Sequential order selection for real time signal processing

In this paper, we address the problem of order selection in signal processing applications for the general case where the observations can have correlations in time and require sequential processing. We consider the finite normal mixtures (FNM) model as an example for which correct order selection is very important. We derive penalized partial likelihood as the information theoretic criterion for order selection for the general case where observations can have correlations. We propose a sequential order selection procedure and investigate its properties and give a number of examples in channel equalization showing the effectiveness of the approach