Marginal likelihood for estimation and detection theory

Yair Noam, Joseph Tabrikian

Research output: Contribution to journalArticlepeer-review

10 Scopus citations

Abstract

This paper derives and analyzes the asymptotic performances of the maximum-likelihood (ML) estimator and the generalized likelihood ratio test (GLRT) derived under the assumption of independent identically distribution (i.i.d.) samples, where in the actual model the signal samples are m-dependent. The ML and GLRT under such a modeling mismatch are based on the marginal likelihood function, and they are referred to as marginal maximum likelihood (MML) and "generalized (sum) marginal log-likelihood ratio test"(GMLRT), respectively. Under some regularity conditions, the asymptotical distributions of the MML and GMLRT are derived. The asymptotical distributions in some signal processing examples are analyzed. Simulation results support the theory via several examples.

Original languageEnglish
Pages (from-to)3963-3974
Number of pages12
JournalIEEE Transactions on Signal Processing
Volume55
Issue number8
DOIs
StatePublished - 1 Aug 2007

Keywords

  • Asymptotic properties
  • Consistency
  • Detection
  • Estimation
  • Generalized (sum) marginal log-likelihood ratio test (GMLRT)
  • Marginal likelihood
  • Marginal maximum likelihood (MML)
  • m-dependent

Fingerprint

Dive into the research topics of 'Marginal likelihood for estimation and detection theory'. Together they form a unique fingerprint.

Cite this