Guassian Random Process and

White Noise
Lecture 15
EEE 352 Analog Communication Systems
Mansoor Khan
Electrical Engineering Dept.
CIIT Islamabad Campus
Gaussian Processes
• Gaussian process is a special random process
– Important for communications because noise is usually modeled
as Gaussian process
• Definition
– A random process X(t) is Gaussian process if at any time instant
t
i
, the random variable X(t
i
) has Gaussian distribution (is
Gaussian random variable)
• A simple property
– when a Gaussian process passes a linear filter, the output is still
a Gaussian process
X(t)
t
t
i
X(t
i
)
• Gaussian process & Gaussian random variable
– Remind: if X(t) is Gaussian process, then at each time instant t
i
,
X(t
i
) is Gaussian random variable
– Gaussian random variable X(t
i
) described by a PDF with mean μ
and variance σ
2
• Gaussian process & Gaussian random variable
– Remind: if X(t) is Gaussian process, then at each time instant t
i
,
X(t
i
) is Gaussian random variable
– Gaussian random variable X(t
i
) described by a PDF with mean μ
and variance σ
2
– Gaussian random process X(t) described by mean function μ(t)
and power spectral density (PSD) S
X
(f)
• or use autocorrelation function R
X
(τ) in place of PSD
– Main property #1: for stationary random process
– Main property #2: power of random process equals to
power of random variable
– Main property #3: If random process has zero mean,
then
mean function of random process is contant,
and equals to mean of random variable,
( ) t u u =
2
( )
X
S f df o
·
÷·
=
í
| |
( ) ( ) ( )
X i i
S f df E X t X t
·
÷·
=
í
• Example
– Let X(t) be zero-mean stationary Gaussian process with PSD
S
X
(f). Determine PDF of X(3).
5, for 500 500 Hz
( )
0, else
X
f
S f
÷ s s
¦
=
´
¹
2
10000
Probability density function of random variable (3) :
1
( )
10000
x
X
f x e
t
÷
=
500
2
500
Mean of random variable (3) : ( ) [ ( )] 0
variance of random variable (3) :
( ) 5 5000
X X
X
X t E X t
X
S f df df
u u
o
·
÷· ÷
= = =
= = =
í í
White Processes
• White process is a special random process with
constant power spectral density
– Power density is a constant for all frequencies
– White process is an ideal math model because the total power is
infinite, analogy to impulse function δ(t).
– But it is useful to model noise, and usually gives very good
approximation and extremely simple math.
( ) , ( ) ( )
W W
S f C R C t o t = =
Noise
• An important type of noise is thermal noise
– The origin is random movement of electrons
– This noise signal can be modeled as Gaussian
process
• The composite effect of a large number of small electrons
becomes Gaussian distributed
– We also model the noise signal by white process
• Without any special filtering, noise power will be almost
constant for all frequencies
– The noise signal is usually added to information
signals, so we call it Additive White Gaussian Noise
(AWGN)
• Model of AWGN (additive white Gaussian noise)
X(t): random or deterministic signal
W(t): AWGN
Y(t)=X(t)+W(t): random process
A realization of
AWGN W(t)
• AWGN W(t) is a white Gaussian random process
– “White”: constant PSD
– “Gaussian”: at time t
i
, random variable W(t
i
) has Gaussian
distribution, usually assumed with zero mean.
0 0
( ) , ( ) ( )
2 2
W W
N N
S f R t o t = =
2
2
2
1
( )
2
x
f x e
o
t o
÷
=
Filtered Noise Processes
• AWGN is an ideal noise model, we often use it
to describe the received noise
– Practical noise can usually be modeled by filtering AWGN, i.e.,
processing noise via filters
• Major difference between AWGN and filtered
noise
– PSD different: filtered noise has limited bandwidth B, and PSD: .
– Distribution of random variable same: still Gaussian random
variable
2
( ) | ( ) | ( )
N W
S f H f S f =
• Example
– If AWGN pass an ideal lowpass filter with bandwidth B and
magnitude 1, what is the output signal N(t)? Determine the PDF
of N(0).
1, for
( )
0, otherwise
B f B
H f
÷ s s
¦
=
´
¹
2 2
0
0
( ) is still Gaussian process, with PSD
( ) | ( ) | ( ) | ( ) |
2
, for

2
0, otherwise
N W
N t
N
S f H f S f H f
N
B f B
= =
¦
÷ s s
¦
=
´
¦
¹
2
0
2
0
0
2
0
(0) is Gaussian random variable with zero-mean, variance
( )
2
1
Its PDF is: ( )
2
B
N N
B
x
N B
N
N
S f df df N B
f x e
N B
o
t
·
÷· ÷
÷
= = =
=
í í
• Random process & random variable
– When an AWGN signal passes a filter with bandwidth B and unit
passband magnitude
• If AWGN has zero mean, then all the random processes and random
variables are zero mean.
• PSD and variance relationship
Random process
PSD S(f)
Random variable
variance σ
2
Input W(t)
Output N(t)
0
( )
2
W
N
S f =
0
, for in passband
( )
2
0, otherwise
N
N
f
S f
¦
¦
=
´
¦
¹
2
0 N
N B o =