Approximate regularised maximum‐likelihood approach for censoring outliers. Issue 21 (24th September 2019)

• Record Type: Journal Article
• Authors: Han, Sudan; De Maio, Antonio; Pallotta, Luca; Carotenuto, Vincenzo; Iommelli, Salvatore; Huang, Xiaotao
• Is Part Of: Journal of engineering. Volume 2019:Issue 21(2019)
• Page Start: 8061
• Page End: 8065
• Publication Date: 2019-09-24
• Subjects: maximum likelihood estimation -- optimisation -- radar signal processing -- combinatorial mathematics -- communication complexity -- approximation theory -- radar receivers; radar scenario -- combinatorial optimisation problem -- cross‐validation technique -- knowledge‐aided sensor signal processing -- expert reasoning data -- outlier excision methods -- satisfactory performance level -- approximate RML estimation procedure -- censoring outlier index set -- approximate regularised maximum‐likelihood estimation approach -- reduced complexity -- ARML procedure; Engineering -- Periodicals; Engineering; Electronic journals; Periodicals; 620.005
• DOI: 10.1049/joe.2019.0717
• Languages: English
• ISSNs: 2051-3305

Abstract

Abstract : This study considers censoring outliers in a radar scenario with limited sample support. The problem is formulated as obtaining the regularised maximum likelihood (RML) estimate of the outlier index set. Since the RML estimate involves solving a combinatorial optimisation problem, a reduced complexity but approximate RML (ARML) procedure is also devised. As to the selection of the regularisation parameter, the cross‐validation technique is exploited. At the analysis stage, the performance of the RML/ARML procedure is evaluated based both on simulated and challenging knowledge‐aided sensor signal processing and expert reasoning data, also in comparison with some other outlier excision methods available in the open literature. The numerical results highlight that the RML/ARML algorithm achieves a satisfactory performance level in the presence of limited as well as sufficient sample supports whereas the other counterparts often experience a certain performance degradation for the insufficient training volume.