Scribd
Upload
23 views

AdvDSP lecture2-RandomProcesses1

1) The document discusses random processes (RPs), including their characterization using probability density functions and autocorrelation functions. 2) RPs can be classified as strictly stationary, wide sense stationary, or ergodic based on properties like having time-invariant statistics or probability distributions. 3) The power spectral density (PSD) of a RP is related to its autocorrelation function by the Wiener-Khinchine theorem. The PSD provides information about the distribution of power over frequency. 4) Filtering a RP changes its autocorrelation and PSD in predictable ways. Filtering can be used to modify the correlation properties of RPs.

Uploaded by

Alireza Dabiry
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
23 views

AdvDSP lecture2-RandomProcesses1

1) The document discusses random processes (RPs), including their characterization using probability density functions and autocorrelation functions. 2) RPs can be classified as strictly stationary, wide sense stationary, or ergodic based on properties like having time-invariant statistics or probability distributions. 3) The power spectral density (PSD) of a RP is related to its autocorrelation function by the Wiener-Khinchine theorem. The PSD provides information about the distribution of power over frequency. 4) Filtering a RP changes its autocorrelation and PSD in predictable ways. Filtering can be used to modify the correlation properties of RPs.

Uploaded by

Alireza Dabiry
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 104

Advanced DSP

Lecture 3
Stochastic Processes
RANDOM PROCESSES
Discrete time

Continuous time
Two Views of RP
• Viewing RP as a sequence of random variables
Two Views of RP

• RP is a collection of functions of time.


Each function is determined by a single
probabilistic event.
Two example
Realization functions from
(sample function) the random bits
process
Probabilistically characterizing RPs

Temperature at noontime, (k=day)


Probabilistically characterizing RPs
Probabilistically characterizing RPs

Time, t
Probabilistically characterizing RPs
The complete characterization of RPs is
practically impossible. Usually one makes do with
1st order i.e., P(x;t) and 2nd order i.e., P(x1,x2;t1,t2)
PDFs.
Probabilistically characterizing RPs
Probabilistically characterizing RPs
An Example Random Process
Time varying PDF of RP

Data for each


day is collected
from different
parts in
the world
Time varying PDF of RP

Variance
varies
Random Processes
Recall:
1st order PDF of RP gives the likelihood
of values occurring at each given time.

2nd order PDF characterizes Probabilistic


coupling between RP values at each t1 and t2.
Random Processes
Auto Correlation Function
Comparing ACFs of two RPs
Auto Correlation Function
Averages
Classifying RPs
Classifying RPs
Ergodicity
(Strictly) Stationary RP
(Strictly) Stationary RP
Time Invariant 1st order PDF
Stationary RP
Stationary RP
Stationary RP
Stationary RP
Stationary RP
Wide Sense Stationary RP
Wide Sense Stationarity
A WSS RP has constant variance
Example
Example (cont.)
Example (cont.)
Example (cont.)
Example (cont.)
Example (cont.)
Discrete Time White Noise
Discrete Time White Noise
Discrete Time White Noise
Discrete Time White Noise
Discrete Time White Noise
Example 2 Filtered RP
Example 2: Filtered RP
Example 2: Filtered RP
Example 2: Filtered RP
Example 2: Filtered RP
Filtered RP
Filters can be used to change the
correlation structure of a RP
Filtered RP
Power of RPs
Power of RPs
Relationship of Power to ACF
Power vs Variance
Power Spectral Density of a RP
Power Spectral Density of a RP
Definition of PSD for WSS RP
Wiener Khinchine Theorem
Proof of W K theorem
Proof of W K theorem (cont)
Proof of W K theorem (cont)
Proof of W K theorem (cont)
Proof of W K theorem (cont)
Proof of W K theorem (cont)
Proof of W K theorem (cont)
Proof of W K theorem (cont)
Proof of W K theorem (cont)
Properties of PSD and ACF
PSD for Discrete Time Processes
Example
Power Spectrum of Harmonic Process
Multiple Random Processes
Jointly WSS Processes

Interested in the case where x(t) and y(t) are


both WSS and also are jointly WSS.
This happens if

Rxy(t1,t2)=E{x(t1),y(t2)}=Rxy(t2-t1)

Jointly WSS processes have crosscorrelation


function that depends only on the relative time

Rxy(t1,t2)=E{x(t1),y(t1+τ)}=Rxy(τ)
Uncorrelated Jointly WSS RPs

Jointly WSS RPs are said to be

Uncorrelated: Rxy(τ)=E{x(t),y(t+τ)}= E{x(t)},E{y(t)}

Orthogonal: Rxy(τ)=0 {special case of uncorrelated with


E{x(t)}=0 and/or E{x(t)} =0}

Independent: For any t1 and t2 the RV x(t1) and


y(t2) are independent

Note that independence implies uncorrelated RPs but


not vice versa
White Noise
Continuous Time White Noise
Discrete Time White Noise
Discrete Time White Noise
Sinusoid with Random Phase
Example: Signal Buried in Noise
Assume that we have a model of a measured signal
x(t) = s(t) + n(t)
where s(t) is the desired signal (speech, music, …)
and n(t) is additive noise.

Usually when the generating mechanisms of the two processes


are different the two RPs are assumed to be uncorrelated.

Assume 1) Both s(t) and n(t) are WSS


2) n(t) is white noise
3) s(t) and n(t) are uncorrelated

Find the PSD of x(t)


First find the autocorrelation function of x(t)
Filtering RPs
Passing a RP through an LTI system reshapes
the ACF of the process. Knowing how the
reshaping is done is useful in modeling RPs.
Filtering RPs
To describe the output RP y(t) one looks at its
(i) Mean, (ii) ACF and (iii) PSD.

Comment: WSS in WSS out

Crosscorrelation Function
Filtered DT RP
Filtered DT RP

You might also like