Scribd
Upload
55 views32 pages

Introduction To Signal Processing: Iasonas Kokkinos Ecole Centrale Paris

This document provides an introduction to random signals and stochastic processes. It discusses sources of randomness in signals, such as noise from imaging or thermal noise. Random signals can be analyzed using statistical properties like their power spectrum. Stochastic processes are families of random signals that can be characterized using first and second order statistics like mean, variance, and autocorrelation. Wide-sense stationary processes have constant mean and autocorrelation that depends only on lag. The power spectrum of a process is equal to the discrete-time Fourier transform of its autocorrelation. Linear time-invariant systems transform stochastic processes in ways that can be understood from the systems' impulse response.

Uploaded by

Shivanshu Trivedi
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
55 views32 pages

Introduction To Signal Processing: Iasonas Kokkinos Ecole Centrale Paris

This document provides an introduction to random signals and stochastic processes. It discusses sources of randomness in signals, such as noise from imaging or thermal noise. Random signals can be analyzed using statistical properties like their power spectrum. Stochastic processes are families of random signals that can be characterized using first and second order statistics like mean, variance, and autocorrelation. Wide-sense stationary processes have constant mean and autocorrelation that depends only on lag. The power spectrum of a process is equal to the discrete-time Fourier transform of its autocorrelation. Linear time-invariant systems transform stochastic processes in ways that can be understood from the systems' impulse response.

Uploaded by

Shivanshu Trivedi
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 32

INTRODUCTION TO

SIGNAL PROCESSING
Iasonas Kokkinos
Ecole Centrale Paris
Lecture 7

Introduction to Random Signals

Sources of randomness

Inherent in the signal generation


Noise due to imaging

Prostate MRI Denoised Version


Thermal noise:
Movement of electrons inside resistor in equilibrium

Sources of randomness
Doppler Radar
Send out wave of frequency
Listen to its reflection
Reflected frequency will depend on targets speed
Noise:
Birds, insects (angels)
Sea motion
Wind on trees

How can we decide if a target is present?

Sources of randomness
Speech generation: deterministic system
Complex mechanical models
Turbulence for fricatives
For engineering: simple model + noise
Texture modeling in images

Same feel as natural images


Sufficient for computer vision, maybe not for graphics

Random Signals

Speech signal:

Scatter plots of successive samples

Random Signals

What can we learn from the one signal about the other?

How can we understand if signals are regular?

Random Signals

How can we identify when the signals behavior changes?

Random Signals

How can we remove daily fluctuations?


How can we predict the signal?

Random Signal Analysis

Extract information for understanding & classification


Spectral Estimation
Parametric Modelling
Applications:
Signal Compression/Transmission
Pattern recognition (e.g. speech recognition)
Signal Detection (e.g. presence of sinusoid/target)
Signal Estimation (e.g. frequency of sinusoid/speed)

Random Signal Processing


Denoising
Signal corrupted by noise
Goal: recover signal
Signal: possibly random
Tracking
Dynamical system
Noisy observations of state
Goal: Recover actual state

7th Lecture Layout


Randomness in Signals
Introduction to Stochastic Processes
First and Second order statistics
Power Spectrum

LTI Systems & Stochastic Processes

Stochastic Process examples

Motion of a particle in a liquid


: the particle we chose to observe
Voltage of AC generator with unknown phase
: the signal we get by plugging in

Also known as: random process, random signal

Stochastic Process
Family of signals

Second order distribution:

Density:
N-th order distribution: joint distribution

of

Discrete-time Stochastic Process


: Realization of the random process
For any : x is a discrete-time signal
At any time n: x[n] is a random variable

Description of Stochastic Processes


In general: joint distribution for any order
and any time.
In practice: only a few statistics are used
Mean at time t: expected value of x(t):

Autocorrelation:

7th Lecture Layout


Randomness in Signals
Introduction to Stochastic Processes
First- and Second- order statistics
Power Spectrum

LTI Systems & Stochastic Processes


Spectral Factorization
Normal & Predictable Processes

Discrete-time stochastic process x[n]


Expected Value
Variance
Autocorrelation
Autocovariance

Stationary Processes

Strict-sense-Stationary process:
Joint distribution of any k observations
does not change with time

Wide-sense-stationary (WSS) process:


Constant mean
Autocorrelation
is a function of
Bounded variance

Autocorrelation of WSS processes


Properties

Hermitian Symmetry:

Maximum:

Average Power:

Positive Semidefinite:

Autocorrelation Matrix
Consider

Autocorrelation matrix:

Properties:

Hermitian:
Toeplitz:
Positive Semidefinite:

For any vector

Autocovariance Matrix:
where

WSS White Noise Process v(n)


WSS White Noise:

Zero mean:

Auto-correlation:

Autocorrelation matrix: Diagonal

Observations are uncorrelated

White Gaussian Noise (WGN)


Each sample is independent of other samples
Each sample follows zero-mean Gaussian
distribution

White Gaussian Noise


White noise: sequence of uncorrelated
random variables
WGN: Gaussian variables

WGN

Figure Credit: Manolakis , Ingle & Kogon

Autocorrelation matrix of sinusoid + WGN

Sinusoid signal:
Autocorrelation of noise:
Autocorrelation of x[n]:

Autocorrelation matrix

where

will have the form:

DT-processes x[n], y[n]


x[n] = rainfall on day n
y[n] = #umbrellas sold on day n
Cross-Correlation

Cross-Covariance

Relation between Covariance & Correlation:

Consider: y[n] temperature on day n.

7th Lecture Layout


Randomness in Signals
Introduction to Stochastic Processes
First and Second order statistics
Power Spectrum

LTI Systems & Stochastic Processes

Power Spectrum
Definition of power spectrum of a WSS
stochastic process x[n]:

Wiener-Khintchine theorem:

The power spectrum of a WSS stochastic


process equals the DTFT of its autocorrelation

Power Spectrum of white noise


Autocorrelation function:
Power Spectrum:

Equal power on all frequencies


Just like white Light
Red, pink, violet noises..

7th Lecture Layout


Randomness in Signals
Introduction to Stochastic Processes
First and Second order statistics
Power Spectrum

LTI Systems & Stochastic Processes

Stochastic Processes & LTI Systems

Stationary Processes and LTI systems


WSS process x(n) drives an LTI system:

Mean:

Input-output crosscorrelation:

Stationary Processes and LTI systems

Output autocorrelation:

Power Spectrum:

Systems & Random Signals


System: Deterministic

Input: Stochastic Process


Output: Stochastic Process
Systems Effect:

You might also like