Development of a kernel for the detection of non stationary signals

Abstract

In many signal detection applications, probabilistic methods are used to model and detect signals in order to account for variances within the signal class. However, probabilistic methods require training data in order to operate adequately, which is not always readily available. This paper presents a method of detecting probabilistic signals using a single training example to form a kernel matrix for use within a compact subspace detector. Results indicate that this method performs well in detecting chirp signals, providing a proof of concept for detection of probabilistic signals using a single training example.