This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Detection and Estimation (Spring, 2009) Bayesian Estimators NCTU EE P.1 Bayesian Estimators 9 Now we assume we have some prior knowledge about θ . To incorporate it, we assume θ is random variable with a given pdf. 9 Bayesian estimation is useful in situations where an MVU estimator cannot be found. ~ Prior Knowledge and Estimation Example: DC Level in WGN, but –A ≤ A ≤ A instead of  ∞ ≤ A ≤ ∞ . MVU estimator: x A = ˆ Truncated sample mean estimator: A ( is better than A ˆ in terms of MSE. We reduce the mean square error by allowing the estimator to be biased! Detection & Estimation (Spring, 2009) Bayesian Estimators NCTU EE P.2 Bayesian Approach: A is considered to be a random variable with a prior pdf. We attempt to estimate the realization of A. e.g., Assume A ~ U [A ,A ]. Classical: Bayesian: Remark: MSE depends on A. BMSE doesn’t. Since p( x ) ≥ 0 for all x , if the integral in brackets can be minimized for each x , then the Bayesian MSE will be minimized. Detection & Estimation (Spring, 2009) Bayesian Estimators NCTU EE P.3 Remarks: 9 The optimal estimator in terms of minimizing the Bayesian MSE is the mean of the posterior pdf p(A x ). 9 : the prior pdf of A 9 The estimator that minimizes the Bayesian MSE is termed the minimum mean square error (MMSE) estimator . Example: p(A) = U [A ,A ]. Assume w[n] is independent of A. where Detection & Estimation (Spring, 2009) Bayesian Estimators NCTU EE P.4 The MMSE estimator: Cannot be evaluated in closed form. Remarks: 9 A ˆ is biased towards zero unless N A 2 σ >> . In general, A ˆ is biased towards the prior mean. 9 As N increases, the MMSE estimator relies less and less on the prior knowledge and more on the data. 9 Before observation, we assume a prior pdf p( θ ). After observation, our state of knowledge about the parameter is summarized by the posterior pdf p( θ  x ). 9 The choice of a prior pdf is critical in Bayesian estimation. A wrong choice will result in a poor estimator. 9 An optimal estimator is defined to be the one that minimizes the MSE when averaged over all realizations of θ and x . MMSE estimator Detection & Estimation (Spring, 2009) Bayesian Estimators NCTU EE P.5 ~ Choosing a Prior PDF 9 Should be based on physical constraints of the problem. 9 Choose prior to allow easy integration. Example: DC Level in WGN – Gaussian Prior pdf Detection & Estimation (Spring, 2009) Bayesian Estimators NCTU EE P.6 Let The posterior pdf is also Gaussian, but with different mean and variance. where Remark: α is a weighting factor (0 < α < 1) that compromise between the prior knowledge and the data knowledge. Detection & Estimation (Spring, 2009) Bayesian Estimators NCTU EE P.7 The prior knowledge improves the estimation accuracy....
View
Full Document
 Spring '09
 ShengJyhWang
 Estimation theory, WGN, Bayesian Estimators, NCTU EE, Bayesian Linear Model

Click to edit the document details