ECE 141 Lecture 6:

Digital Channel Model

Probability Theory and Random Processes, Additive White Gaussian Noise,Vector Channel

ECE 141: DIGITAL COMMUNICATIONS 1

The Channel
𝑠𝑚 𝑡 Channel 𝑟𝑚 𝑡

❑ Signals in communication must pass through an environment

to deliver a message from source to destination.
❑ This is called the channel.
❑ A signal, 𝑠𝑚 𝑡 , passes through a channel and is corrupted due
to the channel conditions and becomes a received signal,
𝑟𝑚 𝑡 .

ECE 141: DIGITAL COMMUNICATION I 2

Signal Corruption in Human
Communication
*noise*

I love you
I lab you I lob you I love you

ECE 141: DIGITAL COMMUNICATION I 3

Additive Noise Channel
Channel

𝑠𝑚 𝑡 𝑟 𝑡

𝑛 𝑡

❑ An external random process, 𝑛 𝑡 , corrupts the transmitted

signal.
❑ Possible sources: random movement of electrons (thermal
noise), other transmitters, cosmic noise, noise added by
transistors (shot noise), etc.
❑ The relationship between 𝑟𝑚 𝑡 and 𝑠𝑚 𝑡 :
𝑟 𝑡 = 𝑠𝑚 𝑡 + 𝑛 𝑡

ECE 141: DIGITAL COMMUNICATION I 4

 Random Variable
❑ Probability Space:
(, Pr )

Sample Probability
Space Measure
❑ Random Variable: X :  →
❑ Distribution:
FX ( x) → Pr { X  x}
Cumulative
Distribution Function

❑ Roughly speaking, we have:

(, Pr ) ⎯⎯
X
→( , Fx )

ECE 141: DIGITAL COMMUNICATIONS 5

Gaussian Random Variable
❑ Gaussian RV: X ~ N (  ,  2 ) (can be completely described by its
mean and variance)
❑ Gaussian PDF:  1 − x− 2 
f X ( x) FX ( x) = exp  
 
x 2  2
2 2

❑ Definition of Jointly Gaussian:
Z1 , Z 2 ,..., Z n  is J .G.  m  n, W1 ,W2 ,...,Wm i.i.d . N (0,1)
Constant vector b  n

n m
Constant matrix a 

 Z1   W1 
Jointly Gaussian Z =   = aW + b where W =  
Random Vector
 Z n  Wm 

ECE 141: DIGITAL COMMUNICATIONS 6

Gaussian Random Variable
❑ Jointly Gaussian Random Vector:
Z ~ N (μ, k )

𝐸 𝑍1 𝑉𝑎𝑟 𝑍1 𝐶𝑜𝑣 𝑍1 , 𝑍2 ⋯ 𝐶𝑜𝑣 𝑍1 , 𝑍𝑛

𝐸 𝑍2 𝐶𝑜𝑣 𝑍2 , 𝑍1 𝑉𝑎𝑟 𝑍2 ⋯ 𝐶𝑜𝑣 𝑍2 , 𝑍𝑛
𝝁=𝐸 𝒁 = 𝑘 = 𝐸 𝒁−𝝁 𝒁−𝝁 𝑇
=
⋮ ⋮ ⋮ ⋱ ⋮
𝐸 𝑍𝑛 𝐶𝑜𝑣 𝑍𝑛 , 𝑍1 𝐶𝑜𝑣 𝑍𝑛 , 𝑍2 ⋯ 𝑉𝑎𝑟 𝑍𝑛

❑ PDF of Jointly Gaussian Random Vector:

1  1 T 
f Z (z ) = exp  − ( z − μ ) k −1 ( z − μ ) 
 2 
n
( 2 ) 2 det(k )

❑ linear combinations of jointly Gaussian random variables are

also jointly Gaussian

ECE 141: DIGITAL COMMUNICATIONS 7

Random Processes
❑ time-indexed collection of R.V. or sample function drawn from an
ensemble
V1 V2

v1(t)

v2(t) Experiment
…

vN(t)

ensemble

❑ Since waveform is not exactly known, we cannot compute the

signals’ Fourier Transform
❑ Stochastic signals, though unpredictable, can be analyzed using their
statistical properties (mean, autocorrelation, etc.)

ECE 141: DIGITAL COMMUNICATIONS 8

Time Average vs. Ensemble
Average
Ensemble average Time average
T /2
1
E[v(t k )] =
1 N

v j (t k )
 v(t ) = lim
T → T
v
−T / 2
j (t )dt
N j =1
for any sample function v j (t )
tk

v j (t )

Process is ERGODIC if ensemble average = time

average

ECE 141: DIGITAL COMMUNICATIONS 9

Moments of Random
Processes
❑ First Moment:  X (t ) = E[ X (t )]

❑ Second Moment:
K X (t1 , t2 ) = Cov ( X (t1 ), X (t2 ) )
Auto-covariance function
= E ( X (t1 ) −  X (t1 ) )( X (t2 ) −  X (t2 ) ) 

RX (t1 , t2 ) = E  X (t1 ) X (t2 )  Auto-correlation function

❑ Auto-correlation function is the measure of relatedness of two

random processes
❑ SPECIAL CASE: If t1 and t2 are equal, then the auto-
correlation function is the mean square value.

ECE 141: DIGITAL COMMUNICATIONS 10

Stationary Processes
❑ Stationarity is statistical invariance to a shift of the time
origin
❑ Formal Definition:
A continuous-time process { X (t ), t  } is stationary in the strict-
sense if, for all t1 , t2 , …, tR   and ∆  , the joint CDF of the time
instances of X(t) satisfies
FX ( t1 ), X ( t 2 )..., X ( t R ) ( x1, x2,..., xR )
= FX ( t1 +  ), X ( t 2 +  )..., X ( t R +  ) ( x1, x2,..., xR )
for all real numbers x1, x2, …, xR

❑ For many applications, we do not require the process to be

strictly stationary

ECE 141: DIGITAL COMMUNICATIONS 11

Wide-sense Stationary
Process
❑ We only often require the following conditions in our analysis:
1. The mean of the random process should be constant with respect to time

E[ x(t )] =  x , t
2. The auto-correlation function of the random process only depends on the
time difference
Rx (t1 , t 2 ) = Rx ( ), t ,

❑ Such a process is called weak-sense/wide-sense stationary (WSS)

process
❑ If process is ergodic, then it is WSS BUT NOT THE OTHER WAY
AROUND!
❑ Important Property: Wiener-Khinchine Theorem

Rx ( ) ⎯→
F
Sx ( f )

ECE 141: DIGITAL COMMUNICATIONS 12

Filtering of WSS Random
Processes
❑ We are primarily interested in the following properties of a
random process after passing through an LTI filter:
▪ First and second moments
▪ Stationarity
▪ Gaussianity
❑ Stationarity/wide-sense stationarity is preserved under LTI
filtering
❑ Gaussianity is preserved under LTI filtering
❑ For a WSS process, the PSD of the filtered process is the PSD
of the original process times the squared magnitude response
of the LTI filter

ECE 141: DIGITAL COMMUNICATIONS 14

Filtering of WSS Random
Processes
WSS WSS

X (t ) h(t ) Y (t ) = ( X  h)(t )

❑ First Moment of output: Y = H (0)  X

❑ Second Moment of output: KY ( ) = (h  K X  hrv )( )

❑ PSD of output: SY ( f ) =| H ( f ) |2 S X ( f )

ECE 141: DIGITAL COMMUNICATIONS 15

Preserving Gaussianity of
Filtered Random Processes
Z (t ) h(t ) V (t ) = ( h  Z ) (t )
Gaussian Gaussian

g1 (t ) V1 = Z (t ), g1 (t )

.
. Jointly Gaussian
Z (t )
. Random Vector

g m (t ) Vm = Z (t ), g m (t )

Now let the filters 𝑔1 𝑡 to 𝑔𝑚 𝑡 be orthonormal signals, 𝜙1 𝑡 to 𝜙𝑀 𝑡 .

ECE 141: DIGITAL COMMUNICATIONS 16

Vector Representation of
Noise
❑ If the signals exist in 𝑁 dimensions, then the transmitted signal
can be expressed as:
𝑁

𝑠𝑚 𝑡 = ෍ 𝑎𝑚𝑘 𝜙𝑘 𝑡
𝑘=1
❑ By linearity of the signal space, it follows that the noise and
received signal can also be expressed in terms of the
orthonormal basis:
𝑁 𝑁

𝑟 𝑡 = ෍ 𝑏𝑘 𝜙𝑘 𝑡 𝑛 𝑡 = ෍ 𝑛𝑘 𝜙𝑘 𝑡
𝑘=1 𝑘=1

ECE 141: DIGITAL COMMUNICATION I 17

Additive Noise Channel in
Vector Form
❑ The signals as vectors:
𝑎𝑚1 𝑏1 𝑛1
𝑎𝑚2 𝑏 𝑛2
𝒔𝒎 = ⋮ 𝒓= 2 𝒏= ⋮
⋮
𝑎𝑚𝑁 𝑏𝑁 𝑛𝑁
❑ The relationship between the transmitted signal and received
signal can be expressed as a vector equation:
𝑟 𝑡 = 𝑠𝑚 𝑡 + 𝑛 𝑡 → 𝒓 = 𝒔𝒎 + 𝒏
❑ Note that the elements of 𝒏 are all independent and
identically distributed (IID) random variables.

ECE 141: DIGITAL COMMUNICATION I 18

Noise As a Random Process

NOISY
x(t ) y (t ) = x(t ) + Z (t )
CHANNEL

Some random process

NOISE AS AN ADDITIVE WHITE GAUSSIAN PROCESS

ECE 141: DIGITAL COMMUNICATIONS 19

Channel Noise
❑ Recall that the power of a random process is:
+∞

𝑃𝑛 = 𝑅𝑛𝑛 0 = 𝐸 න 𝑛 𝑡 𝑛∗ 𝑡 𝑑𝑡
−∞

❑ The term inside the expectation operator is the inner product

operation.
❑ In vector form, the noise power can be calculated:
𝑃𝑛 = 𝐸 𝒏 ∙ 𝒏 = 𝐸 𝒏 2

ECE 141: DIGITAL COMMUNICATION I 20

Additive White Gaussian
Noise (AWGN) Channel
❑ A Gaussian distribution for the noise is the most likely*
distribution of the noise that corrupts the channel.
❑ Gaussian distribution with power 0.5𝑁0 :
1 𝑛2
−
𝑝 𝑛 = 𝑒 𝑁0
𝜋𝑁0
❑ By breaking the noise into 𝑁 components, then the probability
of a Gaussian-distributed noise signal vector is:
𝑁
1 𝒏 2
− 𝑁
𝑝 𝒏 = 𝑒 0
𝜋𝑁0
❑ White Noise means that 𝑅𝑛𝑛 𝜏 ≠ 0 = 0.

*will be elaborated in the latter part of the course.

ECE 141: DIGITAL COMMUNICATION I 21
Vector Channel Model
Channel
𝒔𝒎 𝒓
𝑝 𝒓|𝒔𝒎

❑ In digital communications, the signal, 𝒔𝒎 , is selected from a set

of possible signal vectors 𝒔𝒎 , 1 ≤ 𝑚 ≤ 𝑀 , depending on the
symbol sent.
❑ The received signal, 𝒓, depends statistically on the transmitted
vector through conditional probability density functions
𝑝 𝒓 𝒔𝒎 .
❑ Note:
𝑀

𝑝 𝒓 = ෍ 𝑝 𝒓 𝒔𝒎 𝑝 𝒔𝒎
𝑖=1

ECE 141: DIGITAL COMMUNICATION I 22

Digital Signals under an
AWGN Channel
Channel
𝒔𝒎 𝒓
𝑝 𝒓|𝒔𝒎

❑ The received signal is:

𝒓 = 𝒔𝒎 + 𝒏
❑ In vector terms, the symbol 𝒔𝒎 is displaced due to the noise.
2 (t )

𝒏
𝒔𝒎
1 (t )

ECE 141: DIGITAL COMMUNICATION I 23

Digital Signals under an
AWGN Channel

ECE 141: DIGITAL COMMUNICATION I 24