Contact Us
CRLB for Likelihood Functions with Parameter-Dependent Support and a New Bound

CRLB for Likelihood Functions with Parameter-Dependent Support and a New Bound

Wednesday 15/02/2017
  • Yaakov Bar-Shalom
  • Classroom 165, ground floor, Library, Aerospace Eng.
  • University of Connecticut
  • The talk will be given in English

In this talk we discuss the regularity conditions required for the classical Cram´er-Rao lower bound (CRLB) for real-valued (nonrandom unknown) parameters to hold. It is shown that the commonly assumed requirement that the support of the likelihood function (LF) should be independent of the parameter to be estimated can be replaced by the much weaker requirement that the LF is continuous at the end points of its support.

Parameter-dependent support of the LF arises when an unknown parameter is observed in the presence of additive measurement noise and the measurement noise probability density function (pdf) has a finite support. It is also pointed out that the commonly cited requirements of absolute integrability of the first two derivatives of the LF should be replaced by requirements on the log-LF (LLF). Some practical examples of finite-support measurement noises, which lead to parameter-dependent LF support, are discussed in light of the above. For the case where the LF is not continuous at the end points of its support, a new modified CRLB — designated as the Cram´er-Rao-Leibniz lower bound (CRLLB), since it relies on Leibniz integral rule — is presented and its use illustrated. The CRLLB is shown to provide valid bounds for a number of long-standing problems for which the CRLB was shown in the literature as not valid, in particular, for a uniformly distributed measurement noise.

Light refreshments will be served before the lecture
For more info
Please fill in the details

"*" indicates required fields

This field is for validation purposes and should be left unchanged.
*required fields are marked with an asterisk