AV314 - Communication Systems I

(Analog Communication Systems)

Review of important concepts from Signals & Systems

References:
Lathi (BPL) - Chapters 2 and 3
Upamanyu Madhow (UM) - Chapter 2
Complex numbers
• A compact way to represent amplitude and phase information

• An important identity to keep in mind - Euler's identity

• How do we express cos() and sin() in terms of this complex exponential?

Signals
• Signals are functions of time
• Classification of signals

Continuous valued or discrete valued

Continuous or discrete time

Energy and power signals

Periodic and aperiodic signals

• The usual engineering approach applies here !

• When thinking about signals (or anything) we try to approximate it by simpler things that we can handle!

• Examples of “simple” signals


Impulse (Dirac delta) functions

What is the sifting property of impulse function?

Why is the impulse function important?
Signals

• Examples of “simple” signals


Indicator functions

How does the boxcar or indicator function help us in compactly specifying these two signals?
Signals
• Sinc functions

Why is the sinc function so important for us?

A small detour
• Recall that in our definition of the communication problem we had used a function error(d(t), m(t))

• Suppose m(t) is a voice signal. What is a good choice for error() ?


d(t) needs to be “similar” to m(t) - there is no need for them to be equal

• How do we define “similarity”?

• Signals are like vectors !


A simple illustration - a finite extent signal can be represented by a “closely sampled” sequence of values

The sequence of values is a vector!

There are better ways of thinking about signals as vectors - but this will serve as motivation


“Dot - product” of signals
Energy and Power
• Energy of a signal


Power of a signal

Exercise: Compute the power of a sinusoid

Recall that this is also a way to classify signals.

What is a system?
• Examples of systems ...
• We consider systems that transform an input signal x(t) to an output signal y(t)

• We need to think about systems ...


Why?

How do we think about systems?

Mathematical model of systems?

Functions/Compostion of functions

• Specific and useful models


Linear

Time invariant

What do you mean by a linear system?

What do you mean by a time invariant system?

Can you give an example of a non-linear system?

Can you give an example of a time-variant system?

What is a memoryless system?

Can you give an example of a memoryless system which is not (linear and time invariant)?
Linear Time Invariant (LTI) systems
• Why are LTI models useful?

• What an LTI system(remember that it is a model) completely?


Do non-LTI systems have impulse responses?


For any input we can obtain an output using convolution


How do you show this?

• Why are LTI system models useful for communication systems?

Linear Time Invariant (LTI) systems: Example
Linear Time Invariant (LTI) systems: Example

When faced with a need to compute

convolutions - use simple functions like
this!

Decompose your signal to such known

functions!
Linear Time Invariant (LTI) systems: Example

Obtain the convolution of the following signals.

A LTI model for a channel

Consider a wire


Can you suggest a LTI model for the wire?


What is the impulse response?


How valid do you think this LTI model is?
Fourier transforms
•
Why are we interested in Fourier transforms?

•
Signals (belonging to a large class) can be represented as a sum (or an integral) of complex exponentials
•
Complex exponentials are eigen functions of LTI systems

•
Can you derive this property?

•
Therefore, it is easy to find the output for any signal expressed as the sum/integral of complex exponentials
•
The Fourier transform of the input signal is important for us
•
The Fourier transform of the impulse response is important for us

•
Please review Fourier Series on your own (H/W)
Some Fourier transforms to keep in mind

• If
Fourier transforms

What is Fourier transform (and inverse Fourier transform?)


Quick review of some properties of Fourier transforms

•
Work/Derivation
Fourier transforms: some more properties

•
Work/Derivation
Fourier transforms: some more properties

•
Some Fourier transforms to keep in mind

• If
Properties of Fourier Transforms ...

• What is linearity property of Fourier transform?

• Why is this property important for us?
Properties of Fourier Transforms ...
• A shift in time domain corresponds to _____________ ?
• What is the time delay property ? Why is this important for us?

• A shift in frequency domain corresponds to ____________ ? Why is this important for us ?

Properties of Fourier Transforms ...
• Scaling a signal in time domain leads to ____________ ? Why is this important for us?
• Compression and expansion of a signal
Properties of Fourier Transforms ...
• Convolution of two signals in time domain corresponds to ____________ of their Fourier transforms? Why
is this important for us?

Group exercise: Find the Fourier transform of sinc(10t) sinc(4t)

(convolution of two boxes = trapezoid)
Properties of Fourier Transforms ...
• What is the complex conjugation property of Fourier transforms?
• Conjugation of a signal in one domain (say time) leads to conjugation and flipping of argument in the
other domain.
• Why is this important for us?
A problem to work out ...
Work/Derivation
New terminology - Bandwidth

• The set of frequencies occupied by a signal is its bandwidth

• For signals we usually talk about the one sided bandwidth or two sided bandwidth
• For physical (real) signals one sided bandwidth makes sense since the spectrum has conjugate
symmetry
• But we will mostly talk about two sided bandwidth (and two sided Fourier transforms)

• If a signal has all of its energy within a finite bandwidth, then how does the signal look like in time domain?

• But many real world signals are of finite extent in time domain.

• So how do we define bandwidth in these cases?

Bandwidth for time limited signals

• Let us consider a boxcar signal

s(t)
1

• New terminology - Energy spectral density

• Energy spectral density = magnitude squared of Fourier transform

• Bandwidth is defined as fractional energy containment bandwidth = what is the range of frequencies over
which 99% or 95% or 90% or some a fraction of the signal energy is contained

• Another way to define bandwidth is to use null-to-null bandwidth - but we use the above definition here.
New terminology - Baseband (low-pass) and Passband (band-pass) signals

• Baseband signals
• have their Fourier transforms near or containing DC
• so they pass through low pass filters - also called low pass signals
• Can you give some examples of baseband signals?
• Can you give an example of how the magnitude of the Fourier transform would look like?

• Passband signals
• have their Fourier transforms away from DC
• they pass through bandpass filters - also called band pass signals
• Can you give some examples of passband signals?
• Can you give an example of how the magnitude of the Fourier transform would look like?
Using Fourier transform domain for periodic signals

• We note that one of the conditions to be met for a Fourier transform to exist is that the signal be absolutely
integrable. This is not satisfied by any periodic signal with finite energy in a period (or non-zero power).

• In communication systems, we will have to deal with sinusoids and complex exponentials quite a lot. We
would like to have a common framework (that of the Fourier transform domain) to represent and think about
both types of signals.

• What do you think power spectral density is?

Using computers (Matlab) to think about Signals and Systems ...

• In AV314 we work with continuous time signals and systems for the most part

• How do we represent such signals and systems in a computer (Matlab)?

• We need to discretize both time as well as value taken by a signal

• Sampling and Quantization

• What is meant by sampling? (uniform sampling)

• We won’t worry about quantization here

Using computers (Matlab) to think about Signals and Systems ...

• A sampled signal is a discrete time signal

• Suppose the signal is known to be of finite extent,
then the discrete time signal = an array or a vector
in Matlab

• What should be the sampling frequency?

• Nyquist criterion - (but for a finite extent signal?)
• Oversampling for a better visual representation of
the signal.

• Please try out some operations on signals

• How to add two signals together?
• How to find out the product of two signals?
• How to find out the signal energy? (interesting!)
Using computers (Matlab) to think about Signals and Systems ...

• How can think about systems in Matlab?

• A single input single output system is a transformation - a function in Matlab

• A memoryless system is relatively easy to implement

• Elementwise transformation of an array

• Systems with memory

• A procedural definition can always be translated into a Matlab function

• What about LTI systems?

• LTI systems with impulse responses which are finite in time?

• Well, these are also discrete time signals!

Using computers (Matlab) to think about Signals and Systems ...

• How do we think about signals passing through LTI systems?

• Usually what are interested in is the output of an LTI system with an input signal
• The “conv” function can be used to easily obtain the convolution
• Let us try out an example
• Suppose we have two discrete time signals [1,2,3] and [1,2,3]. Compute the convolution of these
two signals using “conv” function in Matlab
Thinking about Fourier transforms
Implementing this approximation in Matlab

Only for k taking values

0,1,2, ... , N - 1
Implementing this approximation in Matlab

Some interesting periodicity

and symmetry properties
here!
Implementing this approximation in Matlab

Ts = 0.01;
fs = 0.2;
Ttotal = 1/fs;
t = 0:Ts:(Ttotal - Ts);

xn = cos(2 * pi * 10 * t) .* exp(-t);
Xk = Ts * fft(xn);

frequencies = (0:(length(t)-1)) * fs;

stem(frequencies, abs(Xk));

Representation of Fourier Series for Periodic

signals?
A special operation - filtering

• Filtering is a very common operation that we do in communication systems

• How can we simulate filtering in Matlab?

• Filtering using an LTI filter ? Well just use conv()!
• But we need the impulse response of the LTI filter
• Sampled impulse response!

• Doing the equivalent of continuous time operations in discrete time (will learn about in DSP)
• But the essential idea is that we can sample the impulse response of a continuous time filter and use that
in Matlab for a filtering operation

• We will discuss a recipe here without going into any details!

A recipe for filtering

• What is a LTI filter?

• A system with a frequency response having a specific “shape”
• Suppose the frequency response is denoted as H(f) (note that frequency response consists of both
magnitude response |H(f)| and phase response. We usually want the phase response to be “linear” for
filters with “good” properties
• So the requirement is put on |H(f)|’s “shape”

|H(f)|
Passband of filter

Stopband of filter

fp f

Low pass filter

A recipe for filtering
|H(f)|
Passband of filter

Stopband of filter

fp f

Ideal Low pass filter

|H(f)|
Transition band Passband of filter

Stopband of filter

fp fstop f

More realistic low pass filter response

A recipe for filtering

|H(f)|
Transition band Passband of filter

Stopband of filter
f
fp fstop
More realistic low pass filter response

|H(f)|
Ripple - tolerance in passband

f
fp fstop Ripple - tolerance in stopband
Still more realistic low pass filter response
A recipe for filtering
(0,1) (fp ,1)
|H(f)|
(fstop,0)

fp fstop f
A piecewise linear approximation is used in Matlab

• So suppose Matlab provides you with a function to obtain the impulse response corresponding to this |H(f)|
• What do you think the input to that function be like?
• Can’t be continuous
• Only need to specify one half - positive frequency part of |H(f)|
• Where does the stopband end?

• We assume that the stopband ends at fs/2

• Piecewise linear approximation

• How can we represent this?
• A sequence of (x, y) points
A recipe for filtering
(0,1) (fp ,1)
|H(f)|
(fstop,0)

fp fstop f
A piecewise linear approximation is used in Matlab

• Piecewise linear approximation

• How can we represent this?
• A sequence of (x, y) points (0, 1), (fp,1), (fstop, 0)
• Also a point at (fs/2, 0)

• Matlab also uses a normalization by fs/2

• So represented as a sequence of points (0, 1), (2fp/fs , 1), (2fstop/fs, 0), (1, 0)

• But also grouped as set of frequency and amplitude points

• Frequency points: (0, 2fp/fs , 2fstop/fs, 1)
• Corresponding amplitudes: (1, 1, 0, 0)
A recipe for filtering
(0,1) (fp ,1)
|H(f)|
(fstop,0)

fp fstop f
A piecewise linear approximation is used in Matlab

• Frequency points: (0, 2fp/fs , 2fstop/fs, 1)

• Corresponding amplitudes: (1, 1, 0, 0)

• Matlab provides multiple functions for filter design

• We use a function called firpm

• firpm produces an output which is a sampled impulse response
• What is the length of this impulse response - this is an input
• Other inputs are the frequency points and corresponding amplitudes

• Example firpm(200, [0, 0.2, 0.4, 1], [1, 1, 0, 0])

Other filters?

• What do you think this will do?

• firpm(200, [0, 0.2, 0.4, 1], [0, 0, 1, 1])

• What do you think this will do?

• firpm(200, [0, 0.1, 0.2, 0.4, 0.5, 1], [0, 0, 1, 1, 0, 0])
An exercise

• Consider a signal x(t) that consists of two cosines - one at a frequency of 10 Hz and the other at 50 Hz.
• Design a filter to filter out the 50 Hz cosine.
• Simulate this system in Matlab.