PRESENTATION

TITLE
Stochastic Proceses
STOCHASTIC PROCESSES

• Text
IV. Stochastic processes
• Definitions
• Stationary processes
• Spectral density function
• Ergodicity of stochastic processes
• Stochastic Data Processing Systems
 IV. Stochastic processes
4.1. Definitions
X (t, x )
• Stochastic processes
• Let x denote the random !
X (t, x n )
outcome of an experiment.
To every such outcome X (t, x k ) !
suppose a waveform X(t, x)
is assigned. X (t, x 2 ) !
• The collection of such X (t, x1 )
waveforms form a t
0 t1 t2
stochastic process.
IV. Stochastic processes
4.1. Definitions
• The set of {xk} and the time index t can be continuous
or discrete (countably infinite or finite) as well.
• For fixed xi ÎS (the set of all experimental outcomes),
X(t, xi) is a specific time function.
• For fixed t, X1=X(t1, xi) is a random variable. The
ensemble of all such realizations X(t, x) over time
represents the stochastic process X(t).
IV. Stochastic processes
4.1. Definitions
• Example
• X(t)=Acos(w0t+j)
• Where j is a uniformly distributed random variable in (0, 2p)
represents a stochastic process.
• Some stochastic processes
• Brownian motion,
• Stock market fluctuations,
• Various queuing systems.
IV. Stochastic processes
4.1. Definitions
• If X(t) is a stochastic process, then for fixed t, X(t)
represents a random variable.
• Its distribution function is given by

FX ( x, t ) = P{ X (t ) £ x}
• Notice that Fx(x, t) depends on t, since for a different t, we
obtain a different random variable.

D dFX ( x, t )
f X ( x, t ) =
dx
• Derivative of Fx(x, t) represents the first-order probability
density function of the process X(t).
IV. Stochastic processes
4.1. Definitions
• For t = t1 and t = t2,
• X(t) represents two different random variables X1 = X(t1) and
X2 = X(t2) respectively.
• Their joint distribution is given by

FX ( x1 , x2 , t1 , t2 ) = P{ X (t1 ) £ x1 , X (t2 ) £ x2 }
• Their joint density function is:

D ¶ 2
FX ( x1 , x2 , t1 , t2 )
f X ( x1 , x2 , t1 , t2 ) =
¶x1 ¶x2
• And represents the second-order density function of the
process X(t).
IV. Stochastic processes
4.1. Definitions
• Similarly fx(x1, ..., xn, t1, ..., tn) represents the nth order
density function of the process X(t).
• Complete specification of the stochastic process X(t)
requires the knowledge of fx(x1, ..., xn, t1, ..., tn) for all ti, i
= 1, ..., n and for all n. (an almost impossible task in
reality).
IV. Stochastic processes
4.1. Definitions
• Characterisctics of stochastic processes
• Mean of a stochastic process
+¥
µ (t ) = E{ X (t )} = ò -¥ x f ( x, t )dx
D
X

• µ(t) represents the mean value of a process X(t). In general, the

mean of a process can depend on the time index t.
• Autocorrelation function of a process X(t) is defined as
D
RXX (t1 , t2 ) = E{ X (t1 ) X (t2 )} = ò ò x1 x2 f X ( x1 , x2 , t1 , t2 )dx1dx2
* *

• It represents the interrelationship between the random variables X1 =

X(t1) and X2 = X(t2) generated from the process X(t).
IV. Stochastic processes
4.1. Definitions
• Properties of autocorrelation function
1. RXX (t1 , t 2 ) = RXX* (t 2 , t1 ) = [ E{ X (t 2 ) X * (t1 )}]*
2. RXX (t , t ) = E{| X (t ) |2 } ³ 0.
(Average instantaneous power)
• 3. Rxx(t1, t2) represents a nonnegative definite function, i.e., for
any set of constants {ai}ni=1
n n
åå i j RXX (ti , t j ) ³ 0.
a a *

i =1 j =1
• This equation follows by noticing that: E{|Y|2}³0 and
n
Y = å ai X (ti ).
i =1
IV. Stochastic processes
4.1. Definitions
• The autocovariance function of the process X(t).

C XX (t1 , t 2 ) = RXX (t1 , t 2 ) - µ X (t1 ) µ X (t 2 )

*

• Examples
• Given T
z = ò -T X (t )dt.
T T
E [| z | ] = ò -T ò -T E{ X (t1 ) X * (t2 )}dt1dt2
2

T T
= ò -T ò -T R XX (t1 , t2 )dt1dt2
IV. Stochastic processes
4.1. Definitions
• Consider process X(t)
X (t ) = a cos(w 0 t + j ), j ~ U (0,2p ).
µ (t ) = E{ X (t )} = aE{cos(w 0 t + j )}
X

= a cos w 0 t E{cosj } - a sin w 0 t E{sin j } = 0,

2p
since E{cosj } = 1 cosj dj = 0 = E{sin j }.
2p ò0
R XX (t1 , t2 ) = a 2 E{cos(w 0 t1 + j ) cos(w 0 t2 + j )}
a2
= E{cosw 0 (t1 - t2 ) + cos(w 0 (t1 + t2 ) + 2j )}
2
a2
= cos w 0 (t1 - t2 ).
2
IV. Stochastic processes
4.2. Stationary processes
• Strict sense and wide sense stationarity
• Stationarity
• Stationary processes exhibit statistical properties that are
invariant to shift in the time index.
• Thus, for example, second-order stationarity implies that
the statistical properties of the pairs {X(t1) , X(t2) } and
{X(t1+c) , X(t2+c)} are the same for any c.
• Similarly first-order stationarity implies that the statistical
properties of X(ti) and X(ti+c) are the same for any c.
 IV. Stochastic processes
4.2. Stationary processes
• Strict sense stationarity
• In strict terms, the statistical properties of a stochastic process are
governed by the joint probability density function.
• A process is nth-order Strict-Sense Stationary (S.S.S) if
f X ( x1 , x2 ,! xn , t1 , t2 !, tn ) º f X ( x1 , x2 ,! xn , t1 + c, t2 + c !, tn + c )
for any c, where the left side represents the joint density function of
the random variables X1=X(t1), ..., Xn=X(tn) and the right side
corresponds to the joint density function of the random variables
X’1=X(t1+c), ..., X’n=X(tn+c).
• A process X(t) is said to be strict-sense stationary if equation
above is true for all ti, i=1, ..., n; n=1, 2, ... and any c.
IV. Stochastic processes
4.2. Stationary processes
• First order strict sense stationary process
• For any c
f X ( x, t ) º f X ( x, t + c )
• In particular, if c = -t, then
f X ( x, t ) = f X ( x )
• That means, the first-order density of X(t) is
independent of t. In that case
+¥
E [ X (t )] = ò -¥ x f ( x )dx = µ , a constant. (1)
IV. Stochastic processes
4.2. Stationary processes
• Second order strict sense stochastic process
• From definition, we have:

f X ( x1 , x2 , t1 , t 2 ) º f X ( x1 , x2 , t1 + c, t 2 + c)
• for any c. If c is choosen so as c = -t2, we get
f X ( x1 , x2 , t1 , t 2 ) º f X ( x1 , x2 , t1 - t 2 )
• The second order density function of a strict sense
stationary process depends only on the difference of
the time indices t1-t2=t.
IV. Stochastic processes
4.2. Stationary processes
• The autocorrelation function is given by
RXX (t1 , t2 ) =D E{ X (t1 ) X * (t2 )}
= ò ò x1 x2* f X ( x1 , x2 ,t = t1 - t2 )dx1dx2
= RXX (t1 - t2 ) D= RXX (t ) = RXX* ( -t ),
(2)
• The autocorrelation function of a second order strict-
sense stationary process depends only on the
difference of the time indices t2-t1=t
 IV. Stochastic processes
4.2. Stationary processes
• Wide sense stationarity
• A process X(t) is said to be Wide-Sense Stationary if:
E{ X (t )} = µ and E{ X (t ) X * (t )} = R (t - t
1 2 XX 1 2 ),
• Since these equations follow from (1) and (2), strict-sense
stationarity always implies wide-sense stationarity.
• In general, the converse is not true
• Exception: the Gaussian process (normal process).
• This follows, since if X(t) is a Gaussian process, then by definition
X1=X(t1), ..., Xn=X(tn) are jointly Gaussian random variables for any
t1,..., tn whose joint characteristic function is given by
n n
j å µ ( tk )wk - åå C XX
( ti , tk )wi w k / 2
f (w1 , w 2 ,!, w n ) = e
X
k =1 l ,k
 IV. Stochastic processes
4.2. Stationary processes
• Examples
• The process:
X (t ) = a cos(w 0 t + j ), j ~ U (0,2p ).
• This process is wide sense stationary but not
t2 strict sense stationary.
• If the process X(t) has zero mean,
T
then s2z is reduced to: z= ò -T
t = t1 -t 2
X (t )dt.
-T T t T T
t1
t -T s z2 = E{| z |2 } = ò -T ò -T R (t1 - t2 )dt1dt2 .
XX

2T - t
• As t1 and t2 varies from –T to T, so t=t2-t1
varies from -2T to 2T.
• Rxx(t) is a constant over the shaded region in
the figure on the left.
 IV. Stochastic processes
4.3. Power spectrum
• Power spectrum
• For a deterministic signal x(t)
• The spectrum is well defined: If X(w) represents its Fourier
transform, +¥ - jw t
X (w ) = ò -¥ x(t )e dt ,
• then |X(w)|2 represents its energy spectrum. This follows from
Parseval’s theorem since the signal energy is given by
+¥ +¥
1 w d w = E.
ò -¥ x (t )dt = ò -¥
2 2
2p | X ( ) |
• Thus |X(w)|2Dw represents the signal energy in the band (w, w+Dw)
| X (w )|2
X (t ) Energy in (w ,w +Dw )

0 t
0 w
w w + Dw
 IV. Stochastic processes
4.3. Power spectrum
• For stochastic processes,
• A direct application of Fourier transform generates a sequence of
random variables for every w
• For a stochastic process, E{|X(t)|2} represents the ensemble average
power (instantaneous energy) at the instant t.
• Partial Fourier transform of a process X(t) based on (– T, T ) is given
by T - jw t
X T (w ) = ò -T X (t )e dt
• The power distribution associated with that realization based on (– T,
T ) is represented by

| X T (w ) |2 1 T - jw t
2

2T
=
2T ò -T X (t )e dt
IV. Stochastic processes
4.3. Power spectrum
• The average power distribution based on (– T, T ) is ensemble
average of power distribution for w
ì | X T (w ) |2 ü 1 T T - jw ( t1 - t2 )
PT (w ) = E í ý = ò- T ò- T
E { X ( t1 ) X *
( t 2 )}e dt1dt2
î 2T þ 2T
1 T T - jw ( t1 - t2 )
=
2T ò- T ò- T XX 1 2
R ( t , t ) e dt1dt2
• We have this represents the power distribution of X(t) based on (–
T, T ).
• If X(t) is assumed to be w.s.s, then
RXX (t1 , t2 ) = RXX (t1 - t2 )
• and we have
1 T T - jw ( t1 - t2 )
PT (w ) = ò -T ò -T XX 1 2
R ( t - t ) e dt1dt2 .
2T
IV. Stochastic processes
4.3. Power spectrum
• Let t = t2-t1, we obtain
1 2T - jwt
PT (w ) = ò - 2T XX
R (t ) e (2T - | t |)dt
2T
2T
= ò - 2T RXX (t )e - jwt (1 - 2|tT| )dt ³ 0
• This is the power distribution of the w.s.s. process X(t) based on
(– T, T ).
• Leting T®¥, we obtain
+¥
S XX (w ) = lim PT (w ) = ò-¥ RXX (t )e - jwt dt ³ 0
T ®¥
• Sxx(w) is the power spectral density of the w.s.s process X(t).
IV. Stochastic processes
4.3. Power spectrum
• Khinchin-Wiener theorem
• The autocorrelation function and the power spectrum of a
w.s.s process form a Fourier transform pair.

RXX (w ) ¬¾®
F×T
S XX (w ) ³ 0.
+¥
RXX (t ) = 1 w jwt
dw
2p ò-¥ XX
S ( ) e

+¥ - jwt
S
• For t = 0, XX (w ) = lim PT
(w ) = ò-¥ XX
R (t ) e dt ³ 0
T ®¥
+¥
1 w w
ò-¥ XX = = } = P,
2
2p S ( ) d R XX
(0) E {| X ( t ) | the total power.
IV. Stochastic processes
4.3. Power spectrum
• The area under Sxx(w) represents the total power of the
process X(t), and hence Sxx(w) truly represents the power
spectrum.

S XX (w ) S XX (w ) Dw
represents the power
in the band (w , w + Dw )

w w + Dw w
0

• The nonnegative-definiteness property of the auto-correlation

function translates into the “nonnegative” property for its
Fourier transform (power spectrum).
RXX (t ) nonnegative - definite Û S XX (w ) ³ 0.
IV. Stochastic processes
4.3. Power spectrum
• If X(t) is a real w.s.s process, then
RXX (t ) = RXX ( -t )
• So that +¥
S (w ) = ò -¥ R (t )e - jwt dt
XX XX

+¥
= ò -¥ RXX (t ) cos wt dt
¥
= 2 ò 0 RXX (t ) cos wt dt = S XX ( -w ) ³ 0
• The power spectrum is an even function, (in addition to
being real and nonnegative).
IV. Stochastic processes
4.4. Ergodicity
• Time averages
• Given wide-sense stationary process X(t).
• Time averages T
1
T ®¥ 2T ò
• Mean n = lim x (t )dt
-T

• Autocorrelation T
1
R (t ) = lim
T ®¥ 2T ò x(t + t )x (t )dt
-T
• These limits are random variables.
• Problems:
? ?
n = E{x (t )} R (t ) = E{x (t + t ) x (t )}
 IV. Stochastic processes
4.4. Ergodicity
• Ergodicity
• X(t) is ergodic if in the most general form if all its
statistics can be determined from a single function X(t, z)
of the process.
• X(t) is ergodic if time averages equal ensemble averages
(expected values)
 IV. Stochastic processes
4.4. Ergodicity
• Ergodicity of the mean X (t, x )

• Time average of a given process X(t) !

T X (t, x n )
1
ò x (t )dt
!
nT = X (t, x k )

• nT is a random variable. 2T -T X (t, x 2 ) !

• Since E{X(t)} is a constant, we have
X (t, x1 )
E{nT} = E{X(t)} = h t
0 t1 t2
• The variance of nT is given by:

1 æ t ö
2T
sn = ò ç1 - ÷[ R(t ) - h ]dt
2 2
T
T 0 è 2T ø
• R(t) is the autocorrelation of X(t).
• If this variance tends to zero with T®¥, then nT tends to its
expected value.
IV. Stochastic processes
4.4. Ergodicity
• Ergodic theorem for E{X(t)}

T
1
lim
T ®¥ 2T
-T
ò x (t )dt = E{x (t )} = h
iff
æ t ö
2T
1
ò0 è 2T ø t - h ]dt = 0
2
lim ç 1 - ÷[ R ( )
T ®¥ T
IV. Stochastic processes
4.4. Ergodicity
• Ergodicity of autocorrelation
• We form the average:
T
1
R T (l ) =
2T ò x(t + l )x (t )dt
-T
• We have T
1
E{R T (l )} =
2T -T
ò E{x(t + l )x (t )}dt = R(l )
• For a given l, RT(l) is the time average of the process
F(t)=x(t+l)x(t)
• The mean of the process F(t) is given by
E{F(t)}=E{x(t+l)x(t)} = R(l)
IV. Stochastic processes
4.4. Ergodicity
• Its autocorrelation
RFF(t)=E{x(t+l+t)x(t+t)x(t+l)x(t)}
• Hence with w(t) = F(t), we have
• Ergodicity theorem for autocorrelation
• For a given l
T
1
lim
T ®¥ 2T
-T
ò x(t + l ) x (t )dt = E{x(t + l ) x (t )} = R(l )
iff
æ t ö
2T
1
ò0 è 2T ø FF t - (l )]dt = 0
2
lim ç 1 - ÷[ R ( ) R
T ®¥ T
IV. Stochastic processes
4.4. Ergodicity
• Ergodicity of the distribution function
• We determine first order distribution F(x) = E{X(t)£x} of a given
process X(t) by a suitable time average.
• Consider the process
ì1 if x (t ) £ x
y(t ) = í
î0 if x(t ) > x
• Its mean is given by
• Its autocorrelation E{y(t )} = 1.P{x (t ) £ x} = F ( x)
• Where F(x,x;t) is the second order distribution of x(t)

E{y(t + t ) y(t )} = 1.P{x (t + t ) £ x, x (t ) £ x} = F ( x, x;t )

IV. Stochastic processes
4.4. Ergodicity
• We form the time average
T
1
yT =
2T ò y(t )dt
-T
• We have
E{yT} = E{y(t)} = F(x)
• The variance of yT is given by
1 æ t ö
2T

ò t - h ]dt = 0
2
ç 1 - ÷[ R ( )
T 0 è 2T ø
• Where R(t) and h are replaced by F(x, x;t) and F(x)
IV. Stochastic processes
4.4. Ergodicity
• Ergodic theorem for distribution function
• For a given x,
T
1
lim
T ®¥ 2T
-T
ò y(t )dt = F ( x)
iff
1 æ t ö
2T
lim ò ç1 - ÷[ F ( x , x; t ) - F 2
( x)]dt = 0
T ®¥ T 2T ø
0è
Stochastic Data Processing System
Models
Stochastic Data Processing
System Models
• Classification of Linear Systems
LTI systems
• Systems without memory
• Systems with memory
• Time Invariant systems
• Autocorrelation of output processes
• Power Spectrum and Khinchin-Wiener Theorem
LTI systems with stochastic input
• Deterministic system transformation
Y (t, x i )
X (t, x i )

¾¾¾®
X (t )
T [×] ¾Y¾®
¾
(t )

t t
• Y(t) = T[X(t)].
• Problems formulation: goal: to study the output
process statistics in term of the input process
statistics and the system function
• Is the output process stochastic ?
• How are statistics of the output process ?
• What is the relation between input and output processes ?
 LTI systems with stochastic input
• Scope: Response of the deterministics systems under
an actions of stochastic processes
Deterministic systems

Memoryless systems Systems with memory

Y (t ) = g [ X (t )]
Time varying systems Time Invariant Linear systems
systems
+¥
Y (t ) = ò - ¥ h(t - t ) X (t )dt Y (t ) = L[ X (t )]
+¥ Linear time-invariant systems
= ò - ¥ h(t ) X (t - t )dt .
X (t ) h (t ) Y (t )
4.1 Systems with stochastic inputs

• Memoryless systems
• Response of memoryless systems
• Y(t) = g[X(t)]
• First order distribution and density: FY(y,; t) và fY(y; t) in
relation with FX(x; t), fX(x; t):
FY ( y ) = P(Y (x ) £ y ) = P (g ( X (x )) £ y ) = P (X (x ) £ g -1 ( -¥, y ]).
• Expectation:
¥
E{Y (t )} = ò g ( x) f
-¥
X ( x; t )dx
4.1 Systems with stochastic inputs

• Second order ststistics:

• Second order density fY(y1, y2; t1, t2 of Y(t) can be
determined in term of fX(x1, x2; t1, t2)
• Autocorrelation: E{Y(t1)Y(t2)}:
¥ ¥
E{Y (t1 )Y (t 2 )} = ò ò g(x )g(x ) f
- ¥- ¥
1 2 X ( x1 , x2 ; t1 , t 2 )dx1dx2

• n order statistics: n order density function of Y(t)

fY(y1, y2, …yn; t1, t2, …, tn)

(t1) = g(X(t1)), …, Y(tn) = g(X(tn)).

45

4.1 Systems with stochastic inputs

Memoryless Systems:
The output Y(t) in this case depends only on the present value of the
input X(t). i.e.,
Y (t ) = g{ X (t )}
Strict-sense Strict-sense
stationary input Memoryless
system stationary output.

Wide-sense Need not be

Memoryless stationary in
stationary input system any sense.

X(t) stationary Y(t) stationary,but

Gaussian with Memoryless not Gaussian with
system R XY (t ) = hR XX (t ).
RXX (t )
4.1 Hệ thống TTBB dưới tác dụng
của quá trình ngẫu nhiên
• Khảo sát tính dừng của tín hiệu đầu ra.
• Khi tín hiệu đầu vào là tín hiệu dừng theo nghĩa hẹp, tín hiệu
ra cũng dừng theo nghĩa hẹp;
• Nếu đầu vào X(t) là dừng theo bậc N, đáp ứng Y(t) cũng
dừng theo bậc N;
• Nếu X(t) là dừng trong khoảng thì Y(t) cũng dừng trong
khoảng đó;
• Nếu X(t) là dừng theo nghĩa rộng, Y(t) có thể không dừng
theo mọi nghĩa.
• Ví dụ:
• Bộ thu nhận theo luật bình phương: Y(t) = X2(t)
4.1 Systems with stochastic inputs

• COnsider the memoryless system

ì 1, x³0
g ( x) = í
î- 1, x<0
48

4.1 Systems with stochastic inputs

Linear Systems: L[×] represents a linear system if
L{a1 X (t1 ) + a2 X (t2 )} = a1 L{ X (t1 )} + a2 L{ X (t2 )}.
Let Y (t ) = L{X (t )}
represent the output of a linear system.
Time-Invariant System: L[×] represents a time-invariant system if

Y (t ) = L{X (t )} Þ L{X (t - t0 )} = Y (t - t0 )
i.e., shift in the input results in the same shift in the output also.
If L[×] satisfies both equations, then it corresponds to
a linear time-invariant (LTI) system.
LTI systems can be uniquely represented in terms of their output
h (t ) to
Impulse
a delta function response of
the system
d (t ) LTI h(t )
t

Impulse Impulse
response
49
4.1 Systems with stochastic inputs Y (t )

X (t )
t
X (t ) Y (t )
t LTI
+¥
Y (t ) = ò - ¥ h(t - t ) X (t )dt
arbitrary
input +¥
= ò - ¥ h(t ) X (t - t )dt
Eq. follows by expressing X(t) as
+¥
X (t ) = ò - ¥ X (t )d (t - t )dt
and we have
Y (t ) = L{X (t )}.
+¥
Y (t ) = L{ X (t )} = L{ò - ¥ X (t )d (t - t )dt }
+¥
= ò - ¥ L{ X (t )d (t - t )dt } By Linearity

+¥
= ò - ¥ X (t ) L{d (t - t )}dt By Time-invariance
+¥ +¥
= ò - ¥ X (t )h (t - t )dt = ò - ¥ h (t ) X (t - t )dt .
4.1 Systems with stochastic inputs

• Output statistics
• Mean of the output
+¥
µ (t ) = E{Y (t )} = ò - ¥ E{ X (t )h(t - t )dt }
Y

+¥
= ò - ¥ µ X (t )h(t - t )dt = µ X (t ) * h(t ).

• Cross-correlation of input-output
R XY (t1 , t2 ) = E{ X (t1 )Y * (t2 )}
+¥
= E{ X (t1 ) ò - ¥ X * (t2 - a )h * (a )da }
+¥
= ò - ¥ E{ X (t1 ) X * (t2 - a )}h * (a )da
+¥
= ò - ¥ R XX (t1 , t2 - a )h * (a )da
= R XX (t1 , t2 ) * h * (t2 ).
4.1 Systems with stochastic inputs
• Autocorrelation of output

RYY (t1 , t 2 ) = E{Y (t1 )Y * (t 2 )}

+¥ +¥
= E{ò X (t1 - b )h( b )db Y * (t 2 )} = ò E{ X (t1 - b )Y * (t 2 )}h( b )db
-¥ -¥
+¥
= ò RXY (t1 - b , t 2 )h( b )db = RXY (t1 , t 2 ) * h(t1 )
-¥

RYY (t1 , t2 ) = R XX (t1 , t2 ) * h * (t2 ) * h(t1 ).

µ (t )
X h(t) µ (t )
Y

(a)

RXX (t1 , t 2 ) ¾
¾® h*(t2) ¾¾ ¾¾®
R XY ( t1 ,t 2 )
h(t1) ¾
¾® RYY (t1 , t 2 )
(b)
4.1 Systems with stochastic inputs
• If X(t) - wide sense stationary µ (t ) = µ
X X
+¥
µ (t ) = µ
Y X ò -¥ h(t )dt = µ X
c, a constant.
RXX (t1 , t 2 ) = RXX (t1 - t 2 )
• X(t), Y(t) are jointly w.s.s
+¥
R XY (t1 , t2 ) = ò - ¥ R XX (t1 - t2 + a )h * (a )da
= R XX (t ) * h * ( -t ) = R XY (t ), t = t1 - t2 .
• Y(t) is w.s.s.
+¥
RYY (t1 , t 2 ) = ò -¥ RXY (t1 - b - t 2 )h( b )db , t = t1 - t 2
= RXY (t ) * h(t ) = RYY (t ).
RYY (t ) = RXX (t ) * h* (-t ) * h(t ).
53

4.1 Systems with stochastic inputs

The output process is also wide-sense stationary.
X (t ) This gives rise to the following representation
Y (t )
wide-sense wide-sense
stationary process LTI system
h(t) stationary process.

(a)

X (t ) Y (t )
strict-sense LTI system
stationary process strict-sense
h(t) stationary process
(see Text for proof )
(b)

X (t ) Y (t )
Gaussian process
Gaussian
process (also Linear system (also stationary)
stationary)
(c)
4.1 Systems with stochastic inputs
• Theorem:
• For linear systems:
E{L[X(t)]} = L[E{X(t)}]
 4.1 Systems with stochastic inputs

• White noise
• W(t) is said to be a white noise process if
RWW (t1 , t 2 ) = q (t1 )d (t1 - t 2 ),
• E[W(t1) W*(t2)] = 0 unless t1 = t2
• W(t) is said to be wide-sense stationary (w.s.s) white
noise if
• E[W(t)] = constant, and

RWW (t1 , t 2 ) = qd (t1 - t 2 ) = qd (t ).

• If W(t) is also a Gaussian process (white Gaussian
process), then all of its samples are independent random
variables
4.1 Systems with stochastic inputs
LTI Colored noise
White noise
h(t) N (t ) = h (t ) *W (t )
W(t)
• For w.s.s. white noise input W(t), we have
+¥
E [ N (t )] = µW ò -¥ h(t )dt , a constant
Rnn (t ) = qd (t ) * h* (-t ) * h(t ) = qh* (-t ) * h(t ) = qr (t )
• where
+¥
r (t ) = h(t ) * h (-t ) = ò -¥ h(a )h* (a + t )da .
*

• Thus the output of a white noise process through an LTI

system represents a (colored) noise process.
• . White noise need not be Gaussian.
• “White” and “Gaussian” are two different concepts.
4.2 Discrete Time Stochastic
Processes
• Definition
• A discrete time stochastic process Xn = X(nT) is a
sequence of random variables.
• The mean, autocorrelation and auto-covariance
functions of a discrete-time process are:

µ n = E{ X (nT )}
R(n1 , n2 ) = E{ X (n1T ) X (n2T )}
*

C (n1 , n2 ) = R(n1 , n2 ) - µ n1 µ n*2

4.2 Discrete Time Stochastic
Processes
• Tính dừng
• Strict sense stationarity and wide-sense stationarity
definitions apply here also.
• X(nT) is wide sense stationary if :
E{ X (nT )} = µ , a constant
E [ X {( k + n )T } X * {(k )T }] = R (n ) = rn = r-*n
• R(n1, n2) = R(n1 – n2) = R*(n2 – n1)
• The positive-definite property of the autocorrelation
sequence æ r r r ! r ö
0 1 2 n
• ç * ÷ , n = 0, 1, 2, …, ¥
ç r1 r0 r1 ! rn -1 ÷
Tn = ç ÷ = Tn
*

"
ç ÷
ç r* r* ! r* r ÷
è n n -1 1 0 ø
4.2 Discrete Time Stochastic
Processes
• Output of the LTI systems
• X(nT) - W.S.S, LTI system h(nT), Y(nT) - response.
• Cross-correlation of input-output are:
R XY ( n ) = R XX ( n ) * h * ( -n )
RYY ( n ) = R XY ( n ) * h( n )
RYY ( n ) = R XX ( n ) * h * ( -n ) * h( n ).

• Thus wide-sense stationarity from input to output is

preserved for discrete-time systems also
4.3. Auto Regressive Moving
Average (ARMA) Processes
• Consider an input
p
– output representation
q
of LTIs
X ( n ) = - å a k X ( n - k ) + å bkW ( n - k ),
k =1 k =0
• where X(n) may be considered as the output of a system
{h(n)} driven by the input W(n)
• The transfer function
W(n) h(n) X(n)
p q
X ( z ) å ak z - k = W ( z ) å bk z - k , a0 º 1
k =0 k =0
-1 -2 -q
¥
X ( z ) b0 + b1 z + b2 z + ! + bq z B( z )
H ( z ) = å h(k ) z - k = = -1 -2 -p
=
k =0 W ( z ) 1 + a1 z + a2 z + ! + a p z A( z )
¥
X ( n ) = å h( n - k )W ( k ).
k =0
4.3. Auto Regressive Moving
Average (ARMA) Processes
n Transfer function has p poles and q zeros
• The output undergoes regression over p of its
previous values and at the same time a moving
average based on W(n), W(n-1) ..., W(n-q) of the
input over (q + 1) values is added to it (ARMA(p, q))
• The input {W(n)} represents a sequence of uncorrelated
random variables of zero mean µ = 0 and constant variance
s2 w so that RWW (n) = s W2d (n).
n If {W(n)} )}is normally distributed then the output {X(n also
represents a strict-sense stationary normal process.
n If q = 0, AR(p) – all-pole process.
n if p = 0, represent MA(q) process
4.3. Auto Regressive Moving
Average (ARMA) Processes
• Autoregressive process AR(1)
• An AR(1) process has the form:
X ( n ) = aX ( n - 1) + W ( n )
• The corresponding system transfer ¥
:
1
H ( z) =
1 - az -1
= å a n -n
z
n =0

• if |a| < 1 - stable system.

• System impulse response: h(n ) = a n , | a |< 1
• Output autocorrelation:
¥ |n|
a
RXX (n) = s W2d (n) * {a - n } * {a n } = s W2 å a |n|+ k a k = s W2
k =0 1 - a 2
 4.3. Auto Regressive Moving
Average (ARMA) Processes
• Tự tương quan chuẩn hóa của đầu ra:
R XX ( n )
r (n) = = a |n |
, | n | ³ 0.
R XX (0)
X

• Trường hợp quá trình X(t) kết hợp với nhiễu V(t):
• V(t) là chuỗi ngẫu nhiên không tương quan với giá trị trung
bình 0 và tự tương quan sV2.
• Y(t) = X(t) + V(t)
|n|
a
RYY (n) = RXX (n) + RVV (n) = RXX (n) + s V2d (n) = s W2 + s 2
d ( n)
1- a 2 V

R (n) ì1 n = 0 s 2

r (n) = YY
= í |n| c= 2 W
< 1.
Y
R (0) îc a
YY n = ±1, ± 2,! s + s (1 - a )
W
2
V
2
4.3. Auto Regressive Moving
Average (ARMA) Processes
• the effect of superimposing an error sequence
on an AR(1) model AR(1)
r X (0) = r Y (0) = 1

r X (k ) > rY (k )

n
0 k
4.3. Auto Regressive Moving
Average (ARMA) Processes
• Autoregressive AR(2)
X (n) = a1 X (n - 1) + a2 X (n - 2) + W (n)
• Transfer function
¥
1 b1 b2
H ( z ) = å h( n) z -n
= = +
n =0
-1
1 - a1 z - a2 z -2
1 - l1 z -1
1 - l2 z -1
h(0) = 1, h(1) = a1 , h(n) = a1h(n - 1) + a2 h(n - 2), n ³ 2
• with l1, l2 - poles of the system, the impulse response:
• Và các hệ thức: h(n) = b1l1n + b2 ln2 , n ³ 0
b1 + b2 = 1, b1l1 + b2 l2 = a1. l1 + l2 = a1 , l1l2 = -a 2 ,
• H(z) stable follows | l1 |< 1, | l2 |< 1.
4.3. Auto Regressive Moving
Average (ARMA) Processes
• Autocorrelation of output process
• Autocorrelation:

RXX (n) = E{ X (n + m) X * (m)}

= E{[ a1 X (n + m - 1) + a2 X (n + m - 2)] X * (m)}
+ E{W (n + m) X * (m)}
= a1 RXX (n - 1) + a2 RXX (n - 2)
• Correlation coefficient:
RXX ( n )
r X (n) = = a1 r X ( n - 1) + a2 r X ( n - 2).
RXX (0)
 4.3. Auto Regressive Moving
Average (ARMA) Processes
• ARIMA processes (Autoregressive Integrated
moving average)
• For modeling time series
• Consider ARMA processes:
p W – white
q noise
X ( n ) = - å a k X ( n - k ) + å bkW ( n - k ),
k =1 k =0

( 1
(1 + $ 𝐷 % ) 𝑋 𝑛 = (𝑏/ + $ 𝐷 0 )𝑊(𝑛)
%&' 0&'
• Di – delay operator, or lag operator
• An ARIMA(p’,d,q) process is a particular case of an ARMA(p,q)
process having the autoregressive polynomial with d unit roots
and p’ = p-d .
4.3. Auto Regressive Moving
Average (ARMA) Processes
• Autocorrelation

RXX (n) = RWW (n) * h* (- n) * h(n) = s W2 h* (- n) * h(n)

¥
= s W å h* ( n + k ) * h( k )
2

k =0

æ | b | 2
( l* n
) b *
b ( l * n
) b b *
( l* n
) | b | 2
( l 2) ö
* n
= sW ç
2 1 1
+ 1 2 1
+ 1 2 2
+ 2
÷
è 1- | l1 | 1 - l1l2 1 - l1l2 1- | l2 | ø
2 * * 2

RXX (n) *n *n
r X ( n) = = c1l1 + c2 l2
RXX (0)
4.3. Auto Regressive Moving
Average (ARMA) Processes
• Moving average process
q
X(n) = ∑ bkW (n − k),
k=0

• q zeros; without poles;

• Non-regressive systems;
• Impulse response and transfer function:
∞
H ( z ) = ∑ h ( k ) z − k =b0 + b1 z −1 + b2 z −2 ++ bq z − q
k =0
Thank you for
your attentions!