Digital Signal Processing and its

Applications
Introduction
Objectives of DSP
 Primary Objective: Transition from continuous-independent variables to
discrete-independent variables.
 Example: In an audio signal, the independent variable is time (continuous),
and the dependent variable could be voltage, current, or magnetic field,
which can also be continuous.
 The course will explore signals where both the independent and
dependent variables are continuous, as is typical in real-life situations.

Concept of Continuum
 Continuum: In continuous signals, no matter how close two values of the
independent variable are, there will always be another value in between.
 This idea extends beyond one-dimensional signals (like time) to multi-
dimensional signals, such as images (2D) or 3D spaces.
 Real-life signals often involve both continuous independent and dependent
variables.

Key Objective of the Course

 The goal is to discretize the independent variable of a continuous
signal while maintaining the continuity of the dependent variable.
 Discrete signals are chosen at strategic intervals to avoid loss of
information.
 For example, an audio signal might be sampled at discrete time points, yet
the audio signal itself is treated as continuous.

Discretization
 Discrete signals break the continuity of the independent variable, making
it impossible to find values between two chosen sample points.
 Why discretize?: Discretization is necessary for processing signals on
modern computing devices.
 Despite the theoretical nature of continuous signals, modern technology
allows us to capture and process them accurately, making discretization
less of a concern.

What is Signal Processing?

 Signal: The variation of a dependent variable with respect to an
independent variable. Examples include voltage over time, temperature
over space, and image intensities.
 Signal Processing: The modification and separation of signals.
 Modification: Transforming a signal’s characteristics (e.g.,
enhancing one part of the signal).
 Separation: Isolating components of a signal, such as separating
male and female voices in an audio recording.
Practical Example: Audio Signal Processing
 In an audio recording with male and female voices, the goal might be to:
 Modify: Enhance or suppress certain components of the signal
(e.g., amplify female voices while suppressing male voices).
 Separate: Isolate the male and female voices into separate signals.
 The practical challenge is that translating these requirements into a
system language for automation requires multiple steps, which will take
time to build and understand.

The Importance of Systematic Approach in Signal Processing

 To automate the process, we need to:
 Build a formal description of the system.
 Identify and categorize signals and their properties.
 Create domains in which systems can be characterized and
analysed.

Introduction to Sampling
What is a satisfactory way to discretize a signal?
 Satisfactory means ensuring that no critical information is lost during
discretization. A loss of information is undesirable, and the challenge is
determining what constitutes critical information in a signal.
 Discretization refers to using only specific values of the signal at
selected points of the independent variable (commonly time).
 Sampling is the process of taking values of the signal at uniformly spaced
intervals.
 In audio signals, the independent variable is time, and uniform sampling
would mean selecting signal values at regular time intervals.

Uniform Sampling
 A signal can be sampled by choosing values at regular intervals, called the
sampling period (T).
 Samples are taken at T, 2T, 3T, … and similarly at negative
intervals, such as −T, −2T, ….
 The key goal is for this sampling process to be satisfactory, meaning it
does not lose important aspects of the signal.
 Does uniform sampling cause any loss of critical information?
 To examine this, the behaviour of a sinusoidal signal (sine wave)
is studied, as sine waves are fundamental in signal processing.

Why Focus on the Sinusoid?

 Sine waves are chosen because they are smooth and have certain
mathematical properties that make them ideal for DSP analysis:
 The derivative, integral, and summation of sine waves of the same
frequency remain sine waves of the same frequency.
 This persistence under mathematical operations makes the sine
wave a fundamental building block in signal processing.
 Limitations of non-sinusoidal waves: When two triangular waves of
the same frequency are added, the result is not a triangular wave but a
collection of straight-line segments. Similar issues arise when adding
 square waves of the same frequency. This shows how more complex
waveforms do not retain their characteristics when manipulated.
 The contrast between sine waves and other waveforms demonstrates why
sine waves are so pivotal in signal processing: their properties are
preserved under operations like addition, differentiation, and integration.
 Mathematically, sinusoids are smooth and analytic, meaning they retain
their form under differentiation and integration. This is useful for electrical
circuits, where relationships often involve derivatives and integrals. For
example, inductors and capacitors relate voltage and current through
derivatives. Thus, sinusoids remain unchanged in form through various
operations, making them ideal candidates for processing in discrete
systems.

Why sine waves are used in the field of communication?

(Optional)
 A sine wave has a single frequency in the frequency domain. In contrast,
waves like square or triangular waves are made up of a fundamental
frequency and many higher frequencies (harmonics). These extra
frequencies cause interference and distortion during transmission. Filtering
out these harmonics increases system complexity and reduces efficiency.
 Sine waves are smooth and continuous, meaning they don’t have abrupt
changes (like sharp edges in square or triangular waves). This avoids
issues like high-frequency noise or power loss during transmission [see
here].
 Most communication devices, like antennas and oscillators, naturally
generate sine waves because they resonate at specific frequencies.
 Sine waves are ideal carriers for modulation techniques (e.g., AM, FM,
PM) because their properties make it easy to embed information in their
amplitude, frequency, or phase. Other waveforms with multiple
frequencies complicate the modulation and demodulation process.
 Sine waves occupy the least bandwidth for a given frequency, which is
critical in communication systems to avoid signal overlap.
 Sine waves are solutions to linear differential equations that describe
natural systems. They’re easy to analyse and process using mathematical
tools like the Fourier Transform, which decomposes signals into sine waves
 Sine waves help in achieving the same goals (carrier signal, efficient
modulation) without introducing harmonics or abrupt transitions, making it
much easier than other waves.

What Properties of Sine Waves Make Them Easy to Embed Information?

 A sine wave can be completely described by just three parameters:
Amplitude, Frequency and Phase
 These properties make it easy to modify the sine wave to encode
information:
 Amplitude Modulation (AM): Change the height of the wave.
 Frequency Modulation (FM): Change how fast the wave
oscillates.
 Phase Modulation (PM): Shift the wave in time.
Why Do Abrupt Changes Lead to Power Loss or High-Frequency Noise?
 Discontinuities in time (like the sharp edges in square waves) translate to
high frequencies in the frequency domain.
 As frequency increases, signals tend to attenuate more when traveling
through mediums like air, cables, or optical fibres. This is because higher-
frequency signals interact more with the medium's particles, leading to
greater energy loss (e.g., heat, scattering).
 To overcome this attenuation, you need to amplify the signal, which
requires more power.

What is Sampling?
Sampling is taking measurements of a continuous wave (like a sine wave) at
regular time intervals. For example:
 A 0.1kHz sine wave (100 Hz) is sampled at 1kHz (every 1 ms).
 You record values like {0, 1, 0, −1, 0, 1, 0, −1, …}.

The confusion during sampling – Time domain view

When you sample a sine wave, other sine waves with different frequencies can
also pass through those same sampled points. For example:
 The 0.1 kHz sine wave gives samples like {0, 1, 0, −1, …}.
 A 1.1 kHz sine wave (fs−0.1) or 2.1 kHz (2fs−0.1) could produce exactly
the same samples at the same times.
 Both waves pass through the same sample points because the sampling
misses enough details to distinguish them. In essence, the "fast" wave’s
extra cycles get folded into the slower pattern – Aliasing.
In addition to frequency, sampling also creates phase ambiguity. This means
that the relative position (upward or downward movement) of the sine wave at
the sample points might belong to:
 A sine wave moving upward at those points.
 Or a sine wave moving downward, but with a phase shift.

The confusion during sampling – Frequency domain view and its

solution
 Sampling a signal of frequency f at a rate 𝑓𝑠 introduces peak at f, and

(𝑘𝑓𝑠, where 𝑘 is an integer).

periodic copies of the signal's spectrum at multiples of the sampling rate

 These replicas occur because sampling a signal in the time domain

corresponds to convolving the signal's spectrum with an impulse train in
the frequency domain (see here).
 When the sampling rate (fs) is too low, the replicas of the spectrum
overlap with the original spectrum, causing aliasing and the confusion.

twice the highest frequency component (fmax) of the original signal: fs ≥

 To avoid aliasing and the confusion, the sampling rate (fs) must be at least

2fmax. This ensures that the original spectrum and its replicas are
sufficiently separated in the frequency domain, and thus we can easily
separate them and get the original spectrum back.
  Practical Anti-Aliasing measures: Before sampling, a low-pass filter
fs
(anti-aliasing filter) is applied to remove frequency components above ,
2
ensuring no aliasing occurs. This filter ensures that only frequency
components below the Nyquist frequency (fs/2) are present.

Why do Spectrum Replicas occur?

 Sampling a signal involves multiplying it by a periodic impulse train in the
time domain:
∞
1
x s ( t ) =x ( t )∗ ∑ δ ( t−nT s ) , where T s= .
n=−∞
fs

 In the frequency domain, this is equivalent to convolving the original

spectrum X ( f ) with the spectrum of the impulse train:
∞
X s ( f )= X ( f )∗ ∑ δ ( f −kfs )
k=−∞
 The convolution creates copies of X ( f ) centred at every multiple of fs, i.e.,
f ± kfs

Types of Sampling
There are two main types of sampling based on the type of signal:
 Nyquist Sampling: Used for baseband signals.
 Bandpass Sampling: Used for passband signals.

Importance of Bandpass Sampling

 Nyquist Sampling: For passband signals, applying Nyquist sampling
would require much higher sampling frequencies, which might not always
be feasible.
 Example: A passband signal centred at 1 GHz with a bandwidth of 5 MHz
requires sampling at a rate greater than 2 GHz according to the Nyquist
criteria.
 Bandpass Sampling: By applying the bandpass sampling theorem,
passband signals can be sampled at lower rates, avoiding aliasing without
the need for higher sampling frequencies.

Sampling Theorem for Bandpass Signals

 A bandpass signal can be recovered from its sampled signal if
the sampling frequency (Fₛ) satisfies the following condition:
2∗f H
F s= where k is the ratio of the upper cutoff frequency to the
k
f
bandwidth: [ k = H ]
B
 Minimum Sampling Frequency:

2∗f H
 If F s= , this is the minimum sampling frequency required to
k
recover the signal from its samples.
 2∗f H
 If F s> , recovery is still possible.
k
2∗f H
 If F s< , it is impossible to recover the bandpass signal from
k
the sampled signal.
 Simplified Formula:
fH
 By substituting k = into the equation for F s, we get:
B
F s=2∗B This is a simplified result that can be used directly for
solving problems.
 Example Problem:
 Upper Cutoff Frequency (Fₕ) = 6 kHz; Lower Cutoff Frequency (Fₗ)
= 4 kHz
 Bandwidth = 6 kHz – 4 kHz = 2 kHz
 Minimum Sampling rate = 2*2 kHz = 4 kHz

Discrete Systems
 A discrete system takes a sampled input (denoted as x[n]) and produces a
sampled output (denoted as y[n]).

 The system operates according to the Nyquist-Shannon sampling theorem,

which ensures that the signal is sampled at a rate sufficient to reconstruct
the original signal.
 The system can perform operations on the discrete samples (such as
separating frequencies) and then reconstruct the output using a low-pass
filter to obtain the desired components.
 The system must ensure that a sine wave input produces a sine wave
output of the same frequency, with changes in amplitude and phase only.

Phasors in Discrete Systems

 A phasor is a complex number used to represent a sinusoidal signal. It
captures the amplitude, frequency, and phase of a sinusoid in a single,
compact form.
 A sinusoidal signal (like a cosine wave) is typically written as:
x (t )= Acos ( ω t +ϕ )
where: A - amplitude; ω - angular frequency; ϕ – phase.
 Instead of working with this directly, we represent it as a phasor in the
complex exponential form/a vector in the complex plane.
o The magnitude of the vector is the amplitude (A).
o The angle of the vector is the phase (ϕ).
 The sinusoid itself is the projection of the rotating phasor onto the real
axis:
x (t )=ℜ [ A e j (ωt +ϕ ) ] =Acos ( ω t+ ϕ )
Why Do We Need Phasors?
 Simplified Calculations:
 Sinusoids involve trigonometric functions (cos and sin), which can be
cumbersome to handle mathematically. By converting them into
complex exponentials, we can use algebra instead of trigonometry.
 For example, adding two sinusoids of the same frequency but different
phases become as simple as adding two complex numbers.
 System Analysis:
 When a sinusoidal signal passes through a linear time-invariant (LTI)
system, its frequency does not change, but its amplitude and phase
might. Phasors let us focus on these changes directly, without worrying
about the sinusoid’s oscillation.
 Frequency-Domain Representation:
 In the frequency domain, signals and systems are analysed in terms of
their frequency components. Phasors are central to this approach,
making tools like Fourier Transform and frequency response much
easier to understand and use.

System Requirements for Phasor Behaviour

A system's behaviour on phasors (or equivalently, sinusoids) should satisfy the
following requirements for analysis to remain simple and consistent:
 Experiment 1: Single-Frequency Phasor Input
o When a sinusoidal signal (phasor) of frequency ω0\omega_0ω0 is
input, the system outputs another phasor of the same frequency
ω0\omega_0ω0, but with:
o Modified Amplitude: The system might amplify or attenuate the
signal.
o Modified Phase: The system might shift the phase of the signal.
o Example: Input: Aej ω 0 n
o Output: H ( ω 0 ) Aej ω0 n, where H ( ω 0 ) is the system's frequency
response at frequency ω 0.
 Experiment 2: Different-Frequency Phasor Input
o When a sinusoidal signal (phasor) of a different frequency ω1\
omega_1ω1 is input, the output must also have the same frequency
ω1\omega_1ω1, but again with changes in amplitude and phase.
The system's response depends on the frequency.
o Example: Input: B e j ω n 1

j ω1 n
oOutput: H ( ω 1) B e
 Experiment 3: Combination of Two Phasors
o When two phasors of different frequencies ω 0 and ω 1 are input
together, the system must process each phasor independently:
o The amplitude and phase of each phasor must change as though
they were input alone.
o The output phasors should add up to match the input combination
after processing.
o Example: Input: Aej ω 0 n + B e j ω n 1
 Output: H ( ω 0 ) Aej ω0 n, + H ( ω 1) B e
jω n 1
o

Key System Demand

The key demand is that the output phasors must follow predictable rules based
on the system's behaviour. Specifically:
 The frequency of each phasor remains unchanged (no distortion or
mixing of frequencies occurs).
 The amplitude and phase of each frequency component are modified
consistently according to the system's frequency response.
This predictability is crucial because it allows us to analyse the system using
phasor algebra rather than solving differential or difference equations every
time.

System Properties for Phasor Behaviour

For a system to behave predictably with phasors, it must satisfy three
fundamental properties:
1. Linearity
o A system is linear if it satisfies the principle of superposition:
 Additivity: The response to a sum of inputs equals the sum
of the responses to each input.
If x 1 ( t ) → y 1 (t )∧x 2 ( t ) → y 2 ( t ) ,then x1 ( t )+ x 2 ( t ) → y 1 ( t )+ y 2 ( t ) .
 Homogeneity: If the input is scaled, the output is scaled by
the same factor.
If x ( t ) → y ( t ) , then c ⋅ x (t ) → c ⋅ y ( t ) .
o Why it matters for phasors: Phasors rely on the ability to add
sinusoids of different frequencies and have the system process
them independently. Without linearity, the output could include
unexpected terms (e.g., frequencies not present in the input)
2. Shift Invariance (Time Invariance)
o A system is time-invariant if its behaviour does not change with
time shifts. Mathematically:

If x ( t ) → y ( t ) , then x ( t−t 0 ) → y ( t−t 0 ) .

o Why it matters for phasors: For phasors, time invariance ensures

that the system's response depends only on the input frequency,
not when the input signal starts. This is necessary for the concept of
a consistent frequency response.
3. Stability
o A system is stable if bounded inputs produce bounded
outputs. This means the system does not "blow up" or produce
infinite outputs for finite inputs.
o Why it matters for phasors: Phasor analysis assumes that the
system response is well-behaved. Stability ensures that the
amplitude of sinusoidal inputs doesn’t grow uncontrollably.
o BIBO stability does not guarantee predictable behaviour for
unbounded inputs—these could result in either bounded or
unbounded outputs.
 o The necessary and sufficient condition for a system to be BIBO
stable (in the context of Linear Shift-Invariant systems) is that the
impulse response must be absolutely summable.
o For non-LSI systems, this conclusion may not hold, and the system
could still be unstable even with an absolutely summable impulse
response

Impulse Response and Characterization of LTI Systems

 Definition:
 The impulse response of a system, denoted as h[n], is the output of a
Linear Time-Invariant (LTI) system when the input is a unit impulse signal.
 Key Insights:
 The impulse response contains all the information about how the system
affects different frequency components of an input signal.
 It "tests" the system by introducing all possible frequencies simultaneously
(since the impulse contains equal energy across all frequencies).
 By observing how the system responds to the impulse, we understand how
it processes each frequency.
 Importance: Knowing allows us to:
 Predict the Output for Any Input: The system's output can be
computed for any arbitrary input using convolution:
 Analyse the System's Frequency Response: The frequency response
describes the amplitude and phase effects of the system on sinusoids of
different frequencies.

Convolution
 A fundamental operation in signal processing that combines two signals to
produce a third signal. It is used extensively in systems analysis to
understand how a system responds to an input signal.
 For two discrete signals x[n] (input) and h[n] (impulse response), the
convolution is defined as:
∞
y [ n ] =x [ n ]∗h [ n ] = ∑ x [ k ] h [ n−k ]
k=−∞

Properties
1. Commutativity
a. The order of convolution does not matter. In other words, order in
which two LTI systems are placed does not matter.

h1 [ n ]∗h 2 [ n ]=h2 [ n ]∗h1 [ n ]

2. Associativity
( x [ n ]∗h 1 [ n ] )∗h2 [ n ] =x [ n ]∗( h1 [ n ]∗h 2 [ n ] )
a. Associativity allows cascading of systems to be analysed in any
sequence.
b. In a cascaded system, the overall impulse response h_total [n] is
the convolution of individual impulse responses.
htotal [ n ] =h 1 [ n ]∗h2 [ n ]
c. Input x[n] produces the output:
 y [n]=x [n]∗htotal [n]=x [n]∗( h ¿ ¿ 1 [ n ]∗h2 [ n ] )¿

3. Distributivity
a. In a parallel system, the overall impulse response is the sum of the
individual impulse responses h1 [ n ] and h2 [ n ]
b. The output of the parallel system is:
y [ n ] =x [ n ]∗( h 1 [ n ] +h2 [ n ] )=( x [ n ]∗h1 [ n ] ) + ( x [ n ]∗h2 [ n ] )
c. In a parallel system, the combined response is:
htotal [ n ] =h 1 [ n ] +h2 [ n ]
d. Input x [n] produces the output:
y [ n ] =x [ n ]∗htotal [ n ] =( x [ n ]∗h 1 [ n ] ) + ( x [ n ]∗h2 [ n ] )

Causality in Systems
In any system where time is the independent variable (such as in signal
processing), the system's behaviour depends on the direction of time. Systems
operate based on the principle that information flows forward in time and not
backward, meaning the system's output at any point in time depends on past or
present inputs, but not on future inputs.

Real-Time vs Offline Systems

 Real-time systems are those where outputs are generated immediately
based on past and present inputs. In these systems, the output at any
time depends only on the input up to that point in time, meaning future
inputs cannot be considered.
 Offline systems are systems where outputs can be computed using both
past and future inputs. These systems are not constrained by the need to
act in real time.

Definition of Causality
A system is considered causal if the output at any time n only depends on
present and past inputs (i.e., x [k ] for k ≤ n). Specifically, for any two input
sequences x 1 [ n ]and x 2 [n], if x 1 [n]=x 2 [n] for all n ≤ n0, the outputs y 1 [n] and y 2 [n]
should also be equal for all n ≤ n0. This ensures that the system does not rely on
future values to generate the current output.

Experiment for Causality

 Given two input sequences x 1 [n] and x 2 [n], if they are identical for all n ≤ n0,
the system should produce identical outputs y 1 [n] and y 2 [n] for all n ≤ n0.
This validates the causal behaviour of the system: if the system has the
same input history up to a point, it must produce the same output up to
that point as well.

Causality and Cause-Effect Relationship

Causality is rooted in a cause-effect relationship where the output is the
effect of the input applied over time. The system reacts to past and present input
values and does not anticipate or rely on future inputs.
Causality in LTI Systems
For LTI systems, causality is closely related to the system’s impulse response:

 An LTI system is causal if its impulse response h [n ] satisfies the

condition h [ n ] =0 for all n< 0. This means that the system’s response at
time n depends only on the current and past values of the input signal, not
on future values.

 If the impulse response has values for n< 0, this implies the system’s
output would rely on future inputs, making it non-causal.

Thus, in causal systems, the output at time n is determined only by input values
at time n or earlier, ensuring that the system doesn’t depend on future inputs.

Frequency Response
 Formal Definition: The frequency response is mathematically defined
using a sum over all values of the system's impulse response h[n],
weighted by a complex exponential. This can be written as:
∞
H (ω)= ∑ h[n]e− j ω nwhere h n is the system's impulse response at time n ,
n=−∞
and ω is the angular frequency of the complex exponential input.
 This sum essentially tells you how the system responds to all the different
frequency components of the input signal.
 Linear Shift-Invariant (LSI) systems are a type of system that behaves in a
predictable manner. They are linear (their response to an input is directly
proportional to the input) and shift-invariant (the system's response to a
shifted input is just a shifted version of the original output).
 A key property of these systems is that when you input a rotating complex
exponential (a signal that rotates in the complex plane), the output of the
system will have the same angular velocity (frequency). In other words,
the system does not alter the frequency of the input signal—it only
modifies its amplitude or phase.
 Stability is essential for this behaviour. If the system is stable, it will
always have a frequency response, meaning it will respond predictably to
complex exponential inputs.
 The frequency response of a system describes how the system reacts to
sinusoidal (rotating) signals. When a rotating complex exponential is fed
into the system, the output will be a complex exponential of the same
frequency but possibly with a different amplitude or phase.
 Only stable LSI systems can have a frequency response because they
provide a stable and predictable way to analyse and understand signals in
the frequency domain.

Why only LSI systems can have Frequency Response? Simple Explanation
(Optional):
 Frequency response is a property that tells us how a system processes
or changes different frequency components of an input signal. It's
essentially a tool to analyse how a system behaves in the frequency
domain.
 For a system to have a meaningful and well-defined frequency response, it
must satisfy certain conditions:
 o Linear: This means that if you scale or combine input signals, the
output behaves in a predictable, proportional way. (Example: If you
double the input, the output also doubles.)
o Time-Invariant: This means the system's behaviour doesn't
change over time. (Example: If you give the system an input today
or tomorrow, the response is the same.)
 Why only LTI systems?
o Linear systems ensure that each frequency component in the input
signal is processed independently. If the system isn't linear,
different frequency components can mix or distort unpredictably.
o Time-invariance ensures that the system’s behaviour doesn’t
depend on when you apply the signal. Without this, the frequency
response would change over time, making it meaningless.
 Non-LTI systems (e.g., nonlinear systems or time-varying systems) don't
preserve this clean relationship with frequency. They might generate new
frequencies (nonlinear distortion) or behave unpredictably over time. As a
result, we can't describe their behaviour using a straightforward frequency
response

Dot Product and its use in LSI System Interpretation

 Dot product of two vectors gives you a measure of how much one vector
points in the direction of the other. For two vectors v=[ v 1 , v 2 ] and u=[u1 ,u 2] ,
the dot product is: v ⋅u=v 1 u1+ v 2 u2, which is a scalar (a single number).
 This scalar can be interpreted as the "component" of the vector v in the
direction of u. By multiplying this scalar with u, we get the actual
component of v in the direction of u.
 The goal here is to apply the same principle (using the dot product) to LSI
systems. The frequency response H (ω) of an LSI system can be thought of
as the dot product between:
o The impulse response h [n ] of the system (how the system reacts to
a single impulse),
o A complex exponential at frequency ω (which represents a pure
tone at a certain frequency).
 The dot product measures how much h [n ] aligns with this specific
frequency. If they align well, H (ω) is large, meaning the system strongly
responds to that frequency (amplifies or allows to pass through). If
they don’t align, H (ω) is small, meaning the system suppresses that
frequency.
 The complex exponential e jωn represents a tone at frequency ω .
o The system’s impulse response h[n] shows how it reacts to different
frequencies.
o H (ω) tells you how much the system's impulse response aligns with
that frequency.
 If the alignment is good (i.e., h [n ] matches the frequency well), H (ω) is
large, meaning the system lets that frequency pass through or amplifies it.
If the alignment is poor, H (ω) is small, meaning the system blocks or
attenuates that frequency.
