Scribd
Upload
12 views

Lecture 1 - Introduction To Det-Est - Annotated

This document discusses parameter estimation and how to evaluate estimators. It provides an example of estimating an unknown DC level (parameter Θ) from a data set with noise. Two estimators are examined: taking the mean of all data points (case 1) and using a single data point (case 2). Case 1 has smaller variance and performs better since it uses more data, making it a minimum variance unbiased estimator. In summary, an estimator's performance is described statistically by its mean and variance, with the goal being a minimum variance unbiased estimator.

Uploaded by

Paras Vekariya
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
12 views

Lecture 1 - Introduction To Det-Est - Annotated

This document discusses parameter estimation and how to evaluate estimators. It provides an example of estimating an unknown DC level (parameter Θ) from a data set with noise. Two estimators are examined: taking the mean of all data points (case 1) and using a single data point (case 2). Case 1 has smaller variance and performs better since it uses more data, making it a minimum variance unbiased estimator. In summary, an estimator's performance is described statistically by its mean and variance, with the goal being a minimum variance unbiased estimator.

Uploaded by

Paras Vekariya
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 14

Estimation Theory

Ch-1: Introduction

Steven M. Kay book


Example: Estimation of parameter
• Determine range R
More Application Areas
Parameter Estimation
• Extracting values of parameter(s) from a discrete data set

• N point data set

depends on an unknown parameter Θ

• Where g is some function


Mathematical Estimation Problem
How good/bad an estimator is ?

• We could model data as (DC level in WGN noise)

• Estimate the DC level 𝐴 from data 𝒙


• Case1: Estimate of A can be
• Case2:

• Will it do better than the earlier?

• Check the mean and variance of each estimator under case 1 and 2.
• Assume zero mean and uncorrelated noise
• First one should perform better as it makes use of all data
• Although is very close to true value 1
• Note: We need to average out over multiple realizations to have
better idea

• Plot the histogram


• No of times estimator produces a given range of values  can be treated as PDF
• Sample mean is better as the values obtained are more close to the true value 1
(i.e., variance is small)
What do we learn?
• An estimator is a random variable
• Its performance can be completely described statistically (by its PDF)

• 1. Mean of estimator  Actual unknown parameter


• 2. Variance of estimator should be as minimum as possible

1 and 2 together  Minimum variance unbiased estimator

You might also like