0% found this document useful (0 votes)
214 views

Detec%on and Es%ma%on Theory: Class Notes Ell 719

This document provides an outline for a course on Detection and Estimation Theory. It discusses key topics that will be covered, including binary and multiple hypothesis testing, detection of continuous time signals, estimation theory fundamentals, and properties of estimators. Example applications are presented, such as communications, radar, sonar, and estimation problems involving signals with unknown parameters in noise. The goals of the course are also stated as formulating and solving various detection and estimation problems using statistical tools and designing optimal processors and estimators.

Uploaded by

Neelkanth Kundu
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
214 views

Detec%on and Es%ma%on Theory: Class Notes Ell 719

This document provides an outline for a course on Detection and Estimation Theory. It discusses key topics that will be covered, including binary and multiple hypothesis testing, detection of continuous time signals, estimation theory fundamentals, and properties of estimators. Example applications are presented, such as communications, radar, sonar, and estimation problems involving signals with unknown parameters in noise. The goals of the course are also stated as formulating and solving various detection and estimation problems using statistical tools and designing optimal processors and estimators.

Uploaded by

Neelkanth Kundu
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 29

Detec%on

 and  Es%ma%on  Theory  

Class  Notes  
ELL  719  
ELL  719:  Detec%on  &  Es%ma%on  Theory  
Tenta%ve  Course  Outline  (subject  to  change)  

•  Introduc)on:  Structure  of  detec%on  &  es%ma%on  problems.  


•  Simple  Binary  Hypothesis  Tes)ng:  Decision  Criteria;  Maximum  Likelihood,  Neyman-­‐Pearson,  
Bayes  risk,  Probability  of  Error  and  Min-­‐Max  Criteria;  Receiver  Opera%ng  Characteris%cs;  
Single  and  Mul%ple  Observa%ons  cases;  Composite  Hypothesis  Tes%ng.  
•  M’ary  Hypotheses  or  Mul)ple  Decision  Theory:  Mul%ple  decisions;  Bayes  risk;  Probability  of  
Error;  General  and  Gaussian  cases;  
•  Detec)on  of  Con)nuous  Time  Signals:  Detec%on  of  Signals  in  Addi%ve  White  Gaussian  
Noise;  Detec%on  in  Non-­‐white  Gaussian  Noise;  Signals  with  Unwanted  Parameters  (Random  
phase,  Random  amplitude  and  phase).  
•  Es)ma)on  Theory  Fundamentals:  Maximum  Likelihood,  Bayes  Cost  and  Minimum  Variance  
Es%mators;  Rela%onship  of  Es%mators.  
•  Es)ma)on  with  Gaussian  Noise:  Linear  Observa%ons;  Sequen%al  Observa%ons;  Non-­‐linear  
Es%ma%on;  E-­‐M  algorithm;  State  Es%ma%on  and  Kalman  Filtering.  
•  Proper)es  of  Es)mators:  Unbiased  Es%mators;  Efficient  Es%mators;  Asympto%c  Proper%es.  
•  Main  Text:  Detec%on,  Es%ma%on  &  Modula%on  Theory-­‐Part  1:  Harry  L.Van  Trees,  John  
Wiley;  Paperback  reprint,  2003.  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   2  


Background:  Pre-­‐requisite  

•  Signal  Theory    
 Probability  Theory,  Random  Variables  and  Random  Processes  

•  Signals  and  Systems  


 Representa%on  of  Signals  and  Systems.  Fourier  Series  and  
 Transforms  and  Related  Topics.  

•  Basic  Communica%on  Theory.  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   3  


Learning  Outcomes  

•  Be  able  to  formulate  different  kinds  of  Signal  Detec%on  


Problems.  
•  Be  able  to  use  sta%s%cal  and  mathema%cal  tools  to  solve  a  
hierarchy  of  Detec%on  Problems  and  design  op%mum  
processors.  
•  Be  able  to  formulate  different  kinds  of  parameter  and  signal  
es%ma%on  problems.  
•  Be  able  to  solve  basic  problems  using  sta%s%cal  and  
mathema%cal  tools  and  design  op%mal  es%mators.  
 

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   4  


Introduc%on:  Detec%on  Theory  
•  Deals  with  detec%ng  the  presence  of  useful  signals  in  the  
presence  of  noise  and  interference.  

•  Applica%ons:  Digital  Communica%ons,  Radar,  Sonar,  


Radio-­‐astronomy;  many  others.  

•  Broadly,  we  may  look  for  


–  Known  signals.  
–  Par%ally  known  signals,  or    
–  Completely  unknown  and  Random  signals  
in  noise.  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   5  


Examples  
•  Known  Signals  in  Noise  Problems  
–  Digital  Communica%on  via  FSK:  
 
s1 (t) = sin ω1t,s0 (t) = sin ω 0 t,0 ≤ t ≤ T
   
•  Transmieed  over  a  communica%on  channel,  which  yields  
an  undistorted  replica,  with  thermal  noise:  
 
r(t) = s1 (t ) + n(t),or
 
  r(t) = s0 (t ) + n(t),0 ≤ t ≤ T
•  Receiver:  RF  amplifier  +  Processor  to  decide  which  of  the  
two  signals  was  transmieed  over  the  %me  interval.  
 
•  Detec%on  Theory  Problem:  Design  and  Evaluate  the  
Processor  
2018     Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   6  
Signals  with  Unknown  Parameters  in  Noise  

•  FSK  problem  with  a  random  phase  drii  in  oscillator  or  


medium:  
r( t) = sin(ω1t + θ 1 ) + n( t),0 ≤ t ≤ T
r( t) = sin(ω 0 t + θ 0 ) + n( t),0 ≤ t ≤ T

–  θ1  and  θ0:unknown,  but  constant  phase  angles.  


€ –  Signal  not  completely  known  even  in  absence  of  noise.  
–  This  uncertainty  must  be  taken  into  account  while  
designing  the  detector.  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   7  


Signals  with  Unknown  Parameters  in  Noise  
•  Ac%ve  Radar/Sonar  Problem:  

•  Tx.  Signal:   s(t) = sin ω c t,0 ≤ t ≤ T


 
•  Received  signal  when  target  present:  

€r(t) = Vr sin(ω c (t − τ ) + θ r ) + n(t),τ ≤ t ≤ τ + T


  = n(t),0 ≤ t ≤ τ ,τ + T < t ≤ ∞.
•  When  target  absent:  

r(t) = n(t),0 ≤ t ≤ ∞

•  Three  
€ unknown  parameters  even  in  the  absence  of  noise.  
   

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   8  
Random  Signals  in  Noise  

•  Passive  sonar  detec%on:  Receiver  listens  for  noise  generated  


by  enemy  submarines:  Engines,  Propellers  and  other  
machinery.  

•  Acous%c  signals  travel  through  ocean  to  hydrophones  in  the  


detec%on  system.  

•  Desired  signal:  Itself  a  sample  func%on  of  a  random  process.  

•  Hydrophone  generates  self-­‐noise  and  picks  up  sea  noise.  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   9  


Random  Signals  in  Noise  

Passive  Sonar  Detec%on  Problem  


r(t) = sΩ (t) + n(t) :Signal  present  

r(t) = n(t) :Signal  absent  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   10  


Random  Signals  in  Noise  

•  Troposcaeer/Underwater  Acous%c/Fading  Channel  


Communica%on  Problem:  
•  Transmieed  signals:  s1(t)  and  s0(t),  as  in  FSK.  
•  Received  signals:  
r(t) = sΩ1 (t) + n(t) :when  s1(t)  transmieed  

r(t) = sΩ 0 (t) + n(t) :when  s0(t)  transmieed  

•  sΩ1(t)  and  sΩ0(t):  sample  func%ons  of  random  processes  


centred  around  ω1  and  ω0    respec%vely.  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   11  


Random  Signals  in  Noise  

•  In  such  cases,  the  signals  to  be  detected  lack  a  determinis%c  


component.  

•  Detector  Design:  Must  be  based  on  the  difference  in  the  
sta%s%cal  proper%es  of  the  random  processes  of  which  sΩ1(t)  
and  sΩ0(t)  are  sample  func%ons.  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   12  


Mul%ple  Decision  Problems  

•  M’ary  digital  modula%ons.  

•  Classifica%on  problems  from  observed  features  data  etc.  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   13  


Hierarchy:  3  levels  

Known  signals  in   •  Synchronous  digital  communica%ons  


noise   •  Paeern  recogni%on  problems  

Signals  with   •  Ac%ve  radar  and  sonar;  target  classifica%on    


unknown   •  Digital  communica%on  systems  without  phase  
reference  or  over  slowly  fading  channels  
parameters  

•  Digital  comm.  Over  scaeer  links;  Passive  sonar  


Random  Signals   detec%on.  
in  noise   •  Seismic  detec%on  systems;  radio-­‐astronomy  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   14  


Classical  Detec%on  Theory  
Defini%ons  

•  Components  of  Decision  Theory  Problems:  


 
 
                                   H1   Probabilis%c  
         Source   Transi%on   Observa%on  
space  
                                   H0   Mechanism  
   
 
 
Decision  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   16  


Components  of  Decision  Theory  Problems  

•  Source:  generates  an  output:  One  of  two  (or  M)  choices.  

•  Digital  Communica%ons:  1’s  and  0’s  of  informa%on  to  be  


transmieed.  
•  Medical  ECG:  H1:  Pa%ent  had  a  heart  aeack;  H0:  absent;  
•  Speaker  Classifica%on:  Speaker  German,  Bri%sh  or  American;  
Male  or  Female;  (6  hypotheses).  
•  Radar/Sonar:  H1:Target  present;  H0:  Target  absent,  in  a  given  
range-­‐azimuth  cell.  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   17  


Components  of  Decision  Theory  Problems  
•  Probabilis)c  Transi)on  Mechanism:  device  that  knows  the  source  output,  
but  modifies  it  to  generate  a  point  in  a  suitable  observa)on  space  
according  to  a  probability  law.  

•  H1  +1            n  
                                                                                                                                                                                 r    Observa)on  space:  
+  
•  H0  -­‐1                  one-­‐dimensional  
 Transi%on  Mechanism  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   18  


Example  1.1  

•  H1:  r  =  1  +  n;    H0:  r  =  -­‐1+n  


•  Probability  densi%es  of  r  on  the  two  hypotheses:  
 

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   19  


Example  1.1  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   20  


Example  1.2  

•  Source  generates  two  numbers  in  a  sequence.  

H1 : r1 =1+ n1;r2 =1+ n2


H0 :r1 = −1+ n1;r2 = −1+ n2
•  Observa%on  Space:  2-­‐dimensional.  



2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   21  
Example  1.2  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   22  


Example  1.2  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   23  


Decision  Rule  

•  Decision  Rule:  Guesses  which  hypotheses  is  true:  Assigns  each  


point  to  one  of  the  hypotheses.  
 
•  Decision  Rule  mapping:  Subject  of  Detec%on  Theory.  

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   24  


Simple  Binary  Hypothesis  Tests  

•  Let  observa%on  space  correspond  to  N  observa%ons:  


(r1,r2,...,rN )
•  Each  observa%on:  A  point  in  an  N-­‐dimensional  space,  and  
denoted  by  r.  
•  R  is  generated  either  under  H0  or  H1,  with  a-­‐priori  
probabili%es  P0  and  P1,  and  known  condi%onal  probabili%es    
   pr / H  (R /H   0 )  and   pr /  H (R  /H1)  respec%vely.  
0 1

•  Use  this  informa%on  to  develop  a  suitable  decision  rule.  

€ €
2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   25  
Decision  Criteria  

•  (i)  Bayes  Criterion:  Assumes  knowledge  of  P0  and  P1;  


•  Each  of  the  4  decisions  assigned  a  cost:  C00,C10,C11,C01.  
•  The  average  cost  or  risk  is  defined  as  

ℜ = C00 P0 Pr(sayH 0 / H 0 true)


+C10 P0 Pr(sayH1 / H 0 true)
+C11P1 Pr(sayH

1 / H1true)

+C01P1 Pr(sayH 0 / H1true)

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   26  


Decision  Rule:  Bayes  Criterion  

•  Decision  Rule  must  say  either  H1  or  H0,  based  on  the  observed  
vector.  
 :  Rule  for  dividing  the  total  observa%on  space  Z  into  two  
parts:  Z0  and  Z1.  
 

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   27  


Decision  Rule:  Bayes  Criterion  

ℜ = C00 P0 ∫ pr / H 0 (R /H 0 )dR
Z0

+C10 P0 ∫ pr / H 0 (R /H 0 )dR
Z1

+C11P1 ∫ pr / H1 (R /H1 )dR


Z1

+C01P1 ∫ pr / H1 (R /H1 )dR


Z0

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   28  


Decision  Rule:  Bayes  Test  

•  It  is  natural  to  assume  that  the  cost  of  wrong  decisions  will  be  
larger  than  the  corresponding  correct  decision:  

•  i.e.,    C10  >  C00  and  C01  >  C11.  

•  Bayes  Test:  Choose  the  decision  regions  Z0  and  Z1  such  that  
the  average  risk  R  is  minimized.  
 

2018   Detec%on  &  Es%ma%on,  SP,  IIT  Delhi   29  

You might also like