1 Lecture 16 Random Signals and Noise (III) Fall 2008 NCTU EE Tzu-Hsien Sang
Dec 22, 2015
11111
Lecture 16 Random Signals and Noise (III)
Fall 2008
NCTU EE
Tzu-Hsien Sang
2
Outline
• Terminology of Random Processes
• Correlation and Power Spectral Density
• Linear Systems and Random Processes
• Narrowband Noise
33
Linear Systems and Random Processes
– Without memory: a random variable a random variable
– With memory: correlated outputs
• Now, we study the statistics between inputs and outputs, e.g., my(t), Ry(), …
• Assume X(t) stationary (or WSS at least) H() is LTI.
3
H X(t) = X(t,) Y(t) = Y(t,)
444
).()()()()()(
).()()()()()(
).()()(
).()()(
.)()()]()([)(
])()()([)]()([)(
:Functionn Correlatio
).0()]([)(
])()([)]([)(
: ofMean
* fSfHfSfHfSfS
RhRhRR
fSfHfS
RhR
duuRuhduutXtXEuh
duutXuhtXEtYtXER
HmduutXEuh
duutXuhEtYEtm
Y(t)
XXXYYX
XXXYYX
XXY
XXY
X
XY
XS
SSy
555functions.n correlatio andmean
by zedcharacteripletely -com is thusandGaussian joint also
is then Gaussian,joint is ifBut . of pdfjoint the
find todifficult very isit , of pdfjoint given the general,In
n).correlatio (mean, statisticsorder 2ndconsider only wefar, So (2)
too. WSSis
,condition) initial (no LTI is and WSSis If (1) :Remarks
).()()()()()(
).()()()()()(
).()()()(
)]()([)(])()()([
)}]()(){([)]()([)(
2*
Y(t)X(t)Y(t)
X(t)
Y(t)
h(t)X(t)
fSfHfSfHfHfS
RhhRhR
RhduuRuh
duutXtYEuhduutXuhtYE
tXthtYEtYtYER
XXY
XYXY
YXYX
Y
66
• Gaussian Random Process: X(t) has joint Gaussian pdf (of all orders).
– Special Case 1: Stationary Gaussian random process
Mean = mx; auto-correlation = RX().– Special Case 2: White Gaussian random process
RX(t1,t2) = (t1-t2) = () and SX(f) = 1 (constant).
6
0
Show that if ( ) is stationary Gaussian, so is ( ).
Case I: ( ) is white ( independent random variables).
( ) ( ) ( ) ( ) ( ) lim ( ) ( )
weighted sum of Gausk
X t Y t
X t
Y t h t X t X h t d X k h t k
1 2 2 1 1
sian random variables.
( ) has a 1st-order Gaussian distribution. Similarly, the higher
order joint pdf of ( ) is joint Gaussian. For example,
( , ) ( | ) ( ) (would be jointly GaY Y Y
Y t
Y t
f y y f y y f y
ussian).
777
1
2 1
Case II: ( ) is Gaussian but not white (more realistic case).
Claim: ( ) is produced by passing a white Gaussian random process
through ( ).
Then, define ( ) ( ) ( ), we have ( )
x t
x t
h t
h t h t h t y t
2 ( ) ( ).
Back to Case I. Therefore, ( ) is stationary Gaussian.
Note: We often use lower-case letters ( and instead of and )
for denoting random processes. You need to judge by the context
h t Z t
y t
x y X Y
whether it is a deterministic signal or a random process.
88
• Properties of Gaussian Processes(1) X(t) Gaussian, H() stable, linear Y(t) Gaussian.(2) X(t) Gaussian and WSS X(t) is SSS.(3) Samples of a Gaussian process, X(t1), X(t2), …, are
uncorrelated They are independent.(4) Samples of a Gaussian process, X(t1), X(t2), …, have
a joint Gaussian pdf specified completely by the set of means and auto-covariance function .
• Remarks: Why do we use Gaussian model?– Easy to analyze.– Central Limit Theorem: Many “independent” events
combined together become a Gaussian random variable ( random process ).
99
• Example: RC filter with white Gaussian input.
0
0
0
3
3
2 0
2
3
0
Input: is white Gaussian and zero-mean with ( ) .2
1 1Filter: ( )
1 2 1
1where =3 dB cutoff frequency = .
21
The output: ( ) ( ) ( ) .2 1 ( )
( ) .4
i
i
n
n n
RCn
NS f
H ffj fRC j f
fRC
NS f S f H f
ff
NR e
RC
1010
0 0
0
0
0
2 2 00
2 0 00 20
0
20
Output power: ( ) (0) .4
Another approach: ( ) ( ) .2 1 4
Mean: ( ) 0 (0) 0.
Another approach: ( ( )) lim ( ) 0.
The first-order pdf: ( ,
n n
n
n
n
Nn t R
RCN Ndx
n t S f dfRC x RC
n t H
n t R
f y t
2
0
2
0
1) .
2
Note: ( ) does not "completely" describes the behaviour of
a random process.
RCy
NeNRC
R
1111
• Noise equivalent Bandwidth
It is just a way to describe a band-limited noise with the bandwidth of an ideal band-pass filter.
2
2
( )1.
2 { ( ) }N
h t dtB
h t dt
1212
Narrowband Noise
• Q: Besides certain statistics, is there a more “waveform-oriented” approach to describe a noise (or random signal)?
2W
-f0 f0 f
0
Interpretation: A baseband radom process is shifted to a higher frequency.
( ) ( ) cos 2 .n t R t f t
0W f
1313
0
0
However, a bandpass random signal can have a random phase:
( ) ( ) cos(2 ( )).
In general, if we allow a time-invariant phase bias , then
(Envelope-phase representation) ( ) ( ) cos( ( ) )
o
n t R t f t t
n t R t w t t
0 0
2 2 1
r (Quadrature-component representation)
( ) ( ) cos( ) ( )sin( )
( )with ( ) ( ) ( ) and ( ) tan ( ).
( )
c c
sc s
c
n t n t w t n t w t
n tR t n t n t t
n t
1414
0 0
How to produce (t) and ( ):
( ) ( ) cos( ) ( )sin( ) (in mean-square error)
Remarks: Here, we assume is a random variable, independent
of ( ), uniformly distributed over (0, 2 )
c s
c c
n n t
n t n t w t n t w t
n t
1 2
or (- , ). If is
not a random variable, ( ) and ( ) are not WSS. We cannot
use LTI theory to predict the outputs of LPF's.
z t z t
1515
Properties of Quadrature-component representation
(1) ( ) ( ) ( ) 0.
Proof: (i) Let ( ) ( ) ( ).
( ) [ ( ) ( )] ( ( ) ( ))( ( ) ( ))
( ) ( ) ( ) ( ) ( ) ( )
c s
X
n t n t n t
X t n t n t
R E X t X t n t n t n t n t
n t n t n t n t n t n t n
2
2
2 2
2
1
( ) ( )
( ) ( ( )) . ( WSS)
( 0) ( ) [ ( ) ( ( )) ]
(0) ( ( )) 0- ( ( )) 0.
The only possibility is ( ( )) 0.
(ii) { ( )
n
X X n
n
t n t
R n t
S f R d R n t d
S n t d n t d
n t
E z t
0
1
} 2 ( ) cos( ) 0.
{ ( )} { ( )} (0) 0. Similarly, { ( )} 0.c s
n t t
E n t E z t H E n t
1616
0 0
0 0
0 0
1
(2) ( ) ( ) { ( ) ( )}
( ) ( ), .
0, otherwise
( ) { ( ) ( )}.
Proof: (i) ( ) 2 ( )
c s
c s
n n n n
n n
n n n n
S f S f Lowpass S f f S f f
S f f S f f W f W
S f j Lowpass S f f S f f
z t n t
1
0
1 1
0 0
0 0 0
0 0 0
cos( ).
( ) { ( ) ( )}
{4 ( ) ( ) cos( ) cos( ( ) )}
2 { ( ) ( )}cos 2 { ( ) ( ) cos(2 2 )}
2 ( ) cos 2 { ( ) ( )} {cos(2
Z
n
t
R E z t z t
E n t n t t t
E n t n t E n t n t t
R E n t n t E t
1
0
0 0
0 0
1
0 0
2 )}
2 ( ) cos .
Thus, ( ) ( ) [ ( ) ( )]
( ) ( ).
(t) is the low-pass portion of ( ).
( ) { ( ) ( )}.
Similarly, c
n
z n
n n
c
n n n
R
S f S f f f f f
S f f S f f
n z t
S f Lowpass S f f S f f
0 0( ) { ( ) ( )}.sn n nS f Lowpass S f f S f f
1717
1 2
1 2
1 2
0 0
0
0 0
1
Proof: (ii) ( ) { ( ) ( )}
{4 ( ) ( ) cos( )sin( ( ) )}
2 ( )sin .
Thus, ( ) [ ( ) ( )].
( ) { ( ) ( )}
{ ( ) (
c s
z z
n
z z n n
n n c s
R E z t z t
E n t n t t t
R w
S f j S f f S f f
R E n t n t
E h u Z t
1 2
1 2
1 2
2
1 2
*
) ( ) ( ) }
( ) ( ) { ( ) ( )}
( ) ( ) ( )
( ) ( ) ( ).
( ) ( ) ( ) ( )
(
c s
Z Z
Z Z
n n Z Z
u du h v Z t u dv
h u h v E Z t u Z t v dudv
h u h v R u v dudv
h h R
S f H f H f S f
H f
1 2
2
2
0 0
0 0
) ( )
( ) [ ( ) ( )]
{ ( ) ( )}.
Z Z
n n
n n
S f
j H f S f f S f f
j Lowpass S f f S f f
18180)(such that and those)(
)(
at eduncorrelat are )( and )( then
, respect to with symmetric NOT is )( If :Remarks
ed.uncorrelat are )( and )(
. 0)(
.0)}()({
, respect to with symmetric is )( :Proof
. and allfor eduncorrelat are
)2( and )1( then , respect to with symmetric is )( If (4)
.)()( )()()()(
:Proof
.)()()( (3)
2121
21
21
0
00
0
21
0
222
222
ttRttii
tti
tntn
ffS
tntn
R
ffSffSLowpass
ffS
tt
tntnffS
tndffStndffSdffStn
Ntntntn
sc
sc
sc
nn
sc
n
sc
nn
nn
n
scn
nsnnc
sc
1919
. ,0 ,2
),(
:)( and )(for and
, ,2
),;,(
is pdfjoint their t,independen are )( and )( If (6)
).( ofn combinatiolinear weightedare )( and )( :Proof
Gaussian. are )( and )( then Gaussian, is )(If (5)
2
2/)(
2
22
reN
rrf
ttRN
etntnf
tntn
tntntn
tntnt n
N
r
Nnn
sc
sc
sc
sc
sc
2020.)()(
2
1)(
2
1
)sin()cos()()(2
)(sin)()(cos)(
})]sin()()cos()({[)(ˆ
.)(ˆ)(ˆ)(2)(})](ˆ)({[
sense) square-meanin (converge .0})](ˆ)({[ show Wish to
).sin()()cos()()(ˆLet :Proof
).( oft independen is and
),(-over ddistributeuniformly variablerandom a is that Note
.0}))]sin()()cos()(()({[
is, that sense; square-meanin
)sin()()cos()(by drepresente becan n(t) then
, =BW with )( process random bandpass WSSaGiven :Theorem
222
00
022
022
200
2
222
2
00
200
00
Ntntntn
tttntn
ttnttn
ttnttnEtn
tntntntntntnE
tntnE
ttnttntn
tn
ttnttntnE
ttnttn
Btn
sc
sc
sc
sc
sc
sc
sc
2121.)cos()()(2
))(cos()()(2
))(cos()(1)(
))(cos()(1)(
})]2)(cos())()[cos((2)()({
})]2)(cos())()[cos((2)()({
})sin()sin()(2)()({
})cos()cos()(2)()({)(ˆ)(
.)sin()(2)()(
.)cos()(2)()]cos()(2[)()(
)]}.sin()()cos()()[({)(ˆ)(
0
0
0
0
00
00
00
00
0
00
00
dRh
duututRuth
dvtvtvRvth
dututuRuth
dvtvtvvntnvthE
dututuuntnuthE
dvtvvntnvthE
dutuuntnuthEtntn
dvvvnvthtn
duuunuthttnthtn
ttnttntnEtntn
n
n
n
n
s
c
sc
2222
Q.E.D. .0)()(2)(})](ˆ)({[
.)()()()]()([)(ˆ)(
),()(
i.e., ),(2
1)(
2
1cos)( and
Theorem) s(Parseval' ,)()()()(
.)cos()()(2)(ˆ)(
2222
200
000
*
0
tntntntntnE
tndffSdffSffHffHtntn
fSR
ffHffHwh
dffYfXdttytx
dRhtntn
nn
nn
n
2323
• Example: A bandpass signal
2424
.6sin)4(sinc16)(
).()}()({)(
).()}()({)()(
.correlated are )( and )(
. respect to with symmetric NOT is )( Hz, 5select If )2(
.24)62(2)(
and eduncorrelat are )( and )(
. respect to with symmetric is )( Hz, 7select If (1)
00
00
00
00
sc
sc
sc
nn
nnnn
nnnn
sc
n
n
sc
n
R
dffSffSLowpassjfS
cffSffSLowpasstSfS
tntn
ffSf
dffSN
tntn
ffSf