample weighted with k0.
§ If all &5 = 0 f = 1 … ≈ ⟹ we have a non-recursive system, otherwise we have a recursive system.
§ Order of the system: max{≈, e}
§ Basic elements of recursive and non-recursive systems: adder, multiplier, delay elements etc.
Ø Easy to realize in hardware or software.

280
7.1 Discrete time LTI systems - Recursive LTI systems
§ Example of a 2nd order recursive system:

| ' = −&9 | ' − 1 − &% | ' − 2 + k# < ' + k9 < ' − 1 + k% < ' − 2

§ The previous values of <['] and |['] are

stored in the blocks « *9 .
§ Block diagram

§ Characteristics of recursive systems

§ The output signal is feedback.
§ As a consequence ℎ['], is (theoretically)
infinitely long.
§ Hence, recursive systems are also called
IIR systems, which is the abbreviation of
infinite impulse response.

281
7.1 Discrete time LTI systems - Non-recursive LTI systems
§ Example of a 3rd order non-recursive system (FIR system)
|['] = k# <['] + k9 <[' − 1] + k% <[' − 2] + kX <[' − 3]

§ Characteristics of non-recursive systems

§ No feedback of the output signals (all &5 = 0)
§ Non-recursive LTI systems have an impulse response ℎ['] of finite length
ℎ['] = {ℎ# , ℎ9 , . . . ℎ7 } = {k# , k9 , . . . k7 }
§ The impulse response of a non-recursive LTI system of e-th order is equal
to the coefficients of the difference equation.

282
7.1 Discrete time LTI systems - Non-recursive LTI systems
Some characteristics non-recursive systems
§ The impulse response of a non-recursive LTI system of '-th order is equal to the
coefficients of the difference equation:
You can see this very easily if you consider how the system reacts on a discrete Dirac Q[0] using the block
diagram: At ! = 0, a “1” is inducted. This “1” moves forward in the tact of the sampling frequency through the
delay line and is multiplied in each case with a k-coefficient. At the output, we obtain the signal {k0, k1, … , kç}.
The “1” disappears from the delay-line after e tacts and the impulse response stays at zero.
§ Hence, non-recursive systems can also be called FIR systems (FIR…Finite Impulse response).
§ Length of the impulse response: eb = e + 1.
Therefore, the length of the impulse response of a FIR systems of e-th order is equal to e + 1.
§ Output signal of a FIR system:
+ 7 7

| ' = ℎ ' ∗ < ' = V ℎ f < ' − f = V ℎ f < ' − f = V k5 < ' − f
5-*+ 5-# 5-#
§ This corresponds directly to the difference equation of a non-recursive LTI system.
Convolution and difference equation of a FIR system are identical.

283
7.1 Description in time domain – Block diagram
Direct form 1 (possibility 1) for recursive LTI systems

![Y] .[Y]

N E
.[Y] = − J *D .[Y − q] + J (D ![Y − q]
D1$ D1/

284
7.1 Description in time domain – Block diagram
Direct form 1 (possibility 2) for recursive LTI systems

![Y] .[Y]

N E
.[Y] = − J *D .[Y − q] + J (D ![Y − q]
D1$ D1/

285
7.1 Description in time domain – Block diagram
Direct form 2 (canonic) for recursive LTI systems

N E
.[Y] = − J *D .[Y − q] + J (D ![Y − q]
D1$ D1/

Ø The two delay lines in

center can be combined.

286
7.1 Description in time domain – Summary
§ Discrete time LTI systems can be characterized in time domain
- by the impulse response or
- by the coefficients &5 and k5 of the difference equation.

§ The input-output characteristics is given in time domain

- by the discrete convolution of the input signal with the impulse response of the system or
- by evaluation of the difference equation.

§ Often, it is more comprehensive to characterize a discrete time LTI system in frequency domain
to describe the input-output characteristics in frequency domain.

§ Possibilities to describe a discrete LTI system in frequency domain

- by the frequency response ñ(:), respectively ñ(Ω)
- by the transfer function ñj («)

287
7 Discrete time LTI systems
7.1 Description in time domain
§ Linearity and time-invariancy
§ Impulse response
§ Causality and stability
§ Discrete convolution
§ Difference equation
§ Block diagram
7.2 Description in frequency domain
§ Frequency response
§ z-transform
§ Transfer function
§ Poles and zero points

288
7.2 Description in frequency domain
§ Repetition: by the discrete time Fourier transform
+

>4 : = V < ' e*$%&"(,B

"-*+

the spectrum >4 (:) of the sampling signal <['] is calculated.

§ The frequency variable : is often substituted by the standardized variable (notation ; ≜ Ω)
; = 25:)4
We get
+

>4 (e$H ) = V <[']e*$H" .

"-*+

§ The inverse discrete time Fourier transform is given by

&
1
<['] = T >4 (e$H )e$H" d; .
25
*&

289
7.2 Description in frequency domain – frequency response
§ The discrete time Fourier transform ñ(:) of the impulse response ℎ['] of a discrete time LTI system
is called frequency response (or transfer function).
+ +

ñ : = V ℎ ' e*$%&(",B or with ;: ñ e$H = V ℎ ' e*$H" ; = 25:)4

"-*+ "-*+
§ The frequency response ñ(:) is periodic with the period :4 (sampling frequency).
§ The frequency response ñ(ejé) is periodic with the period 25.
§ Using the convolution theorem, we obtain:

<['] ℎ['] |['] = <['] ∗ ℎ[']

ñ(:)
>](:) Ç4 (:) = >](:) ñ(:)

290
7.2 Description in frequency domain –
Amplitude and phase of frequency response
§ The frequency response of a discrete time LTI system is a complex-valued function and, thus,
can be written in magnitude and phase form:
?]
ñ(e$H ) = ñ(e$H ) e$⋅Zkl O(D )
§ |ñ(ejé)| is called amplitude response or magnitude of the frequency response.
§ arg ñ(ejé) is called phase of the frequency response.

§ Amplitude and phase of the frequency response are also :4 or (25)-periodic.

§ Amplitude response is an even function: |ñ(e*$H )| = |ñ(e$H )|
§ Phase of the frequency response is an odd function: arg ñ e*$H = −arg(ñ e$H )

§ Amplitude response is often given in dB:

ñ(e$H ) mn
= 20 ⋅ log9# ñ(e$H )

291
7.2 Description in frequency domain –
Amplitude and phase of frequency response
§ LTI systems have the typical characteristic that a sine-shaped input signal is processed to
a sine-shaped output signal:

<['] = cos(;') |['] = |ñ(ejé)| cos(;' + arg ñ(ejé))

ñ(ejé)

This means that the sine-shaped output signal results from the input signal by an
amplitude change with the factor |ñ(ejé)| and a phase shift arg ñ(ejé).

§ The frequency response can easily be monitored by a sine generator.

292
7.2 Description in frequency domain –
Amplitude and phase of frequency response
§ Example: impulse response and frequency response
of a discrete LTI system

§ Note: Often only the interval 0 < : < :4 /2,

respectively 0 < ; < 5 is displayed because this
interval contains all information about the frequency
response of the system.

293
7.2 Description in frequency domain –
Phase and group delay of frequency response
§ To characterize the delay of a LTI system, two parameters are used. The more common one is the
group delay 7P (ejé)
d arg( ñ(e$H ))
7P (e$H ) = −
d;
§ The other one, less often used, is the phase delay 76 (ejé)
$H
arg( ñ(e$H ))
76 (e ) = −
;
§ With the phase delay 76 , the output of a LTI system on a sine-shaped input signal of the frequency ;
can be written in the form:
|(') = ñ(e$H ) cos( ;(' − 76 ))

§ Consequently, the phase delay is equal to the number of the sampling intervals with which a sine-signal is
delayed during the running through a discrete time LTI system. The group delay is equal to the number of the
sampling intervals with which the envelope of a modulated signal is delayed.

294
7.2 Description in frequency domain –
z-transform

§ To understand FIR filters, the knowledge of the Fourier transform is totally sufficiently.
For IIR filters, we need another mathematical tool – the z-transform.

(To use a feedback without the knowledge of the z-transform is like working with high voltage without fuse.
Who is able to do it and who is good at it, may never have any problems with it. However, if something goes
wrong, it can destroy the whole day.)

§ The z-transform is needed for the description of discrete signals and systems in the filtering theory
and in the digital controller theory.
§ There are many discrete time signals (e.g., the Heaviside step function) and systems, which do not have a
Fourier transform, because the sum of the discrete time Fourier transform does not converge.
By means of the z-transform all signals and systems, which do not have a Fourier transform, can be described.

295
7.2 Description in frequency domain –
z-transform

§ The z-transform helps us with the design and the analysis of digitals filters. The following explanations will
show how this new transform fits in with the already discussed transforms.

§ Fourier‘s great idea: Baron Jean Baptiste Joseph Fourier realized that it is possible to express signals as sums
of sine functions. He showed how periodic signals can be described by the Fourier series.

§ Fourier‘s second approach: Furthermore, Fourier realized how his idea could be applied to non-periodic
functions. This extension is called Fourier transform.

§ Fourier got famous: The Fourier transform got very popular, because it helps scientists and engineers to solve
differential equations and to avoid convolutions. For electric filters, optic filters, mechanical systems,
micro systems, and for a lot of other systems the following applies:

FT(output signal) = FT(input signal) ( FT(impulse response of the system)

296
7.2 Description in frequency domain –
Problems with the Fourier transform (repetition)

§ People soon realized that the Fourier series or the Fourier transform gave strange results in several applications:
Gibbs noticed that the Fourier series had problems with steps.
§ If we use only a finites number of harmonics, we get overshoots and ripples. The height of these overshoots does
not decrease with increasing the number of harmonics. The typical user, however, ignores this problem (called
Gibbs’ phenomenon), because it
- has little practical effect
- can be, if needed, reduced by multiplying the higher harmonics with a
adequate smoothing weighting function.

§ Many people realized that the Fourier series can be applied to periodic signals and the Fourier transform can be
applied to non-periodic signals, but neither of them can be applied to the sum of a periodic and a non-periodic
signal. Dirac solved this problem (more than 100 years later) with his Dirac impulse.
§ The Dirac impulse allows the Fourier transform of signals with a periodic part. The Dirac impulse, however, is a
mathematical abstraction, which is hard to imagine, which is impossible to create technically, and with which
nowadays everybody calculates.

297
7.2 Description in frequency domain –
Problems with the Fourier transform (repetition)

§ Dirichlet discovered that there are many problems with the Fourier transform: He showed that signals do not
have Fourier transforms if they
- show an infinite amount of local maxima and minima within a finite interval
- show an infinite amount of discontinuity points within a finite interval
- are not absolutely integrable.

§ The first two constraints hardly disturb most users, because they only exclude several crazy functions such as

1
<(!) = sin
!

0 for irrational t
< ! = E
1 for rational t

298
7.2 Description in frequency domain –
Problems with the Fourier transform (repetition)

§ The third Dirichlet condition, however, causes many more problems. It demands:
+
T <(!) d! < ∞
*+

§ The third Dirichlet condition says that the sum of the magnitude of a rectified signal has to give a finite
parameter. This constraint is almost the same as the request that the signal has to show a finite energy or an
average power of zero. By that, a lot of signals are excluded, such as DC, sine functions, steps, ramps,
exponential functions, and so on. The expansion with the Dirac impulse (like a patch) solves the problem for
some functions, but not for all.
§ The third Dirichlet condition gives hardly any problem as long as you limit yourself to passive systems.
However, it will get severe when you start to develop active filters, control systems, filter with feedback, or
systems with switches.
§ The Fourier transform and the Fourier series are still a very useful tool, but we are looking for a new and more
improved transform, which overcomes some of above-mentioned problems.

299
7.2 Description in frequency domain –
Lacking convergence of the Fourier transform

§ The integral of the Fourier transform ranges from ! = −∞ to ! = +∞:

+

> : = T < ! e*$%&() d!

*+
§ In practice, we solve this integral as a limit value:
,

>(:) = lim T <(!)e*$%&() d!

,→+
*,
§ For many functions <(!), this limit value converges. In several cases, however, the limit value grows beyond all
limits or oscillates within a range. In such cases, we say that >(:) is indefinite at these frequencies.
§ Several signals, as for example a sine signal, a DC signal, or the step function, are at one or more frequencies
indefinite. An exponentially growing signal is indefinite at all frequencies.
§ Such signals, however, are not mathematical curiosities, but appear practically. For example, the impulse
response of an instable filter is an increasing exponential function.

300
7.2 Description in frequency domain –
Lacking convergence of the Fourier transform
Example: step function
s(t)
0 for ! < 0
< ! =D ! =E 1
1 for ! ≥ 0

+ + + 0
Third Dirichlet condition: ∫*+ <(!) d! = ∫# d! = ! # = not finite t
Ø 3rd Dirichlet condition is violated.
And really:
, ,

> : = lim T < ! e*$%&() d! = lim T e*$%&() d!

,→+ ,→+
*, #
1 , j
= lim e*$%&() = lim cos( 25:)) + j sin( 25:)) − 1
−j25: ,→+ # 25: ,→+
§ Does not converge because the real part of the expression in brackets oscillates between –2 and 0, and the
imaginary part oscillates between – j and +j. The prefactor j⁄25: lowers the amplitude of the oscillation for
large :, but this does not solve the problem.
9 $
§ Using the Dirac impulse, we can write down a result: >(:) = % Q(:) − %&(

301
7.2 Description in frequency domain –
Enforcement of convergence by means of weighting function

§ The idea of Heaviside was now to multiply <(!) with a weighting function Ω(!) before doing the Fourier
transform. Or to say it differently, if FT(<(!)) does not exist, maybe at least FT(<(!) · Ω(!)) will exist.
One can reverse this weighting by calculating the IFT and multiplying it then with 1/Ω(!).

§ We have already used weighting functions by the discrete Fourier transform of finite signals.
They helped there to reduce the leakage. When designing FIR filters, we will use similar weighting
functions for a similar purpose.

§ The weighting function in the Laplace transform has a slightly different purpose. It has two tasks:
- to ensure the convergence of the Fourier integral for Ω(!) · <(!) for very many functions <(!)
- to maintain that the fundamental relation for LTI filters still applies:

LT(output signal) = LT(input signal) · LT(impulse response of the system)

302
7.2 Description in frequency domain –
Enforcement of convergence by means of weighting function

§ For most technical problems, two well suited weighting functions are:
Ç # = e&V, and Ç # = D # ⋅ e&V,

§ Contrary to the weighting functions, we have used up to now, the value of Ø is now not fixed but remains
as a variable in the transform.

:(!)

:(!) ⋅ e*o)

303
7.2 Description in frequency domain –
Enforcement of convergence by means of weighting function
D(!)
0 for ! < 0 1
Example: step function < ! = D ! = E
1 for ! ≥ 0 0 t
, ,

>(:, Ø) = lim T <(!) ⋅ Ω(!) ⋅ e*$%&() d! = ,→+

lim T e*o) ⋅ e*$%&() d! jQ
,→+
*, #
area of convergence
1 o@$%&( ) ,
= lim e*
− Ø + j25: ,→+ # É

The integral converges only for ℜ(.) = Ø > 0 (right s-semi plane) with . = Ø + j;
1
>(:, Ø) = pp < ! =
Ø + j25:
1/N
1/_ 2 N
Display of >(:, Ø) for a special Ø ⇒ −N

−N N 304
7.2 Description in frequency domain –
Laplace transform

§ Józeph Miksa Petzval (1807–1891) studied very intensively the Fourier transform
and Heaviside‘s extension:
+

> :, Ø = T < ! ⋅ e*o) ⋅ e*$%&() d!

*+
§ Petzval realized that you can combine the weighting term Ω ! = e*o) with the complex exponential function of
the Fourier transform e*$%&() . Then, you get e* o@$%&( ) . Thus, he introduced the complex variable . via
. = Ø + j25: and obtained the Laplace Transform in its mathematically most beautiful representation:
+

> . = T < ! e*!) d!

*+

§ The transform was not named Petzval-Transform, because Petzval and his student, Simon Spritzer, who both
worked intensively on the transform, fighted each other intensively in later times. Spritzer claimed that Petzval
had stolen the ideas from Laplace, who in reality never worked on the Laplace transform.

305
7.2 Description in frequency domain –
Laplace transform

The Laplace transform became very popular because

§ It is very similar to the Fourier series / Fourier transform and it can be employed for the same problems. In the
case of LTI filters, the following applies:
LT(output signal) = LT(input signal) · LT(impulse response of the systems)

§ The Laplace transform can be applied to a wider range of signals than the Fourier transform. There is no
problem to Laplace transform a step function, a ramp, or an exponential function. For some signals,
Dirac impulses are still needed and there are still signals, which do not have a Laplace transform.
However, those signals hardly show up in practice.
§ A version of the Laplace transform exists, which permits it to consider switches and initial conditions in the
system analysis.
§ The Laplace transform contains the Fourier transform as special case.

306
7.2 Description in frequency domain –
Laplace transform
§ The Laplace-transform did not completely rule out the Fourier series / Fourier transform because
- It is difficult to numerically calculate the Laplace transform of a signal.
- It is difficult to display the Laplace transform of most signals.

§ Typically, you use the most simple transform, with which you can solve your problem (Fourier sequence,
Fourier transform; or Laplace transform).

§ The Laplace transform is difficult to display. For solving this problem (especially before the era of computers),
the users limited themselves mostly to a representation of the poles, zero points, and the area of
convergence. This representation of the impulse response of a filters shows, if the filter
- Is causal, not causal, or anti-causal,
- Is stable, partially stable, or instable, and which order it has,
- Constitutes a low-pass, band-pass, high-pass, all pass, notch, or comb filter,
- Is linear-phased or non-linear-phased,
- Is a minimal phased or maximal phased filter.

307
7.2 Description in frequency domain –
Representation of Laplace transform

§ With the Fourier transform, we usually have an input function <(!) and an output function >(:). Thereby, ! and : are
real, <(!) and >(:) are complex. In other words, the Fourier transform changes ! against the considered frequency :.
§ The Laplace transform is slightly more complex, as we do not only have to determine the frequency, but also Ø.
We have thus: !, :, and Ø are real, <(!) and >(:, Ø) are complex.
§ If we calculate FT < ! , then we will obtain a function of the frequency, >(:). The frequency : thereby is a real
number between -∞ and +∞ (Hz or rad/sec). For each frequency :, we obtain a complex number, which tells us much
of a cosine and sine function of the considered frequency are contained in the analyzed function <(!). For a chart of the
Fourier transform, we often plot the magnitude and phase (or real and imaginary part) of >(:) as a function of :.
§ The Laplace transform adds another dimension to this problem. The output signal is still complex, but we have to
specify both the frequency : and the value Ø of the weighting function. For reasons, which we will see later on, we do
not simply choose a corresponding Ø, but leave it indefinite. To represent the Laplace transform, we have to draw the
magnitude / phase or real-/imaginary part over a 2-dimensional plane for Ø and :.
§ Such curve displays used to be extremely hard to generate in „pre-computer times“ and today, they are still time-
consuming and laborious.
§ It is surprising with the Laplace transform that we do not need a representation of this curve shape but a pole / zero
point-diagram is completely sufficient.

308
7.2 Description in frequency domain –
Representation of Laplace transform

§ We can imagine the Laplace transform >(.) as a sequence of Fourier transform strips. The value of >(.) along
each vertical line in the .-plane, thereby, represents the Fourier transform of < ! e*ℜ ! ) .
§ There is a special line, the imaginary axis: along this axis, the following applies ℜ . = 0, or e*ℜ ! ) = 1.
The Laplace transform along this line is the Fourier transform.

§ The weighting function Ω(!) grows with Ø beyond all limits, either for ! → ∞ or for ! → −∞. This means that
there is a limited range for Ø, for which the Fourier transform of <(!) · Ω(!) exists. This limited range for Ø
defines a vertical strip in the .-plane, which is called area of convergence.
§ This is a difference between the Fourier transform and the Laplace transform. The Fourier transform is usually
defined for all frequencies or for none at all. The Laplace transform is usually only defined in a vertical strip in
the .-plane and not for all other .-values. The exact area and the width of the domain of convergence depends
on the function <(!).

309
7.2 Description in frequency domain –
Convergence of Laplace transform

Example: negative inverted step function −D(−!)

−1 for !≤0
<(!) = −D(−!) = E
0 for !>0
0 t
, #
# -1
1 *!)
>(.) = Lpp < ! = lim T −D(−!) ⋅ e*!) d! = − lim T e*!) d! = lim e
,→+ ,→+ ,→+ . *,
*, *,
The integral converges only for ℜ . = Ø < 0 (left s-semi plane) with . = Ø + j;
1 1 jQ
Lpp < ! = 1 − 0 =
. . area of
convergence
Ø The Laplace transform of D ! and of −D(−!) are identical. The information of the É
area of convergence of the Laplace integral is necessary to decide to which function
the Laplace transform belongs.

310
7.2 Description in frequency domain –
Convergence of Laplace transform

Example: causal exponential function < ! = D ! eA) jQ

+ + 9 + area of
>(.) = Lpp < ! = ∫*+ D ! eA) e*!) d! = ∫# e A*! ) d! = e A*! )
convergence
A*! #
The integral converges only for ℜ . > ℜ & É
1
Lpp D ! eA) = ℜ . >ℜ &
.−&

Example: inverted anti-causal exponential function < ! = −D(−!) eA) jQ

+ #
#
1 area of
>(.) = Lpp < ! = − T D −! eA) e*!) d! = − T e A*! ) d! = e A*! )
&−. *+ convergence
*+ *+
The integral converges only for ℜ . < ℜ & É

1
Lpp −D(−!) eA) = ℜ . <ℜ &
.−&

311
7.2 Description in frequency domain –
Convergence of Laplace transform

§ Usually, we limit ourselves to signals and systems with fractional rational Laplace transforms >(.).
Thereby, >(.) is given by a rational function in the complex .-plane:
k" . " + k"*9 . "*9 +. . . +k9 . + k# ∏"r-# . − .#r . − .#9 . − .#% ⋅⋅⋅ . − .#"
>(.) = = ># = > #
&R . R + &R*9 . R*9 . . . +&9 . + &# ∏R 5-# . − "5 . − "9 . − "% ⋅⋅⋅ . − "R
§ >(.) is defined in the whole .-plane, apart from the zero points of the denominator polynomial.
For . = "5 ≠ .#r , >(.) has poles and goes to ±∞.

If <(!) is additionally a causal function, so jQ

jQ
zeros that < ! = 0 ∀ ! < !# , the domain of zeros
convergence is at the right of a straight line poles
poles
through the pole with the biggest real part. area of convergence
É
É
If all poles are within the left semi-plane:
<(!) → 0 for ! → ∞, which means stable.

312
7.2 Description in frequency domain –
Table of the most important Laplace correspondences

!(#) 2(8) Area of convergence

s(t)

s(t)

s(t)

s(t)
s(t)

s(t)

313
7.2 Description in frequency domain –
Inverse function of the Laplace transform

§ The inverse transform of the Laplace transform is not as simple as in the case of the Fourier transform.
The inverse Laplace transform needs an integration in the complex plane.
§ For derivation of the formula for the inverse transform, we write . = Ø + j25:
+

> . = T < ! e*o) e*$%&() d!

#*
§ The right integral again is the Fourier transform of the signal <œ ! = < ! e*o)
+
§ Using the inverse transform < ! = ∫*+ > : e$%&() d:, we obtain
+ +

<œ ! = < ! e*o) = T >(:) e$%&() d: = T >(Ø + j25:) e$%&() d:
*+ *+
o) + o@$%&( )
§ Both sides multiplied with e : < ! = ∫*+ > . e d:
m!
§ Substitution: . = Ø + j25:; d: = $%& ; limits: : = −∞ ⟶ . = Ø − j∞ ; : = ∞ ⟶ . = Ø + j∞
o@$+
1
<(!) = T > . e*!) d. Line integral in the complex plane. Calculation is possible
25j by means of function theory and residue formula.
o*$+
314
7.2 Description in frequency domain –
Inverse function of the Laplace transform

Inverse transform if >(.) has only single poles

k" . " + k"*9 . "*9 +. . . +k9 . + k# ∏"r-# . − .#r . − .#9 . − .#% ⋅⋅⋅ . − .#"
>(.) = = ># = >#
&R . R + &R*9 . R*9 . . . +&9 . + &# ∏R 5-# . − "5 . − "9 . − "% ⋅⋅⋅ . − "R
§ For real fractional rational functions, the following holds: ' < æ
If not, one should separate from >(.) a polynomial with the degree ' − æ.

§ All poles are simple, which means the denominator has æ different zero points "9 … "R .
If this holds, >(.) can be separated into partial fractions:
k" . " + k"*9 . "*9 +. . . +k9 . + k# Ö9 Ö% ÖR
> . = = + +. . . +
. − "9 . − "% ⋅⋅⋅ . − "R . − "9 . − "% . − "R

9
§ Then, inverse transform with: D(!)e6Y) ∘−• for ℜ . > ℜ "5
!*6Y

< ! = D ! Ö9 e6G) + D ! Ö% e6") +. . . +D(!)ÖR e6X)

315
7.2 Description in frequency domain –
Inverse function of the Laplace transform

Inverse transform if >(.) has multiple poles

§ Our real fractional rational function now has at least one l-fold pole at "5
ç(.) Ö9 Ö% Ö:
>(.) = = + +. . . + 
+ >(.)
—
. − "5 : e(.) . − "5 . − "5 % . − "5 :

remaining poles
)O 6 ) 9
§ Then, inverse transform with: D(!) e Y ∘−• for ℜ . > ℜ "5
:! !*6Y O`G
! :*9 6 )
< ! = D ! Ö9 e6Y) + D ! Ö% ! e6Y) +. . . +D ! Ö: e Y + <(!)
œ
l−1 !

§ If >(.) has more multiple poles, then repeat above procedure, otherwise use last slide for the rest.

316
7.2 Description in frequency domain –
Inverse function of the Laplace transform
!
Example: > . = !@9 " !@%
i.e., at "1 = −1 a double zero point, at "2 = −2 a single zero point.
Ö9 Ö% ÖX
>(.) = + +
.+1 .+1 % .+2
If we change the right side of the last equation to one common fraction, we obtain
Ö9 . + 1 (. + 2) Ö% . + 2 ÖX . + 1 %
> . = + +
.+1 % .+2 .+1 % .+2 .+2 .+1 %
Ö9 . % + 3. + 2 + Ö% . + 2 + ÖX (. % + 2. + 1)
> . =
.+1 % .+2
Ö9 + ÖX = 0; 3Ö9 + Ö% + 2ÖX = 1; 2Ö9 + 2Ö% + ÖX = 0 ⇒ Ö9 = 2; Ö% = −1; ÖX = −2
2 1 2
>(.) = − −
.+1 .+1 % .+2

< ! = 2D ! e*) − D ! ! e*) − 2D(!)e*%)

317
7.2 Description in frequency domain –
z-transform

Why z-transform became very popular

§ The z-transform has the same advantages and disadvantages as the Laplace-transform and is used in a very
similar way. Through the variable substitution of . = Ø + j25: to « = e o@$%&( , , we obtain the z-transform
+

>j « = V < ' « *"

"-*+
§ The pole/zero point-diagram and the domain of convergence in the z-plane looks different when compared to
the s-plane. Despite this difference, however, the key to understand the z-transform is to understand the
Laplace transform, and then, in the second step, the variable substitution of s-to-z.

Points in the s- and z-plane

§ The s- as well as the z-values are complex numbers.
§ Usually, people interpret the imaginary axes of the s-plane as frequency and
the real part as associated exponential damping factor.
§ In the z-plane, however, the frequency is given by the phase and the exponential damping factor
by the magnitude of a phasor.

318
7.2 Description in frequency domain –
z-transform
§ The z-transform of a discrete time signals <['] is given by+
>j « = V < ' « *"
"-*+
with the complex variable «.
§ Similar to the other transforms we discussed already, <['] and >j « form a transform couple.
<['] ∘−• >j («)
Im(ë)
!(#)
z-transform
2W (ë)

# Re(ë)
§ The z-transform can be seen as generalization of the discrete time Fourier transform.
§ For discrete time signals and systems, it plays the same role as the Laplace transform for analog signals and systems.

319
7.2 Description in frequency domain –
z-transform j;

Points in the s- and z-plane

t a
> . = T e* o@$H )
< ! d!
A
+ +

>j («) = V <[']« *" = V <['] ⋅ e*o," ⋅ e*$H," Continuous FT

"-*+ "-*+

§ The Laplace transform is composed of vertical Fourier transform strips, whereas

ℑ(ë)
each strip uses a different weighting function e*ℜ ! ) . Hereby, we have one special
strip at ℜ . = 0, because this one represents the common Fourier transform. Unit circle
§ The z-transform is composed of Fourier transform circles around the origin. Each ℜ(«)
0
circle in the z-plane is the DFT of <(!) · Ω(!), whereas the radii correspond to the ws
different damping factors ( « = e*o,u" ). Hereby, we have one special circle with
the radius 1, which corresponds to the DFT of <(!). This circle should always be
drawn into the pole/zero point diagram.
Non-causal Discrete FT
signals 320
7.2 Description in frequency domain –
z-transform

Connection between the z- and the discrete time Fourier transform

25 3 = 2W ë = e'!(*+m

25 e'X = 2W ë = e'X

§ This formulae of course will only apply, if the discrete time Fourier transform of the signals exists.

§ The Fourier transform (i.e., the spectrum) of a discrete time signal, thus, is equal to the z-transform of
the signal, evaluated at the point
« = e$%&(,B
« = e$H

321
7.2 Description in frequency domain –
z-transform

Points in the s- and z-plane: Complement images of the s- and z-plane

s-plane z-plane
Im ( = Q Im Ü
Unit circle

Re ( = É 0
Re Ü
ws
H = e”‘

§ The j;-axis in the s-plane is transformed to the unit circle in the z-plane.
§ The right semi-plane of the s-plane is transformed to the region outside the unit circle.
§ The left semi-plane of the s-plane is transformed to the region inside the unit circle.

322
7.2 Description in frequency domain –
z-transform

Points in the s- and z-plane

t + +
ä(.) = T e* o@$H )
:(!)d! >j («) = V <[']« *" = V <['] ⋅ e*o," ⋅ e*rH,"
A "-*+ "-*+
s-plane z-plane
Im(Ü)
Unit circle =
á jQ − axis

a
H = e”‘
Re(Ü)

The range for . or «, for which the Fourier transform exists, is called the domain of convergence. In the s-plane, the
domain of convergence is always a vertical strip, corresponding to a ring-shaped region in the z-plane.

323
7.2 Description in frequency domain –
z-transform

Example 1: causal exponential function ![Y] = D[Y] ⋅ *4

+

> « = V < ' « *"

"-*+
+ +

= V &" « *" = V (&« *9 )"

"-# "-#
9 j
= = for |«| > |&|
9*Aj AG j*A

Region of
convergence Radius ä = V
Geometric sequence
9
0 ∑+ :
:-# < = for < < 1
9*0
Zero point
Pole at « = &
at « = 0

324
7.2 Description in frequency domain –
z-transform

Geometric sequence
1 + # + # ! + # Y +. . . = &
+
1
# + # ! + # Y +. . . = & − 1
V <: = for < < 1 # ⋅ 1 + # + # ! +. . . = & − 1
1−<
:-#

<⋅ú =ú−1
ú⋅ 1−< =1
1
ú=
1−<

325
7.2 Description in frequency domain –
z-transform

Example 2: negative anti-causal exponential function: ![Y] = −D[−Y − 1] ⋅ *4

Domain of
convergence
0

+ *9 *9 Pole at « = &
*" " *" *9 "
>(«) = V <[']« = V −& « = − V (&« )
Zero point
"-*+ "-*+ "-*+
at « = 0
Substitution: ± = −'; ' = −∞ ⇒ ± = ∞; ' = −1 ⇒ ± = 1
*9 + +
1 &−« &
− V (&« *9 )" = − V(«&*9 )W = 1 − V(«&*9 )W = 1 − = −
1 − «&*9 & − « & − «
"-*+ W-9 W-#
j
Result: > « = for « < |&|
j*A

326
7.2 Description in frequency domain –
z-transform: Domain of convergence
§ Examples 1 & 2 give the same z-transforms, although both functions of time are different.
Very different functions of time can have the same z-transforms – they differ from each other in the domain
of convergence (ROC):
- In example 1, the sum converges: >(«) = ∑+ " *" only for « < &
"-# & «

- In example 2, the sum converges: >(«) = ∑*9 " *"

"-*+ & « only for « > &
§ Hence, the entire region of the z-transform belongs to the domain of convergence.
§ There are 4 types of domains of convergence, corresponding to the 4 types of functions of time:
- Sinistral
- Dextral
- Two-sided (unlimited)
- Two-sided limited

327
7.2 Description in frequency domain –
z-transform: Domain of convergence
§ The domain of convergence is a Domain of
« > & particular value (range outside the circle) dextral function of time: convergence
if the time signal is zero for −∞ < ' < '0 Im(Ü)
and shows only values for '0 + 1 < ' < + ∞
(All causal functions of time are dextral).
time, 'ΔT sec
Re(Ü)

§ The domain of convergence is a

« < & particular value (range outside the circle)
if the time signal shows only values for sinistral function of time: Im(Ü)
−∞ < ' < '0 and is zero for '0 + 1 < ' < + ∞
(sometimes « = 0 is not included).
time, 'ΔT sec Re(Ü)

328
7.2 Description in frequency domain –
z-transform: Domain of convergence
Two-sided function of
§ Two-sided function of time: Look at the function time
< ' = D ' ⋅ &" − D −' − 1 ⋅ k "
time, 'ΔT sec
Using of the results of example 1 and 2 we obtain:
|«| > |&| and « < k
Domain of
The domain of convergence is the intersection of both ranges, i.e., convergence
|k| > |«| > |&| Im(Ü)

§ Finite series of the form: Re(Ü)

< ' ≠ 0, '9 < ' < '%

The domain of convergence comprises the whole z-plane Im(Ü)

(sometimes with the exception « = 0)

Re(Ü)

329
7.2 Description in frequency domain –
z-transform: Causality, stability and domain of convergence
§ One can show that LTI systems will be completely stable, if the domain of
convergence comprises the unit circle. Im(«)
§ This condition is identical with the demand that the DFT exists.
§ A causal system is a dextral function of time. In this case, the domain of
convergence is outside of the circle, around the origin.
1
§ A causal, stable LTI system has, thus, a domain of convergence outside of a circle,
but inside the unit circle. Re(«)

Ø Condition for a causal stable LTI system:

domain of convergence |«| > |&| with |&| < 1

330
7.2 Description in frequency domain –
z-transform

Example 3: ![Y] = D[Y] ⋅ Y ⋅ *4

+ + + +
d " *"
> « = V < ' « *" = V ' &" « *" = & V ' &"*9 « *" = & V & «
d&
"-*+ "-# "-# "-#

+
d d « « − & ⋅ 0 − « ⋅ −1 &«
=& V &" « *" = & =& =
d& d& « − & «−& % «−& %
"-#
Domain of
for |«| > |&| convergence

MATLAB: look at the commands ztrans and iztrans. 0

Double pole
Zero point at « = &
at « = 0
331
7.2 Description in frequency domain –
z-transform
§ When using:
z-transform «
<['] = D['] ⋅ &" >(«) =
«−&
§ & = 1: step function z-transform «
<['] = D['] >(«) =
«−1
§ & = e±t, : exponential function
z-transform «
±t," >(«) =
<['] = D['] ⋅ e
« − e±t,

§ When using: z-transform &«

>(«) =
<['] = D['] ⋅ ' ⋅ &" «−& %

§ & = 1: ramp function z-transform )⋅«

<['] = D['] ⋅ ) ⋅ ' >(«) = %
«−1

332
 Description within the frequency domain - z-transform
7.2
Function P(#) r(() <(Ü)
Dirac pulse E(#) 1 1
Shifted Dirac E(# − UΔ&) e1Äè%" Ü 1Ä
Step C(#) 1 Ü
( Ü−1
(for the z-transform, Ramp # 1 Δ&Ü
the following applies: (( (Ü − 1)(
:(!) → :(' Δ)) = <[']) #( 2 Δ& (Ü(Ü + 1)
(w (Ü − 1)w
Exponential e1^? 1 Ü
(+V Ü − e1^è%
#e1^? 1 Δ&Üe1^è%
(( + V)( (Ü − e1^è% )(
Sine sin( Q#) 1 Ü ( in QΔ&
Q( + (( Ü ( − 2Ü cos QΔ& + 1
Cosine cos( Q#) 1 Ü ( − Ü cos Q#Δ&
Q( + (( Ü ( − 2Ü cos( Q#Δ&) + 1

333
7.2 Description in frequency domain –
z-transform: Characteristics
§ Linearity P"I [8] + S"H [8] ∘−• P.I,P (T) + S.H,P (T)

§ Convolution "I [8] ∗ "H [8] ∘−• .I,P (T) ⋅ .H,P (T)

§ Time reversal "[8] ∘−• .P (T) "[−8] ∘−• .P (T 3I )

P
§ Scaling V> "[8] ∘−• .P ( )
=
Q
§ Differentiation in the z-domain 8 ⋅ "[8] ∘−• −T . (T)
QP P

§ Conjugated complex sequence " ∗ [8] ∘−• .P∗ (T ∗ )

§ Limit value lim (" 8 ) = lim 1 − T 3I .P T

>→4 P→I

§ For causal systems, the following applies "[0] = lim .P (T)

P→4

334
7.2 Description in frequency domain –
z-transform: Characteristics

§ Time delay: < ' − f ∘−• « *5 >j «

§ Proof for 1 unit delay:
+ +

<[' − 1] ∘−• V <[' − 1]« *" = « *9 ⋅ V <[' − 1]« * "*9

= « *9 >j («)
"-*+ "-*+

<['] Delay <[' − 1] <['] <[']ñ(«) = <[' − 1]

= « *9
1 unit

<['] <[' − 1]
« *9

§ Relation between differential equation and difference equation:

◊ ◊%
| ! = k# < ! + k9 < ! + k% % < !
◊! ◊!
§ Corresponding difference equation:
|['] = k# <['] + k9 <[' − 1] + k% <[' − 2]
335
7.2 Description in frequency domain –
z-transform: Inverse z-transform

J ![Y] ⋅ ë &4

![Y] 2W (ë)

336
7.2 Description in frequency domain –
Inverse z-transform
§ The inverse z-transform is calculated using:
1
!Y = ñ ë 4&$ 2 ë dë = J Residuals
2=j
-

§ The z-transform >(«) converges in the range ã9 < |«| < ã% . The contour å runs in a mathematically positive
direction within the domain of convergence.
§ If the unit circle of the z-plane is within the domain of convergence of the z-transform, the contour å can lie
along the unit circle. In this case, the inverse integral turns into the inverse transform of the discrete time
Fourier transform.
§ In addition to the direct calculation of the inverse z-transform via integration in the complex plane,
there are also 2 other calculation methods.
- Polynomial division and arranging in powers of «
- Partial fraction decomposition
Ø Usually, partial fraction decomposition is used.

337
7.2 Description in frequency domain –
Inverse z-transform: Partial fraction development
§ As with the Laplace transform, the inverse z-transform can be determined using
partial fraction development of the rational z-transform.
§ In the inverse Laplace transform, we are looking for terms of the form:
Ö9 Ö% ÖX
ä . = + + +. . . .•−∘ :(!) = Ö9 e6G) + Ö% e6") + ÖX e6a) . . . .
. − "9 . − "% . − "X

§ Such a beautiful formula does not exist for the inverse z-transform. Instead, we are looking for terms
of the following form:

VD ë
⇔ VD óD4
ë − óD

338
7.2 Description in frequency domain –
Inverse z-transform: Partial fraction development
Assumptions: (i) order of numerator of the z-transform >(«) must be smaller than the order of the denominator;
(ii) >(«) contains only simple poles "5 , (iii) and <['] must be a dextral function.
§ Partial fraction development (æ number of poles):
R
Ö5
> « =V , where Ö5 = « − "5 ⋅ > « ÿ
« − "5 j-6Y
5-9
§ For each partial fraction, a series can be determined:
Ö5 *9
«
= Ö5 ⋅ « ⋅ •−∘ Ö5 ⋅ "5 "*9 ⋅ D[' − 1]
« − "5 « − "5
§ The inverse transformed series is therefore:
R R
Ö5
>(«) = V •−∘ <['] = V Ö5 ⋅ "5 "*9 ⋅ D[' − 1]
« − "5
5-9 5-9
§ In the case of z-transforms with multiple poles, as with the Laplace-transform, partial fraction decomposition
has to be performed, with multiple poles. The correspondence for this inverse transform is:
«: '+l−1 !
:
•−∘ ⋅ "5 " ⋅ D[']
« − "5 '!⋅ l−1 !

339
7.2 Description in frequency domain –
Inverse z-transform: Partial fraction development
§ Example of a partial fraction expansion for a double pole:
C
:(B) =
1 − D/ B 1/ 1 − D( B 1/ (
§ It is easier for the partial fraction development if the fraction is multiplied by B3:
C Bw
:(B) =
B − D/ ⋅ B − D( (
§ In order to make the order of the numerator polynomial smaller than the order of the denominator polynomial,
we divide by B: (
: B C B F/ F( Fw
:ê B = = = + +
B B − D/ B − D( ( B − D/ B − D( B − D( (

F/ = B − D/ : ê (B)|ë.$6 Fw = B − D( (
: ê (B)|ë.$7
§ To determine the „difficult“ coefficient F( , the above equation is multiplied by (B − D( )( :
C B( F/ B − D( (
= + F( B − D( + Fw
B − D/ B − D/

340
7.2 Description in frequency domain –
Inverse z-transform: Partial fraction development
l «% Ö9 « − "% %
= + Ö% « − "% + ÖX
« − "9 « − "9
§ Then, we differentiate the above equation with respect to « and swap the sides:
%
Ö9 « − "% − 2Ö9 « − "% « − "9 d %
+ Ö% = « − "% ⋅ > J («)
« − "9 % d«
§ Now, we set « = "% . In doing so, the left side then simplifies to Ö% , and the right side becomes well defined.
d %
Ö% = « − "% > J («) Ÿ
d« j-6"

341
7.2 Description in frequency domain –
Inverse z-transform: Partial fraction development
§ If the z-transform was performed using a real sequence <['], the corresponding complex pole « = "∗ always
occurs as well at a complex pole « = ". This also applies for zero points.
§ For conjugated complex poles, the following rule applies:
Ö Ö∗ Ö − Ö"∗ « *9 + Ö∗ − Ö∗ "« *9
>9 « = + =
1 − "« *9 1 − "∗ « *9 1 − "« *9 − "∗ « *9 + ""∗ « *%
§ The following relations also apply:
§ ""∗ = " % and " + "∗ = 2 Re "
§ Ö + Ö∗ = 2 Re Ö ,
§ Ö"∗ − "Ö∗ = −2 Re Ö"∗
§ All terms are real. Therefore, >(«) can also be set as
k9 « *9
>9 « = k# +
1 + &9 « *9 + &% « *%
with the real coefficients k# , k9 , &9 , and &% .

342
7.2 Description in frequency domain –
Inverse z-transform: Partial fraction development
Xj*%
Example 1: >(«) =
j "*9,Kj@#,KV

§ Poles at "1 = 0.6 and "2 = 0,8

§ Partial fraction decomposition: Ö5 = « − "5 >(«)|j-6Y

3« − 2 3« − 2 1.8 − 2
Ö9 = « − 0.6 Ÿ = Ÿ = =1
« − 0.6 « − 0.8 j-#.x « − 0.8 j-#.x 0.6 − 0.8
3« − 2 3« − 2 2.4 − 2
Ö% = (« − 0.8) Ÿ = Ÿ = =2
« − 0.6 « − 0.8 j-#.V « − 0.6 j-#.V 0.8 − 0.6
1 2
>(«) = +
« − 0.6 « − 0.8

§ Inverse transformed sequence: <['] = 0.6"*9 + 2 ⋅ 0.8"*9 ⋅ D[' − 1]

343
7.2 Description in frequency domain –
Inverse z-transform: Partial fraction development
($&Z}íìv )W
Example 2: 2(ë) = poles at "1 = 1 and "2 = e*A∆,
(W&$)(W&Z}íìv )
§ Partial fraction decomposition is not performed on >(«), but on >(«)/«:
J
> « 1 − e*Az,
> « = = *Az,
, Ö5 = « − "5 ⋅ > J («)ÿ
« «−1 «−e j-6Y
(1 − e*Az, ) (1 − e*Az, )
Ö9 = (« − 1) ⋅ ⁄ = ⁄ =1
(« − 1)(« − e*Az, ) (« − e*Az, )
j-9 j-9
*Az, *Az,
*Az,
(1 − e ) (1 − e ) 1 − e*Az,
Ö% = (« − e )⋅ ⁄ = ⁄ = *Az,
(« − 1)(« − e*Az, ) (« − 1) e −1
j-DAWbM j-DAWbM

> « 1 1 « «
>J « = = − > « = −
« « − 1 « − e*Az, « − 1 « − e*Az,

j j
§ Transform couples: •−∘ D['] and •−∘ e*Az,"
j*9 j*DAWbM
§ Inverse transformed sequence: <['] = D ' − e*Az,⋅" ⋅ D[']
344
7.2 Description in frequency domain –
Inverse z-transform: Partial fraction development
9
Example 3: > « = ROC: « > 1
9@j AG 9*j AG "

ja
§ Adapted function: > « =
9@j 9*j "

I(j) {G {" {a
§ Division by «: > J («) = = + +
j 9@j 9*j 9*j "

§ Partial fraction development and inverse z-transform leads to result.

345
7.2 Description in frequency domain –
Transfer function
§ The z-transform ñj («) of the impulse response ℎ['] of a discrete time LTI system is called a transfer function.
+

ñj « = V ℎ ' « *" ℎ['] ∘−• ñj («)

"-*+

§ As a result of the convolution theorem, the following applies:

<['] |['] = <['] ∗ ℎ['] 4W (ë)

ℎ['] dW (ë) =
2W (ë)
ñj («)
>j («) Çj « = >j («)ñj («)

§ Both these equations describe the fundamental connection between the input and output signal
of a discrete LTI system.
§ ℎ['] and ñj («) form a transform couple. A discrete LTI system can be described either by its
impulse response ℎ['] or by its transfer function ñj («).

346
7.2 Description in frequency domain –
Transfer function / frequency response
§ For causal and stable systems, the following connection between transfer function and frequency response
applies:
d(e'X ) = dW (ë = e'X )
§ The frequency response of a discrete time system is, thus, equal to the transfer function, evaluated on the unit
circle of the complex z-plane. ñ(e$H ) is 25-periodic.
§ If one knows the sampling time )4 and wants to display the frequency response as a function of frequency :,
then one has to substitute ; with 25:)4 :

d(3) = dW (ë = e'!(*+m )

§ The frequency response ñ(:) of a discrete time LTI system is :4 -periodic.

§ Reminder: often only the range 0 < : < :4 /2 respectively 0 < ; < 5 is displayed, since all information
about the frequency response of the system is contained in this range.

347
7.2 Description in frequency domain –
Transfer function of IIR systems
§ Difference equation
i 7

|['] = − V &5 |[' − f] + V k5 <[' − f]

5-9 5-#

§ By means of the linearity and the shifting characteristics of the z-transform, it follows:
i 7

Çj « = − V &5 « *5 Çj « + V k5 « *5 >j «
5-9 5-#

and the transfer function is thus:

Çj («) k# + k9 « *9 +. . . +k7 « *7
ñj («) = =
>j («) 1 + &9 « *9 +. . . +&i « *i

348
7.2 Description in frequency domain –
Transfer function of FIR systems
§ Difference equation
7

|['] = V k5 <[' − f]
5-#

§ Impulse response
ℎ['] = {ℎ# , ℎ9 , . . . ℎ7 } = {k# , k9 , . . . k7 }

§ Transfer function
ñj («) = k# + k9 « *9 +. . . +k7 « *7

349
7.2 Description in frequency domain –
Transfer function of FIR systems
§ Example 1: difference equation of a causal FIR system ![Y] .[Y]
delay
< ' = | ' = 0 for ' = −1, −2, −3 , ... 1 unit
|['] = <[' − 1]
new old

§ Impulse response, i.e., excitation with a pulse <[0] = 1 and < ' = 0 for ' > 0

| 0 = 0; | 1 = 1; | 2 = 0; | 3 = 0; … .

§ Step response, i.e., excitation with a step < ' = 1 for ' ≥ 0

| 0 = 0; | 1 = 1; | 2 = 1; | 3 = 1; … .

350
7.2 Description in frequency domain –
Transfer function of FIR systems
9 9 9
§ Example 2: difference equation of a causal FIR system |['] = X <['] + X <[' − 1] + X <[' − 2]

<(') < ' = | ' = 0 for ' = −1, −2, −3 , ...

T T
1/3 1/3 moving mean value
1/3
+ + |(')

1 1 1
|[0] = <[0] + <[−1] + <[−2]
Impulse response, i.e., 3 3 3
excitation with a pulse 1 1 1 1
| 0 = ⋅1+ ⋅0+ ⋅0= n=0
3 3 3 3
1 1 1 1
< 0 =1 | 1 = ⋅0+ ⋅1+ ⋅0= n=1
<['] = 0 for n > 0 3 3 3 3
1 1 1 1
| 2 = ⋅0+ ⋅0+ ⋅1= n=2
3 3 3 3
1 1 1
|[3] = ⋅ 0 + ⋅ 0 + ⋅ 0 = 0 n=3
3 3 3
351
7.2 Description in frequency domain –
Transfer function of FIR systems
9 9 9
§ Example 2: difference equation of a causal FIR system |['] = X <['] + X <[' − 1] + X <[' − 2]

<(') <['] = |['] = 0 for n = −1, −2, −3 , ...

T T
1/3 1/3 moving mean value
1/3
+ + |(')

1 1 1
|[0] = <[0] + <[−1] + <[−2]
Step response, i.e., 3 3 3
excitation with a step 1 1 1 1
| 0 = ⋅1+ ⋅0+ ⋅0= n=0
3 3 3 3
1 1 1 2
<['] = 1 for n ≥ 0 | 1 = ⋅1+ ⋅1+ ⋅0= n=1
3 3 3 3
1 1 1
| 2 = ⋅1+ ⋅1+ ⋅1=1 n=2
3 3 3
1 1 1
|[3] = ⋅ 1 + ⋅ 1 + ⋅ 1 = 1 n=3
3 3 3
352
7.2 Description in frequency domain –
Transfer function of IIR systems
§ Example 1: difference equation of a causal IIR system |['] = <[' − 1] + Ø|[' − 1]
<['] = |['] = 0 for n = −1, −2, −3 , ...
<['] + T |[']

|[0] = <[−1] + Ø|[−1]

Impulse response, i.e.,
|[0] = 0 initial value
excitation with a pulse
| 1 =1+Ø⋅0=1 n=1
| 2 =0+Ø⋅1=Ø n=2
< 0 =1
| 3 = 0 + Ø ⋅ Ø = Ø% n=3
<['] = 0 for n > 0
|[4] = 0 + Ø ⋅ Ø % = Ø X n=4

|['] = Ø "*9 ⋅ D[n − 1]

353
7.2 Description in frequency domain –
Transfer function of IIR systems
§ Example 1: difference equation of a causal IIR system |['] = <[' − 1] + Ø|[' − 1]
<['] = |['] = 0 for n = −1, −2, −3 , ...
<['] + T |[']

§ Impulse response, i.e., excitation with a pulse < 0 = 1 and <['] = 0 for n > 0
Input Output
Ø%
0<É<1 0 < É < −1 É>1
Ø# 9 Ø# Ø9
1 1 Ø 1 1
%
Ø Ø% Ø#

X[0]X[1]X[2]X[3] y[0]y[1]y[2] y[3]

Ø9 |['] = Ø "*9 ⋅ D[n − 1]
354
7.2 Description in frequency domain –
Transfer function of IIR systems
$&W }~
§ Example 2: given the following transfer function d(ë) =
$#/,[W }~
§ One can determine the impulse response using the corresponding power sequence and
then using the difference equation.
§ Power series expansion:
1 − « *9 *9 *% *X
ñ(«) = = 1 − 1.5« + 0.75« − 0.325« +⋯
1 + 0.5« *9
§ The values of the impulse response become:
ℎ[0] = 1; ℎ[1] = −1.5; ℎ[2] = 0.75; ℎ[3] = −0.325
§ Difference equation: transform the transfer function into a difference equation and solve for |(').
1 − « *9 Ç «
ñ « = *9 = Ç « + 0.5« *9 Ç « = > « − « *9 > «
1 + 0.5« > «
| AG
| ' + 0.5| ' − 1 = < ' − < ' − 1
|['] = <['] − <[' − 1] − 0.5|[' − 1]

355
7.2 Description in frequency domain –
Transfer function of IIR systems
§ Example 2: | ' = < ' − < ' − 1 − 0.5| ' − 1
§ We can calculate the impulse response by using:

<['] = Q['] and the initial condition: < 0 = 1, < ' = 0 for ' ≠ 0

| 0 =Q 0 −Q −1 − 0.5| −1 = 1 − 0 − 0 = 1
| 1 =Q 1 −Q 0 − 0.5| 0 = 0 − 1 − 0.5 = −1.5
| 2 =Q 2 −Q 1 − 0.5| 1 = 0 − 0 − 0.5 −1.5 = 0.75
| 3 =Q 3 −Q 2 − 0.5| 2 = −0.5 0.75 = −0.325
⋮

§ The impulse response then becomes:

ℎ[0] = 1; ℎ[1] = −1.5; ℎ[2] = 0.75; ℎ[3] = −0.325

356
7.2 Description in frequency domain –
Transfer function of IIR systems
§ Calculation of the frequency response of a causal IIR system
| ' + 1 − 0.8| ' = < ' + 1 with < ' = | ' = 0 for n = −1, −2, −3 , …

z-transform: Ç « « − 0.8 = > « «

<['] + T |[']
Ç(«) « 1
ñ(«) = = =
0.8 >(«) « − 0.8 1 − 0.8« *9

§ Frequency response: « = e$H

$H
1 1 1
ñ e = = =
1 − 0.8e*$H 1 − 0.8 cos ; − j sin ; 1 − 0.8 cos ; + j0.8 sin ;
1 1
ñ(e$H ) = =
1 − 0.8 cos ; % + 0.8 sin ; % 1.64 − 1.6 cos ;
& + jk = &% + k %
k
#.V EFG H ∠& + jk = tan*9
§ Phase of frequency response: ∠ñ e$H = 0 − tan*9 &
9*#.V }~E H

Note the 0 of the numerator and the 2 term of the denominator stem from ñ e$H .
357
7.2 Description in frequency domain –
Transfer function – zeros and poles
§ Poles and zeros of a transfer function contain useful information about a system.
§ Transfer function of a IIR system can be transformed as follows

k# + k9 « *9 +. . . +k7 « *7
ñj « =
1 + &9 « *9 +. . . +&i « *i

« 7 + (k9 ⁄k# )« 7*9 +. . . +(k7 ⁄k# )

= k# « i*7
« i + &9 « i*9 +. . . +&i
§ If one converts the numerator and denominator polynomial into their roots, one can write:

(« − «9 )(« − «% ). . . (« − «7 ) «5 …zeros
ñj («) = k# « i*7
(« − "9 )(« − "% ). . . (« − "i ) "5 …poles
§ This form is called the pole-zero representation of the transfer function.

358
7.2 Description in frequency domain –
Transfer function – zeros and poles
« − «9 « − «% … « − «7 «5 …zeros
ñj « = k# « i*7 "5 …poles
« − "9 « − "% … « − "i

§ The transfer function, apart from the constant k0, is completely defined by the poles and zeros.
§ The number of the zeros equals the order of the numerator polynomial.
§ The number of poles equals the order of the denominator polynomial.
§ In addition to the ≈ poles, there will be e − ≈ poles within the origin if e > ≈,
and additional ≈ − e zero points within the origin if ≈ > e.
§ Since these poles and zeros are rarely interesting, they are not represented in a pole-zero point diagram.
§ In a pole-zero diagram, the poles are represented by small crosses and the zeros by small circles
in the complex z-plane.

359
7.2 Description in frequency domain –
Transfer function – zeros and poles
Characteristics of poles and zeros for a recursive LTI system
§ Systems with real &- and k-coefficients have real and/or conjugated complex poles and zero points.
§ A recursive discrete time LTI system will be completely stable, if all poles are inside the unit circle.
§ Poles and zeros in the proximity of the unit circle cause the amplitude response to increase / decrease.
§ The conjugated complex pole couple, which is closest to the unit circle, is called dominant. It causes oscillations
during switch-on procedures, which last longer the closer the pole couple is to the unit circle.

360
7.2 Description in frequency domain –
Transfer function – zeros and poles
Possible pole-zero point distributions of a LTI system Im(«) z-plane
§ Real zero point
ñ(«) = k − « Im(«)
o Re(«)
o
§ 2 conjugated complex zeros
ñ(«) = (k − «)(k ∗ − «)
Re(«)
Im(«)
§ Real pole o
ñ(«) = (& − «)*9
Im(«)
x Re(«)
§ 2 conjugated complex poles
ñ(«) = (& − «)*9 (&∗ − «)*9 x
Re(«)
All filters can be represented as series connection of x
these basic filters.

361
 W L-
é= =

Description in frequency domain –

2 4

7.2 Im(«) ç = ejé

Transfer function – zeros and poles Unit circle 1

§ The frequency response of a filter results from the ω

z-transform, evaluated on the unit circle. é=W =
L-
Re(«)
2 -1 1
§ The pole-zero representation of a recursive LTI system is é = 0 (= DC)
(« − «9 )(« − «% ). . . (« − «7 ) «5 …zeros
ñj («) = k# « i*7 "5 …poles z-plane
(« − "9 )(« − "% ). . . (« − "i )
§ Each difference term in the pole-zero representation of
ñ(«) can be represented as a complex vector. é=
3W 3 -1
2 4L-

Im(«)
§ Example:
«9 = e$H̑ G
ñ(«) = k − «
ñ(«) = fi «% = e$H̑ "
∠ñ(«) = 9 k
fi9
o Re(«)
fi%
k − «%
k − «9
362
7.2 Description in frequency domain –
Transfer function – zeros and poles
Ü W ä
§ Extrema: the maximum of the frequency response
(=minimal damping) will occur, if k and « are Ñ = 0; fi =k+1
anti-parallel.
§ Extrema: the minimum of the frequency response ä W Ü ñ(«) = k − «
(=maximal damping) occurs, if k and « are parallel.
Ñ = 5; fi =k−1

§ Mathematics: ñ « = k − « = k − e$H̑ = (k − cos ;)

̑ − j sin ;̑

ñ(«) = ̑ % + (sin ;)
(k − cos ;) ̑ %
∠ñ(«) = arctan( − sin ;/(k
̑ − cos ;))
̑
Product: « = «9 «% = «9 «% e$(oG@o")
j jG
Quotient: « = jG = j"
e$(oG*o")
"

363
7.2 Description in frequency domain – z-plane
é=
W
=
L-

Transfer function – zeros and poles Im(Ü)

2 4
« = ejé
Unit circle
Pole-zero representation of a recursive LTI system å1
(« − «9 )(« − «% ). . . (« − «7 ) ç1 W
ñj («) = k# « i*7 L- ä1
(« − "9 )(« − "% ). . . (« − "i ) é=W = Re(Ü)
2 23ω0 Ü1
§ Each difference term in the pole-zero representation of é = 0 (= DC)
ñ(«) can be represented as a complex vector. ç2

§ The frequency response of a filter results from the å2

z-transform, evaluated on the unit circle « = ejé.
§ The amount of these vectors result from the distance of the poles
and zeros from the frequency analysis point on the unit circle.
|h(ej5)| |Ü − Ü/| ä/
§ The resultant angle is between the vector and the horizontal axis. =
|Ü − å/||Ü − å(| ç/ ç(
$H
fi9 e$CG fi% e$C" … fi" e$CQ
ñ e = k#
‡9 e$ÄG ‡% e$Ä" … ‡R e$ÄX
fi9 fi% … fi" $[ C @C @ …@C * Ä @Ä @ …@Ä ]
= k# e G " Q G " X
‡9 ‡% … ‡R
Q = 23P ⋅ Δ&
ω0
364
7.2 Description in frequency domain –
Transfer function – zeros and poles
Pole-Zero representation of a recursive LTI system
(« − «9 )(« − «% ). . . (« − «7 ) z-plane
ñj («) = k# « i*7 W L-
(« − "9 )(« − "% ). . . (« − "i ) é=
2
=
4
« = ejé
§ The amount of these vectors result from the distance of the pole-zero Im(Ü)
point from the frequency analysis point on the unit circle. Unit circle
å1
§ The resultant angle is between the vector and the horizontal axis. é/
ç1 W
L- ä1
é=W =
2 23ω0 è/ Re(Ü)
Ü1 é = 0 (= DC)
fi9 fi% … fi" $[ ç2
ñ e$H = k# e CG@C"@ …@CQ * ÄG@Ä"@ …@ÄX ]
‡9 ‡% … ‡R é(
å2
$H
arg ñ e = Ñ9 + Ñ% + … + ÑR − (·9 + ·% + … + ·R )

365
7.2 Description in frequency domain –
Transfer function – zeros and poles
z-plane
Band stop filter: a zero point on the unit circle leads to a zero point Im(«)
in the frequency response. The corresponding frequency is
completely suppressed.

k−« =0 ë4
ñ(e$H ) = k − « = 0 o
Symmetry
·" = ;̑ = 25:)4
Iö
̑
<['] = e$H⋅"
Re(«)
| ' = ñ e$H̑ e$H⋅"
̑
=0

§ Band stop filter can be used for suppression of the 50 Hz o

interference.
§ A zero point in the proximity of the unit circle leads to a minimum
of the frequency response, which means the corresponding
frequency is partially suppressed.

366
7.2 Description in frequency domain –
Transfer function of IIR systems
§ Calculation of the frequency response of a causal IIR-system <['] + |[']
|[' + 1] − 0.8|['] = <[' + 1]
T
0.8
1
ñ « =
1 − 0.8« *9 ñ(e$H )
$H
1 5
ñ(e ) =
1.64 − 1.6 cos ;

p 2p 3p ;
∠ñ e$H
0.8 sin ;
∠ñ e$H = − tan*9 53.13
1 − 0.8 cos ;
p 2p 3p ;

367
7.2 Description in frequency domain –
Transfer function – zeros and poles
Im(ë) Pass band
Simple band-pass filter
(poor man’s filter) o o
o
o
o Rejection range
3̑
o
Re(ë)
o
o
o
o o

The more zero points, the better the filter, but also the more complex the filter, and the longer the impulse response.

368
7.2 Description in frequency domain –
Transfer function – zeros and poles
§ A pole on the unit circle leads to a resonance within the frequency
response. The corresponding frequency is infinitely amplified, i.e., z-plane
the transfer function is instable. Im(ë)
ó−ë =0
1
x &
d(e'X ) = =∞
ó−ë 'Y
Re(ë)

x
§ A pole in the proximity of the unit circle leads to a maximum within the
frequency response, which means the corresponding frequency is
strongly amplified.

369
 7.2 Description in frequency domain –
Transfer function – zeros and poles
« − «9 « − «% … « − «7
ñj « = k# « i*7
« − "9 « − "% … « − "i
Examples of pole-zero point fi9 fi% … fi" $[ C @C @ …@C * Ä @Ä @ …@Ä
distributions and their corresponding ñ e$H = k# e G " Q G " X ]
‡9 ‡% … ‡R
frequency responses:
ñ ñ
Im(«)
;! ;!
‡9 x o x
·9 5 5
"9 x
fi9 ‡% fi% −5/2
Ñ9 −5
Ñ% ∠ñ ∠ñ
o o Re(«)
«% «9
· ñ ñ
"%x %

;! x ;!
o x o x
5 x 5
−5 −5
∠ñ ∠ñ
370
7.2 Description in frequency domain –
Transfer function – zeros and poles
Zero points and poles at the same point cancel each other out.
Im(H)
(* − ë) (*∗ − ë)
d(ë) = ⋅ ∗ =1
(* − ë) (* − ë)

xo
Re(H)
Pole=Zero point -> one (identity)
Pole~Zero point -> approximately one
xo

371
7.2 Description in frequency domain –
Transfer function – zeros and poles
Inverse filter: all poles are converted to zeros and all zeros are converted to poles.
Im(«) Im(«)
z-plane
z-plane

o x Unstable

3x 3x
x o o x
Re(«) Re(«)

o x

Inverse filter
372
7.2 Description in frequency domain –
Transfer function – zeros and poles
Summary
§ Magnitude of the frequency response at the frequency ; can:
- Be increased by placing a pole in the proximity of the unit circle.
- Be reduced by placing a zero point in proximity of the unit circle.

§ Poles or Zero points on the origin:

- Do not influence the frequency response .
- Add a phase of ±Q& to the phase of frequency response, which means a pure delay.

§ Adjacent poles and zero points can cancel each other.

§ Stable causal system: all poles are inside the unit circle.

373
7.2 Description in frequency domain –
Transfer function – stability

§ A LTI system will be BIBO-stable (bounded input – bounded output) if for the impulse response,
the following applies:
+

V ℎ(l) < ∞
:-#
Im(«)
§ The effect of discrete poles on the
Higher oscillation
impulse response can be illustrated frequency
as follows: Longer decay
time

Stable Re(«)

|«| = 1
Unstable

374
7.2 Description in frequency domain – Impulse response caused
Poles and zeros – impulse response by poles at the marked
Im(«) point.

x
x

x
x
x x

-1 -0.5 0 0.5 1
x x x x x x
Re(«)

Stability limit:
unit circle « = 1 375
7 Discrete time LTI systems - summary
z-domain
Poles and
1(H) Zeros

orm ë=
f
ns
- t ra e jÉ
z

Frequency domain
n-domain
ℎ[B] Fourier transform „
1(e‚‰ )

Impulse response, Pass band and

output sequence Stop band

376