Networks , Signals & Systems

Prasad Sir Hours

Office :

Mon
& Tru : 5 : 10 PM to 6 PM

A3-217
Vindhya
COR)
MS Teams to prasad .
message on

Krishna of iiit . ac .
in
1 .

Signals
Natural/Artificial variations that some
info
·

carry
.

OR

f) independent variables) =
Some value
of
dependent variable

Eg :. Seismo
signals Temp .
at 2 PM =
f (u ,
y, 2)
Optical ↑
signals pt . in
space
And so on .

brightness of f (r
·
Photos-i
b)
e
.

9,
, =
,

each
pixel
In system
·
a
,

u(t) > SYSTEM >

y(t)
BUT in
"y(t) is a
fu of u(t)"
YI*E]
that a at
NOTE
are entire
sigs y(t) is an
output t
x(t) is an entire .
signal

a) Ways to
classify signals
b complexe signals
O Real-valued . 3 Even and Odd
signals
·
Even : x(t) x( t)
ur
= -

- 4
# tER .

time complex ·

Odd : u(t) = -u( t) -

↓ tEIR
Continuous Time
G Discrete
2
Time

Continuous Time ⑦ periodic and Aperiodic

M
: The
signal
is
defined at

i every spot. Signals

Notation (t)
f >
: x
x(i +) + =
n(t) Periodic
Time
Discrete : The
signal is
defined at
·

Aperiodic
↑ /T
x(T + +)
specific time intervals· + x(T)

+ /111] Notation : 2 [n]

-
nth step -
for ALL tER
b) Different functions to
represent signals
② ②
Real Exponential signals Real Sinusoids

xe(t) = Ceat ,
a
,
CER x(t) =
Acos(wt 4) +
x

ce-at
=
u = 1 wo

>

x = ceat
~
cost

3
Complex Exponential Signals
4
Complex Sinusoids

z(t) = Ceat ,
a, 24 z(t) Ce)(wt
=
+
4)

Suppose C = /C ej0 Split into Re(2) & Fm(2) .

at D Re(z) =
cos (wt 4) +
Then 2 (t) = Jejorat ,

CERR [m(z) = sin

(wt + 4)

·
real rt
Take the part.
M -

Ce coscut

· U
eatcos(wot
⑧

Re(z) = (c) +
4)

E Basically its , a cos fr ·

with

amplitude changing with time .

c) Some quantities associated with

imp .

signals
① Total
Energy ② Average Power

EguStd =
OR

EEPrLH
that to < C then
Note
if
·

,
.

Ad
it is called a
finite energy
.
power signal
the
The
magnitude of
HW do
signal How
periodic
:
is taken to avoid nonzero

getting
a
complex value
signals have so
energy
?

for energy.
NOTE :
If Eg =
finite, ③ Power
of A Periodic
Signal
Po = 0

Plim It is
.

Why ?
We know
+

0 /ke(t)ldt] Es
Now se(t) is periodic
if so selt)?
is
,
-
T ,

dividing by 2T
, n(t kT)) + (u(t)) = -k+ z

So
for periodic signal ult)
Po
<

And in limits terms

,
Po =
linIt T
Po =
lin Eco

T- 8 2T

=
8

d) Special signals
O Unit Impulse/Impulse signal ② Unit
Step Function

joc)du
y(t) =

& J , Xt0 n(t) =

- 8

n(t)
1
Defined : =

So
! O(t)=lim ut
M

Area
So it's like
extremely
an I
= 1

7
small jerk lasting for ↓
O
&
t O sec
-

NOTE

A more
formal definition :
Do NOT write

ggotid
t
5(t) f + Fo n(t) =
= 0
fo(t)dt -

↓ D
-

Dirac-Delta
-. and we
represent it
using It is as nonsense as

arrow .
[8(2)d2
an
Function
N
This represents u(2) =
11 AREA
-
8
e) Transforming Signals
② Time Shifts
Suppose : 1 1 1

& ↑
S & S & 7

u(t) 2(t -

to) x(t +
to)
Forward
Shift Backward
Shift

② Amplitude Modification
1 1 1

& ↑
S & ↑
S & ↑
S

u(t)
du(t) < > du(t) <
, ,

③
Scaling
Time

1 1 1

&
f

S &
f

S & S

x(t) u(2t) u( + (2)

f) Some Examples of Systems

NOTE
t

u(t) = f8(z)di
↓ a

y(t)
-

x(t) >
System <
value at t

Eg Whatever transformations
uIE2)
: we dat
discussed above , i e
. :

Time Shift Time

scaling
> >
-

signal
-

Entire
>
Amp
-

modulation
 NOTE : You move forward with time . So
you
x(t to) Time
delay the signal at + +
=
=
-

see ,

(t + to) Time advance

u But for x(t-to) signal shifts
the
=

ahead observe the

why ? M
to . So
you signal
at t =
+ + t
,

t,
Hence it is a time delay
T
t= p
> ,
.
YOU

Other
Systems :

① Differentiator ② Integrator/Accumulator

gy()Schid
r(t)
u(t), du ,

the rate
Captures of change Accumulates the output values
of the
system
.
al
of a
goi until time to

NOTE :
Smoothening
t+ 5
a
signal/Moving average
Then how do we still get a

continuous
graph of y(t) will graph ?
(u(z)dz
Let The

Instead intervals like

+ -
0 now look like a smoothened of taking
y(t) =

of x(t) since
this
(
...
25 version

·
we
,
I Es
1 ,

basically replacing
H
are
( Where (t
M
it 5 ++ o]
-

small intervals values with

of
their average
. > 7

Take them like this .

9) Some important classes

of systems
& LTIs /Linear Time -

Invariant Systems)
They satisfy the
property of
of additivity
.
Eg :
(t)
ultiS
n (t) >
-
>
- y,
if it's
,
So an
+ 7 3 t

LTI
n
,
(t) > >
-

Yz(t) ,
③
Scaling
Eg x(t) >
-
>
- y(t) then a x(t) >
-
- a .
y(t)

② Time Invariance

If
n (t) >
-
- y(t) then ult- to) >
-

y(t to)

It is called linear because it

follows additivity and
scaling .
② Causal Systems
If y(2) depends on u (2) # It

BUT NOT On 2St

3 Stable Systems
If ly(t)) < const 1Et
.
wherever li(t)/ < const . 2 Et.

is stable.
Then system

y(t) =
e
+
x2(t) X
NOTE that
causality does NOT

Prove that this is causal .

depend on u(t) . We
always measure

u(t) for It causality only with

reference to

y(t) depends on
Sol : ,
wherever the input .
starts

is causal.
HENCE the system
 Harmonics
.
2
of A Sinusoid

·
instit =
caswotjeint
(Both are
periodic .

)
to
1) frequency of the above
sinusoid)
so fro of the form eikwot is called the 1kIth harmonic
of a just.
any
(kez 50]) -

·
Now let
,

u(H) =
[Geot (where any
! Depends
K
"Our
on

biryani"
n(t) is
Obviously ,
periodic .

But
any periodic signal
can the reverse is true ? I e
.

can be
we
say ,

written sinusoids ?
as a sum
of
Yes !

I I
But with careats :

u(t) must have finite energy in time-period

any
·
n(t) must be continuous

a) Fourier Series

Let se(t) have

fundamental angular frequency w

=
Then n(t) can be
expressed as

u(t) =
[aejkwt fe some an e D ke2 .

For some w let

,

eikwat lenght
=2 an
e
time interval
u(t) ,
where T is some
of
 eik-kw

I
ejk'utI

S
t 2π
u (t) a Think about this
integral .
Since time period =
S

(k k')w
-

= K'
Integrating on both sides
, ① if K
,
we will
integrate over (K- K) time

T T
int j(k k') rot periods .
(u(t)e J]axe
-

= dt
0
8
=>
Integral is .

= Jane
j(k k') wt
k'

(
-

② if =
k, a
=
.
j(k-kat
O

&t =>
Integral becomes T .

Jeilkkotdt [k
Since =
=
Eg :
x(t) = sin" Soot

k = k
Write it
using FS
coeffs .

=
a
.
T
(since for every other k
,
integral is
&
.
0

Sol : sin' Soot = 1- cas1000t

2
-

Hence

I fult) eikot
,

- (ecot 1000
an =
de +
=

2 2
< TX

this coeff of any specific

the
Using
= eviot-ejo
,

term can be
found .

Assume (t) is real .

u (t) =

[a
ent (FS of ul ·
Notice how
a
and
a are

(conjugate x() conjugates

Fatekut
(t) of
.

u =

ofn
*
The
coeff
k -

conjugate of [K) eff

uTt) =

[ at
eiwt (Es of ul
) ·

If
on

u
u

*
(t).
(+ ) =
x(t) (ieu(t) =R]
a
=
k
(Replacing with
-k')
Konjugate Symmetry of FS coeft)

that also be used

signals .
Note Fourier series to reconstruct
·

can

In

eskt an depends K r (t)

Za
on w
u(t) = , ,

eskut depends on K
,
.
w
 FS
If selt) is known ,
we can find its FS
representation E>
coefficients
the xe(t)- all
as would
Now someone need not transmit entire
signal transmitting
allow the receiver to reconstruct x(t) themselves

numbers ?
Note :
If u(t) is real
, why is it written as a sum
of complex
elk lear
(i)
Let
u(t)
= Zar an tan"
=

%=
=
9
+
[ (a eikut + ae-kuy
Then

Xaeikut ae-k
=
a +

[2Re((a /ei(Pk kwt)

+ +
u(t) =
a +
~

(ax/cos(d knt)
Z
Z
+ 2
=
a +

Since 2 +z Re(2)
2
,
=

Which is real
9 +
[2re(an ejkwt)
=

So from what we've done so

far ,
the
diff. forms of an FS are :

jkwt

=2
·
u(t) ae

=a
·
u(t)

u(t) =
a +
2[(a(cs(kwot +
du)

u(t) =
Go +
2 Boskut- Geinkot (if an
=
Br +
j()
6) Properties of FS
expansions
②
Linearity
A (t) B(t) both
If g have same
angular frequency we

A(t) =
[a eskwt Z
,
,
Then
if < (t) =
Ea eikut
B(t) =
[beikut
Then
( =
an + br Ke

And let a C (t) = Alt) + B(t) ·

② Multiplication
A(t) =
[aeskwt Z
,
,
Then
if < (t) =
Ea eikut
B(t) =
[beikut

Y
And let 2 (t) =
A(t) B(t) -

Canewt) (aneimet)
=
So in
= bun

Which means ((t) becomes

C(t) =
(abek) einwt

③ Convolution Consider an
example :

We want to out how

find much smoke

= Zebr was emitted

from fireworks over time T .

S(t) = smoke
from a
firework
This series is called at time+
after ignition
kind
of a a CONVOLUTION
.

(t) of fireworks at
ignited
F
Let u(t) y(t)
= no
have
period
.

a T.
,
time t
Then
if
2(t) =
(x(t)y(t -
+)di Now to find total smoke at a time :

T 1t =
0

Y(0) F(0) S(0)

(t)
r(t)y(t)
=

called periodic
·

2 is a convolution of ·

2 t=
(t) Also has
2
period T .

jut Y (1) =
F(1)S(0) + F(0)S(1)
z(t) =
Ice
...
and so .
on So :

Then Cr = Tarbk Y(t)

=[
F(t) S(t-

continuous ,
Or
ifIn .
is

2(t) =

[x(t)y(t -
+ )di
c) Laplace Transform
Recall the FS
of a
periodic signal .
The K* FS
coeff (which multiplies w/ eskwt)

= (x(t)e jkwt
-

a
v It
 For
signal (not necessarily periodic)
·

any ,
we
define :

x (b) = gu(t)e+ dt
-

X(s) Laplace x(t)

The value
of is the
Transform of at
complexe frequencys ,
where

3 5 +
ju
=

The
Laplace Transform basically converts
ult) (from t -domain) to x(0) (to s-domain)
We
define a
'Region of Convergence' for s
,
which make the
integral a finite value .

convolutions
Note : The
sifting property of Now
if you integrate
Let's assume we take both ,

y(t) =
x(t)J(t to) -

Scaling Impulse
Jy(t) (u(t)o(t =
-
+ )di
T
Const .

=(x(t)o(t
Then
,
(2(t) -

+)di

245(0)
+ +
y(t) = to
x(t)
(0(t t)di
= -

+ =
to
,

Notice how
y(t) and (t) are
x(to) O(t-to)
2
Now take 2 (t) =

the thing
basically
same
.
Then
toto
S i to)5(o)
which means :

z(t) = 0

,
t =
to

(x(t)y(t -

t)di =

x(t))y(
+ -
t)dt

You should see now that

Laplace Transform is
basically a convolution too.

Let :

(j(t
e+
0)e
-

X (g) =
-

dt

T
I
2

( Applying easenty)
e(Ult-o) te
 - a+

Eg :
e u(t) (a(r) BUT s = 5 +
jw .

Apply LT on this
. So

I East I
it
-

x (s) =
2

x (s)
= Tu(t)e et
e

Now we can compare &

Ea .

-Get
-
(r + a)
If U +a 70
, e
converges .

C + a) +
x(s) =
sa
=(e
-

If r +a <0 -

(r + a) + >0 .

Ie"I
(b a)t
+
Which
for
-

:
means

t >
-
0
(s + a)
-
,

-
(5 + a)t
2 >
-
D

So X/s) becomes unbounded .

the ROC
Finding
:

pres
De Rosand be

diagram

6) Properties of Laplace Transform

1 Convolution in time domain
Note that A is the

LT(x(t) =
y(t)) =
x(s)Y(g)
convolution
symbol.
if u(t) +
X(e) has Roc = R
,
(t 4)
y(t) y(s) ROC Rz
+
has =

The ROC
of the combination contains R
,
1R2
.
 Linearity
2

x(t) - x(9)
y(t) -
Y(s)
Then <x(t) +
By(t)c(X(s) BY(s) +

3 Derivative

u(t) = X (s) Es Xl
If ,
thendult)
.

2) Inverse Laplace Transform

following formula
We use the :

where is Re(s) FgE ROC of X (g)

any
o

r + j) + c)
St

x(t) =

+j
2
(x(s)e de

S = r +
j) b)
-

f) Fourier Transform
Let u(t) be .
any signal
The Fourier Transform of ult) is a
signal in the w-domain (orf-domain)
* (w) =
FT(u(t)]
=
x(jw)
=
LT(x(s)) (ats jw) =

-Gult)e -stat at s =
ju .

-
D

= gult) ejut

S
Note : LTI and convolution

Let h(t) be the output to an input signal being O(t).

Now if the input signal were instead x(t),

y(t) =
x(t) = n(t)

But
why ? a(t) 1

Proof :

established that
Since
already
it is :

x(t)f(t T) -

5(t) > h(t)

Then c .

J(t) > c .

h(t) [LTI ] ·

functions
Now

a scaled
let us

5(t)
represent
at every
n(t)
t
as "slices"

for example ,
of
-O
↑
a slice at t : I could be : =

x(t) j(t t) - -

- 2034

Now
,
output is :
x(t)

x(t)b(t -
t) - x(t)h(t T) -

all
Integrating
T
over ,

just) O(t--idTu(t)h( + t)di

-

-
8

= x(t) * h(t) .

Note that this ONLY

applies to LTI
systems cause otherwise we can't
say for
sure

that a sum
of inputs guarantees a sum
of their
respective outputs
.

~
Response of LTI
systems to
exponential sinusoids(est)
Output of the
system is

St

y(t) =
h(t) = e

The
s(t T)
-

est)ihte)
 =
H(s] est

Where H(s) is the LT

of h(t) (at specific s
)
.

St
the scale
If you notice
,
whole operation does
nothing but e
;

est > LTI >

H(s) est
This correlates back to a
property of vectors called
eigenfunctions .

Eigenvectors
Av X true where
If = is
,
:

A =
Transformation
X = Some scalar

An e-domain view
of LTI systems
Poles and Zeroes of LT
Output
u (t) - x(t) = h(t)

Pole():Wherehegoesa
o

LT
LT

x(e)
v

Output
(s) S
>
X(s)H(s)
Eg X =
:

s +a

S = 0 is a zero .

s = - a is .
pole
a

2
Important LT pairs
<T
·
J(t) > 1
,
ROC = K

·
a ER
,
at

al
-

u(t))
(e
I
2 ROC =
D , Rele)) -

3 +a

e -at u) t)c
I

&st al
3
- -

Roc = K
, Rels]< -

S+ a

Eg : x (g) = Re(s)) -
1

(e + 1)(b + 2) Note that the ROC

of X
(e)
I
=
1
-
contains the ROC
of both
S + 18 +2

↓
transforms .

↓
Roc is ROC is

f) -
1 37 -
2
3 LTI
systems described
by linear
differential eq . s .

1) Explicit

inputa (t)
For
arbitary
·
an
,

y(t) =
(BlaBla)x(t)
·
Via Impulse Response :

By putting x (t) = J(t) ,

we
get n(t).

Since
y(t) = x(t) = h(t)

H(s) with ROC .

Then we can use 2 methods to find response to se(t).

>
Find h(t) then find a (t) * h(t)
-

>
-

Find X(9) then find Y(s) = x(9) H(s)

,

Then find ILT( Y (s)) =

y(t).

B) Implicit
We can use
differential equations.

Eg
: y(t) +
2y(t) =
x(t)v0y(e) =

g
+ 2
X(f)

Sol
Apply
= siz
: LT on both sides. so H(s)
=
u

=> sY(f) + 2y(s) =

x(s)
=> Y(f)(f 2) + =
x(s) From
given ROC and that H(s) =

et2
+u(t)
- *
ILT <
n(t) =
e