Signal Detection in Correlated Gaussian and Non-Gaussian Radar Clutter

A machine learning approach to distribution identification in non-Gaussian clutter

“…Many of the non-Gaussian distributions that have been empirically fit to measured clutter data belong to the class of SIRVs (e.g. the K distribution and the Weibull distribution for certain values of the shape parameter) [6], [7], [9], [13].…”

“…While not admissable as a SIRV [7], it has been suggested that the log-normal distribution is a good fit to measured clutter [14], [15]. Multivariate, correlated log-normal clutter may be generated by taking the complex exponential of zero mean, complex Gaussian random vectors [15].…”

“…Here we extend the work of [6], [7] to a general machine learning framework for distribution, and ultimately threshold, estimation. Current radar signal processing research emphasizes the role of knowledge-aided and cognitive approaches to future radar systems.…”

“…Therefore, to establish a reliable detection threshold the true distribution of the clutter must be identified and any relevant parameters must be estimated. Previous work proposed a graphically-based approach to distribution identification [6], [7].…”

Stochastic Resonance In Signal Detection And Human Perception

“…For the endogenous model, the desired non-Gaussian PDF and correlation function is realized using a zero memory, non-linear transformation on a real Gaussian process. In this approach, however, it is not possible to control both the PDF and the correlation independently [Rangaswamy, 1993]. Further, the nonlinear transformation may give rise to non-Gaussian processes with a non-negative definite covariance matrix.…”