Performance analysis of reduced-dimension subspace signal filtering and detection in sample-starved environment

摘要

For multichannel signal filtering or detection in unknown noise, it is usually difficult to obtain sufficient independent and identically distributed (HD) training data in real-world applications, which considerably degrades the performance of adaptive algorithms. In this paper, we consider the problem of subspace signal filtering and detection in sample-starved environment. A simple reduced-dimension approach is adopted, which alleviates the requirement of HD training data. First, the test and training data are projected onto the signal subspace. Then we adopt the criterion of the generalized likelihood ratio test (GLRT) to devise a detector, which can also serve as a filter. The resulting detector can properly work in sample-starved environment, where the number of HD training data is less than the dimension of the test data. Moreover, the devised approach is superior to the existing adaptive subspace processor in filtering and detection, even in some sample-abundant situations. Analytical expressions for the probabilities of detection and false alarm are derived for the proposed approach. Numerical examples are given to verify its effectiveness. (C) 2018 The Franklin Institute. Published by Elsevier Ltd. All rights reserved.