Random Process Examples - Binghamton Universityws2.binghamton.edu/fowler/fowler personal...

1/23

Random ProcessRandom ProcessExamplesExamples

2/23

Ex. #1: D-T White NoiseLet x[k] be a sequence of RV’s where…

each RV x[k] in the sequence is uncorrelated with all the others:

E{ x[k] x[m] } = 0 for k≠m

Physically, uncorrelated means that knowing x[k] provides no insight into what value x[m] (for m≠k) will be likely to take (roll a die; the value you get provides no insight into what you expect to get on any future roll)

This DEFINES a DT White NoiseAlso called “Uncorrelated Process”

3/23

Ex. #1: D-T White Noise

Also, let x[k] be Gaussian with Zero mean and Variance σ²

Define RP’s PDF; also called “Normal”

Note: Does not depend on kat least the 1st order PDF is time invariant

X[k] ~ N (0,σ²)

Mean Variance

⎟⎟⎠

⎞⎜⎜⎝

⎛ −

=2

2

2

21);( σ

πσ

x

ekxpPDF :

Recall: Gaussian process are very common, so…

4/23

Ex. #1: D-T White NoiseTASK : We have a model…. Find the mean, ACF, and check if WSS (also find variance of process)

MEAN of Process :

ACF:

E { x[k] } = 0 CONSTANTBy definition !

RX(k1,k2)=E { x[k1] . x[k2] }

By our definition of white noise, ….this is 0 if k1≠ k2

5/23

Ex. #1: D-T White Noise

{ } 2221 ][),(

21σ==== kxEkkR kkkx

By definition of variance for zero-mean case

Now for k1=k2=k:

][

,0,),(

212

21

212

21

kk

kkkkkkRx

−=

⎪⎩

⎪⎨⎧

≠==

δσ

σThus,

][)( 2 mmRx δσ=⇒

= m(like τ for cont-time ACF)

ACF for DT White RP

6/23

Ex. #1: D-T White NoiseACF displays lack of correlation between any pair of any time instants:

Now since we have constant mean and ACF depends only on m = k2-k1 ⇒ WSS

RX[m] = σ²δ[m]

m

σ²

-3 -2 -1 1 2 3

……

7/23

Ex. #1: D-T White NoiseVariance

2

22

]0[]0[

σ

σ

=

=

−=

x

xx

RxR

=0

For this case:Variance of the Process = Variance of the RV

8/23

Ex. #2: Filtered D-T RPStart with White RP x[k] in previous example

Recall : Zero Mean ProcessRX [m] = σ² δ[m]⇒ WSS

x[k] D-T Filterh[n] = [1 1]

y[k] = x[k] + x [k-1]

Two–Tap FIR filterTaps = [1 1]

9/23

Ex. #2: Filtered D-T RPTASK: Is y[k] WSS?

⇒ need to find mean & ACF

MEAN: Using filter output expressions gives

{ } { }{ } { }]1[][

]1[][][−+=

−+=kxEkxE

kxkxEkyE

⇒ E {y[k]} = 0

10/23

Ex. #2: Filtered D-T RPACF :

{ }{ }

{ } { }

{ } { }))1()1((

21)1(

21

)1(21

)(21

2211

2121

1212

1212

]1[]1[][]1[

]1[][][][

])1[][])(1[][(

][][),(

−−−+−

−−−

−−+−+

−+=

−+−+=

=

kkRkkR

kkRkkR

y

xx

xx

kxkxEkxkxE

kxkxEkxkxE

kxkxkxkxE

kykyEkkR

Plug in Eq. for output

11/23

Ex. #2: Filtered D-T RP

⇒ RY(m) = σ²[2δ[m] + δ[m-1] + δ[m+1]

( )][

12

]1[

12

]1[

12

][

1221

22

22

)1()1()1(

)1()(),(

m

x

m

x

m

x

m

xy

kkRkkR

kkRkkRkkR

δσδσ

δσδσ

−−−++−+

−−+−=

+

−

12 where kkm −=

ACF for 2-Tap Filtered White RP

y[k] is WSS

12/23

Ex. #2: Filtered D-T RP

Note : Filter introduces correlation between adjacent samples - but still no correlation for samples 2 or more samples apart (for this filter)

Ry[m]

m

2σ²

-3 -2 -1 1 2 3

…… σ²

13/23

Big Picture: Filtered RPFilters can be used to change the correlation structure of aRP:

x[k] D-T Filterh[n] = [1 1]

y[k] = x[k] + x [k-1]

0 50 100 150 200 250 300

0

White Noise

-20 -10 0 10 200

0.5

1

ACF Rx[m] of Input Input RP (One Sample Function)

km

0 50 100 150 200 250 300

0

-20 -10 0 10 200

0.5

1

ACF Ry[m] of Output Output RP (One Sample Function)

km

14/23

Big Picture: Filtered RP (cont)

km-20 -10 0 10 20 0 50 100 150 200 250 300

0Filter: ones(1,15)

0

0.5

1

ACF Ry[m] of Output Output RP (One Sample Function)

0 50 100 150 200 250 300

0

Filter: [1 1 1 1]

-20 -10 0 10 200

0.5

1

ACF Ry[m] of Output Output RP (One Sample Function)

km

0 50 100 150 200 250 300

0

Filter: ones(1,10)

-20 -10 0 10 20

0

0.5

1

ACF Ry[m] of Output Output RP (One Sample Function)

km

15/23

Filtered RPs: Insight

Narrow ACF Rapid FluctuationsBroad ACF Slow Fluctuations

Our study of the ACFs of filtered random processes and the degree of “smoothness” of the sample functions shows the following general result:

16/23

Ex. #3: Exp. Sig. + White Noise

Define a RP as:

][][ kwxaky k +=

Deterministic Functiona = deter. #

Random Variable(picked once randomly and then fixed)

White Noise(get a new random value at

each k – uncorrelated from sample-to-sample)

17/23

Ex. #3: Exp. Sig. + WN Further Definition of this RP:

• w[k] is white noise

• Each w[k] is a Gaussian RV with

• X is a Gaussian RV with

• RV X is independent of each RV w[k]⇒ E{ x w[k]} = E{x} E{w[k]} = 0

• The # a is a deterministic number

),0(~ 2XNX σ

),0(~][ 2wNkw σ

18/23

Ex. #3: Exp. Sig. + WNTASK: Is this WSS?

MEAN: E{ y[k] } = ak E{X} + E{ w[k] } = 0 (Constant)

ACF: Ry(k,k+m) = E{ y[k] y[k+m] }= E {(akX + w[k]) (ak+m X + w[k+m])}

= a2k+m E{X2} + ak E{X w[k+m]} + ak+m E{X w[k]} + E {w[k] w[k+m]}

= 0 = 0

= 0Due toIndep.

][mwδσ= 2

2Xσ

19/23

Ex. #3: Exp. Sig. + WN

][),( 222 mamkkR wmk

xy δσσ +=+ +

Not WSS !!!

In General: Depends on k not just m

But… If a = ±1 then a2K+M = a2K aM = aM

Then it is WSS

20/23

Ex. #3: Exp. Sig. + WN

If a = -1

If a = +1RX[m]

m1 2 3-3 -2 -1

2xσ

22wx σσ +

RX[m]

m

12

3-3-2

-1

2xσ

22wx σσ +

2xσ−

21/23

Ex. #4: Weird Func of WNLet z[k] be white noise with zero mean

& variance of

Define x[n] = Z [int(n/2)]

Then x[0] = z[0]x[1] = z[0]x[2] = z[1]x[3] = z[1]

2zσ

Always appear in pairs

int(D3D2D1.d1d2d3d4)= D3D2D1

22/23

Ex. #4: Weird Func of WNTASK: Is this RP WSS ?

MEAN: E { x[n] } = E { z[k] } = 0

RX(n, n+m) = E{ x[n] x[n+m] }

If |m| ≥ 2 then x[n] & x[n+m] are two different z[.] values ⇒ uncorrelated

⇒ RX(n,n+m) = 0 for |m| ≥ 2

k = n/2 if n is even= (n-1)/2 if n is odd

23/23

Ex. #4: Weird Func of WN

n dependence causes it not to be WSS

If m = 0 then RX(n, n ) = E { z2[k] } =

If m = 1 … and n is even: RX(n, n+1) =

… and n is odd: RX(n, n+1) = 0

2zσ

2zσ

m

n