Expected likelihood for temporally correlated (over-sampled) training data: Experiment results

In this study, we analyse sensitivity of the likelihood ratio (LR) p.d.f with respect to the residual temporal correlation of antenna training samples, caused by non-rectangular bandpass filters and over-sampling. This temporal correlation causes deviation of the actual LR p.d.f. from the theoretical one derived for independent (Gaussian) training samples. This in turn affects performance of “expected likelihood” detection-estimation techniques. This paper outlines methodologies for correcting for this temporal correlation, allowing for the application of expected likelihood, and validates the methods using multichannel receiver data collected by the Australian OTHR JORN facility.