SIGNALS AND SYSTEMS

Asst Prof. Betül Gürünlü

 Sampling
• In signal processing, sampling is the reduction of
a continuous-time signal to a discrete-time
signal. A common example is the conversion of
a sound wave to a sequence of "samples".
A sample is a value of the signal at a point in
time and/or space; this definition differs from
the usage in statistics, which refers to a set of
such values.
• A sampler is a subsystem or operation that
extracts samples from a continuous signal. A
theoretical ideal sampler produces samples
equivalent to the instantaneous value of the
continuous signal at the desired points.
Signal sampling representation. The continuous
• The original signal can be reconstructed from a
sequence of samples, up to the Nyquist limit, by signal S(t) is represented with a green colored
passing the sequence of samples through a type line while the discrete samples are indicated by
of low pass filter called a reconstruction filter. the blue vertical lines
Theory
• Sampling can be done for functions varying in space, time, or any other
dimension, and similar results are obtained in two or more dimensions.

• For functions that vary with time, let s(t) be a continuous function (or
"signal") to be sampled, and let sampling be performed by measuring the
value of the continuous function every T seconds, which is called the
sampling interval or the sampling period. Then the sampled function is
given by the sequence:

s(nT), for integer values of n.

The sampling frequency or sampling rate, fs, is the average number of
samples obtained in one second, thus fs = 1/T. Its units are samples per
second or hertz e.g. 48 kHz is 48,000 samples per second.
Time and Frequency Domain Analysis
• A time domain analysis is an analysis of physical signals, mathematical
functions, or time series of economic or environmental data, in reference
to time. Also, in the time domain, the signal or function's value is
understood for all real numbers at various separate instances in the case of
discrete-time or the case of continuous-time. Furthermore, an oscilloscope
is a tool commonly used to see real-world signals in the time domain.
• Moreover, a time-domain graph can show how a signal changes with time,
whereas a frequency-domain graph will show how much of the signal lies
within each given frequency band over a range of frequencies.
• In general, when an analysis uses a unit of time, such as seconds or one of
its multiples (minutes or hours) as a unit of measurement, then it is in the
time domain. However, whenever an analysis concerns the units like Hertz,
then it is in the frequency domain.
How is Time Domain Analysis Different from
Frequency Domain?
• Frequency domain is an analysis of signals or mathematical functions, in
reference to frequency, instead of time. As stated earlier, a time-domain graph
displays the changes in a signal over a span of time, and frequency domain
displays how much of the signal exists within a given frequency band concerning
a range of frequencies. Also, a frequency-domain representation can include
information on the phase shift that must be applied to each sinusoid to be able to
recombine the frequency components to recover the original time signal.
• Furthermore, you can convert a designated signal or function between the
frequency and time domains with a pair of operators called transforms.
Moreover, a perfect example of a transform is the Fourier transform. Which
converts a time function into an integral of sine-waves of various frequencies or
sum, each of which symbolizes a frequency component. The so-called spectrum
of frequency components is the frequency-domain depiction of the signal.
However, as the name implies, the inverse Fourier transform converts the
frequency-domain function back to the time function.
Nuances Between Frequency and Time
Domain
• Time domain analysis provides the transitory response of a system to
be analyzed, and it permits a better understanding of the flow of both
mechanical and electrical energies. In general, this includes wave
propagation, the structural changes of a system, and electric potential
generated by external excitations.
• Whereas for the frequency domain, visualization tools such as a
spectrum analyzer are commonly in use when visualizing electronic
signals. Also, some specialized signal processing techniques make use
of transforms, and this results in a joint time-frequency domain.
Moreover, the instantaneous frequency is a critical link between the
time domain and the frequency domain.
Will Time Domain Analysis or Frequency
Domain Analysis be Used More Often?
• Time domain analysis is particularly useful for circuit designs with antennas where a designer may
encounter stray signals, reflections, or ground bounce signals. Time domain signal processing
enables an engineer to separate extraneous signals in time from the desired signal, thereby
identifying the contaminated signals.
• In general, using a frequency domain will simplify analysis mathematically for the system running
it. Many prominent SPICE tools will primarily function through the frequency-domain for this
relevance, efficiency, and accuracy for their analytical functions.
• Seeing a system from the viewpoint of frequency will often provide an innate understanding of
the measured quality that encompasses the behavior of the system. The scientific community
now offers various terminology to describe such characteristic physical system behavior in
reference to time-varying inputs. This includes terms like frequency response, bandwidth, phase
shift, gain, and resonant frequencies, to name a few.
• One of the most familiar and universal examples of frequency content in signals is perhaps audio
signals, such as music. In this case, the frequency-domain analysis gives a better understanding
than time domain analysis because music is tacitly based on the breaking down of intricate
sounds into their separate component frequencies.
Sampling and Aliasing
What Is this Course All About ?
• To Gain an Appreciation of the
Various Types of Signals and Systems
• To Analyze The Various Types of
Systems
• To Learn the Skills and Tools needed
to Perform These Analyses.
• To Understand How Computers
Process Signals and Systems
 Discrete-time Signals and Computers
• Up to now we have been studying continuous-time signals (also
called analog signals) such as

x(t)  Acos(o t   )
• However, digital computers and computer programs can not
process analog signals.
• Instead they store discrete-time versions of analog signals
x[n]  x(nTs )
• This is because digital computers can only store discrete
numbers.
– There are computers called analog computers which do process
continuous-time signals
• Since the computer only stores numbers, how does one know
what continuous-time signal it represents?
 Sampling
• We can obtain a discrete-time signal by sampling a
continuous-time signal at equally spaced time
instants, tn = nTs
x[n] = x(nTs) -∞ < n < ∞
• The individual values x[n] are called the samples of
the continuous time signal, x(t).
• The fixed time interval between samples, Ts, is also
expressed in terms of a sampling rate fs (in samples
per second) such that:
fs = 1/ Ts samples/sec.
Continuous-to-Discrete Conversion
• By using a Continuous-to-Discrete (C-to-D)
converter, we can take continuous-time signals and
form a discrete-time signal.
• There are devices called Analog-to-Digital converters
(A-to-D)
• The books chooses to distinguish an C-to-D converter
from an A-to-D converter by defining a C-to-D as an
ideal device while A-to-D converters are practical
devices where real world problems are evident.
– Problems in sampling the amplitudes accurately
– Problems in sampling at the proper times
 Discrete-Time Signals
• A discrete-time signal is
a sequence of numbers
and carry no 1

information about the 0.8

0.6
0.4

time-sequence. 0.2
0
-0.2 00 2 4 6 8 10

• Looking at the -0.4

-0.6
-0.8

following diagram, -1

which (gray or solid)

waveform are these
(red) samples associated
with?
 Discrete-Time Sinusoidal Signals
• Since a Fourier series can be written for any continuous-time
signal, let’s concentrate on sinusoids
• We define a normalized frequency for the discrete sinusoidal
signal.
x[n]  x(nTs )  Acos(nT s   )
 Acos(̂n   )

ˆ  Ts  
fs
̂ is the normalized or discrete-time frequency
Since we can have different signals with the same ̂, then
there can be an infinite number of continuous-time signal
which yield the same discrete-time sinusoid!
 Two Problems with Sampling
• Problem 1: How many samples are enough to
have to represent a continuous time signal?
1

0
0 2 4 6 8 10

-1

• In this figure, we have a continuous-time signal

sampled every .4 seconds (red samples) and
every 1 second (black samples).
 Discrete-Time Sinusoidal Signals
• Problem 2: Can a set of samples be represent
more that one continuous-time signal
• The discrete-time sinusoid
shown in the figure has 1

which can be obtain from, for

example, either a 1 second
sampled continuous-time 0
0 2 4 6 8 10
sinusoid with f = 0.2 Hz or 1.2
Hz.
• In the first case, where f = 0.2 -1

Hz, we have:
̂  2 (0.2)(1)  .4
 Discrete-Time Sinusoidal Signals

• In the first case, where f = 0.2

1
Hz, we have:
̂  2 (0.2)(1)  .4
• Since a sinusoid is periodic in 0
0 2 4 6 8 10

2, then for the case where

f=1.2 Hz
-1

x[n]  Acos(̂n   )
̂  2 (1.2)(1)  2.4  2.4  2  .4
x[n]  Acos(2.4 n   )  Acos(2 n  0.4 n   )  Acos(0.4 n   )
 Aliasing
• This example
1
illustrates that two
sampled sinusoids can
produce the same 0
0 2 4 6 8 10

discrete-time signal.
-1
1. cos [2π(0.2) t]
2. cos [2π(1.2) t]
• When this occurs we say that that these signals
are aliases of each other.
 Aliasing
• There are more alias signals for this example:
• 1. x(t) = cos (2π(0.2) t) => x[n] = cos (2π(0.2) 1n) = cos (0.4π n)
• 2. x(t) = cos (2π(1.2) t) => x[n] = cos (2π(1.2) 1n) = cos (2.4π n) = cos (0.4 π
n + 2 π n) = cos (0.4π n)
ˆl  0.4  2l for l  0,1,2,3,…
Since cos(2 -  )  cos( ), ̂l  0.4  2l for l  0,1,2,3,…
3. x(t) = cos (2π(.8) t) => x[n] = cos (2π(.8) 1n) = cos (1.6π n) =
cos (2 π n - 0.4 π n) = cos (0.4π n)
• In summary, (for l = positive 1

or negative integer):
0

̂o, ˆo  2 l, 2 l  ̂o

0 2 4 6 8 10

where ˆo is called the principal alias

-1
 Aliasing
• Let’s look at signals of the form: cos(ωlt)
cos(l t)  cos(̂l n)
sampled

ˆ l
where l   ˆl f s and ˆl  ˆo  2l, ˆo is the principal alias, and l is an integer.
Ts
(ˆ o  2l) f s
Therefore, l  ̂l f s  (̂o  2l) f s and f l 
2
(2l  ˆ o ) f s
since cos( )  cos( )  cos(2   ), then we can have ̂ l  2l  ̂o and f l 
2
(2l  ˆ o ) f s
or l  2l  ̂o and f l 
2

In our example, ̂o is 0.4 and f s  1. Then,

l  2l  ˆo  2l  0.4  0.4 ,1.6 , 2.4 , 3.6 , ....rad/sec
(2l  ˆ o ) f s 2l  0.4
  l  0.2  0.2, 0.8 1.2, 1.8, ..... Hz
2 2
 Shannon’s Sampling Theorem
• How frequently do we need to sample?
• The solution: Shannon’s Sampling Theorem:
A continuous-time signal x(t) with frequencies
no higher than fmax can be reconstructed
exactly from its samples x[n] = x(nTs), if the
samples are taken a rate fs = 1 / Ts that is
greater than 2 fmax.
• Note that the minimum sampling rate, 2 fmax ,
is called the Nyquist rate.
 Spectrum of the Discrete-time
Signal
0.5

-2.4 -1.6 -0.4 0.4 1.6 2.4

• There are an infinite number of frequency

components of discrete-time signal
• They consists of the principal along with the
other aliases (an infinite number of them).
 Nyquist Rate
• Shannon’s theorem tell us that if we have at
least 2 samples per period of a sinusoid, we
have enough information to reconstruct the
sinusoid.
• What happens if we sample at a rate which is
less than the Nyquist Rate?
– Aliasing will occur!!!!
 Ideal
Reconstruction
• The sampling theorem suggests that a process exists
for reconstructing a continuous-time signal from its
samples.
• If we know the sampling rate and know its spectrum
then we can reconstruct the continuous-time signal by
scaling the principal alias of the discrete-time signal
to the frequency of the continuous signal.
• The normalized frequency will always be in the range
between 0 ~ π and be the principal alias if the
sampling rate is greater than the Nyquist rate.
Ideal Reconstruction Continued
• If continuous-time signal has a frequency of ω, then the discrete-time
signal will have a principal alias of

ˆ  T  
s
f s
• So we can use this equation to determine the frequency of the
continuous-time signal from the principal alias:
  ˆfs  
ˆ
Ts
• Note that the normalized frequency must be less than  if the Nyquist
rate is used
2f MAX 2f MAX
̂  MAX Ts  2f MAX Ts   
fs f s ( 2 f MAX )
• And the reconstructed continuous-time frequency must be
̂ f s  f s f s
  2 f  ˆ f s  f     f max
2 2 2
 Oversampling
• When we sample at a rate which is greater than the Nyquist
rate, we say we are oversampling.
• If we are sampling a 100 Hz signal, the Nyquist rate is 200
samples/second => x(t)=cos(2π(100)t+π/3)
• If we sample at 2.5 times the Nyquist rate, then fs = 500
samples/sec
• This will yield a normalized frequency at 2π(100/500) = 0.4π
1

0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08

-1
 Oversamplin
•
g
Since we are greater than the Nyquist rate, the normalized
frequency will be < π which means it is the principal alias.
• And we get back the original continuous frequency when we
do the reconstruction
• f = 0.4πfs / 2π = 0.4π500 / 2π = 0.2 (500) = 100 Hz

0.5

-2.4π -1.6π -0.4π 0.4π 1.6π 2.4π

 Undersampling and Aliasing
• When we sample at a rate which is less than the Nyquist rate,
we say we are undersampling and aliasing will yield
misleading results.
• If we are sampling a 100 Hz signal, the Nyquist rate is 200
samples/second => x(t)=cos(2π(100)t+π/3)
• If we sample at .4 times the Nyquist rate, then fs = 80 s/sec
• This will yield a normalized frequency at 2π(100/80) = 2.5π
1

0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08

-1
 Undersamplin
•
g
Since it is > π, 2.5π is NOT the principal alias
• The principle alias is 2.5π - 2π = 0.5π

• Using 0.5π as the principal alias and performing a reconstruction we then

have:
f = 0.5π fs / 2π = 0.5π 80 / 2π = 0.5 (40) = 20 Hz and we have
reconstructed the wrong signal!!!

0.5

-2.5π -1.5π -0.5π 0.5π 1.5π 2.5π

 The Alias Problem due to
Undersampling

0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08

-1
 Aliasing and Folding
• Your book treats undersampling in terms of
aliasing and folding
• During reconstruction, both of these
phenomenon will produce erroneous results.
• The difference between aliasing and folding
has to do with which part of the spectrum
created the alias.
Discrete-to-Continuous Conversion

• An D-to-C converter uses the samples to

reconstruct the continuous-time signal by
interpolation.
• There are various interpolation algorithms
which may be used:
– Zero-Order Hold
– Linear
– Cubic Spline
 Interpolation
1 1

0 0
0 2 4 6 8 10 0 2 4 6 8 10

-1 -1

Zero-Order Hold Linear

• Oversampling always improves the reconstruction

• Best reconstruction is Low Pass Filter or what the text
calls: Ideal Bandlimited Interpolation
 Non-sinusoidal Signals
• Since a Fourier series can be written for any
continuous-time signal, the sampling and
reconstruction processes for any continuous-
time signal is the same
– Shannon’s Sampling theorem
– Nyquist Rate fs ≥ 2fmax to eliminate aliasing
– Oversampling to improve interpolation
– Ideal (low pass filter) Bandlimited interpolation
End of The Lecture