E1 244: Detection and Estimation
Lecture 1: Introduction
Course information
I Instructor:
– Sundeep Prabhakar Chepuri.
Email: spchepuri AT iisc.ac.in
I Class schedule:
– Tuesdays and Thursdays 2.00-3.30pm (Online via MS Teams).
I Course webpage:
https://ece.iisc.ac.in/∼spchepuri/classes/e1244.html
2
Course information
I Teaching Assistants
Sravanthi Gurugubelli Prasobh Sankar
sravanthig prasobhr
I Exercise/tutorial sessions:
– By TAs, on 1st and 3rd Saturdays 11.00-12.30 pm
3
Textbooks
I Fundamentals of Statistical Signal Processing, Volume I: Estimation
Theory, S.M. Kay, Prentice Hall 1993, ISBN-13: 978-0133457117.
I Fundamentals of Statistical Signal Processing, Volume II: Detection
Theory, S.M. Kay, Prentice 1993, ISBN-13: 978-0135041352.
4
Other resources
I Statistical Signal Processing, L.L. Scharf, Pearson India, 2010,
ISBN-13: 978-8131733615.
I An Introduction to Signal Detection and Estimation, H.V. Poor,
Springer, 2nd edition, 1998, ISBN-13: 978-0387941738.
5
Grading and course requirements
I Prerequisite: Matrix theory (or equivalent) and random processes
(or equivalent).
I Three assigments (problem and programming set): 10% each, i.e.,
30% in total
I Midterm exam: 20%
– Open book exam.
I Project: 30%
I Final exam: 20%
– Open book exam.
6
Estimation theory
Time delay estimation or ranging
7
Estimation theory
Sensor localization and positioning
8
Estimation theory
9
Estimation theory
Denoising in hearing aids
10
A simplified model
How to determine ŝ - an estimate of s?
11
A simplified model
How to determine ŝ - an estimate of s?
I ŝ1 = y1 ?
PN
I ŝ2 = N1 i=1 yi ?
How good are these estimators? Are there better estimators?
12
A simplified model
How to determine ŝ - an estimate of s?
I ŝ1 = y1 ?
PN
I ŝ2 = N1 i=1 yi ?
Suppose noise at each microphone, ni , has variance σi2 . Then, how
about:
PN yi
i=1 σ 2
I ŝ3 = PN i
1 ?
i=1 σ 2
i
What is the recipe to determine optimal estimators?
13
Estimation theory
Typical formulation in estimation theory
y[n] = fn (θ) + w[n].
The noise is usually assumed to stochastic, the parameter vector of
interest θ may be
I an unknown deterministic quantity: classical estimation theory.
I an unknown random quantity: Bayesian estimation theory.
14
Detection theory
Radar - target detection
Delay/ranging might not be always required
15
Detection theory
Cognitive radio - spectrum sensing
16
Detection theory
How to choose γ?
17
Detection theory
Typical detection problem formulation, also referred to as binary
hypothesis testing problem:
(no target) H0 :x[n] = w[n]
(target) H1 :x[n] = s[n] + w[n]
We wish to infer the state of nature, i.e., to decide on H0 or H1 using a
detector of the form:
T (x[n]) > γ
I How to make an optimal decision: how to choose T (·) and γ?
I detection of deterministic signals: Neyman-Pearson detectors.
I detection with priors: Bayes detectors.
I detection with unknown parameters.
18
Content
I Review of linear algebra and random processes.
I Minimum variance unbiased estimator
I Cramér-Rao bound
I Maximum likelihood estimator
I Best linear unbiased estimator
I Least squares and recursive least squares.
I Structured covariance estimation
I Bayesian estimators (MMSE and MAP estimators)
I Kalman filtering
I Neyman-Pearson detector
I Bayes detector
I Multiple hypothesis testing
I Composite hypothesis testing
I Sequential probability ratio test (SPRT)
19
Course objectives
Generally, a solution to estimation and detection problems depend on the
underlying data model and the statistical description of the noise and/or
unknowns.
I How to mathematically formulate such problems?
I Determine optimal estimators, characterize performance of these
estimators, and compute the estimation bound.
I Determine optimal detectors and characterize the performance of
these detectors.
20
