(A) study on radar detection algorithms in heterogeneous noise environment

Abstract

In single input single output (SISO) radar detection, the constant false alarm rate (CFAR) detection methods have been studied during decades. The most famous ones are the cell-average (CA) CFAR and the order statistic (OS) CFAR. The major difference between them is how to estimate the statistics of background noise. CA-CFAR uses an average value obtained from the signal in the reference window. The performance of CA-CFAR degrades under the non-stationary environment such as clutter power transition. The greatest of (GO) CFAR and the smallest of (SO) CFAR are proposed to solve the problem of performance degradation in the non-stationary background noise despite the performance loss of detection. OS-CFAR uses the kth largest sample as a statistic instead of an average value used in CA-CFAR. It yields a better performance than CA-CFAR when there exist multi-targets in the reference window.

As the numbers of transmitter and receiver antenna increase, the detection performance of the radar system also increases. When there is two or more transmit antenna and receive antenna, it is defined as multiple input multiple output (MIMO) radar. The generalized likelihood ratio test (GLRT) is widely used in target detection problems in MIMO radar systems. GLRT is derived using the probability density function (pdf) of the noise signal in a homogeneous noise environment.

In a non-homogeneous noise environment, however, the received signal is represented as a summation of several signals with different distributions such that it is difficult to estimate the pdf of the received signal. In addition, GLRT requires knowledge of the number of received signals which contain non-homogeneous noise components, which is not easy to obtain from the received signals. In this thesis, we propose an adaptive selection method for GLRT (ASMGLRT) that overcomes the noise non-homogeneity problem in the MIMO radars. The proposed algorithm estimates the noise statistics via the received signals which contain homogeneous noises only, instead of using all the received signals in a reference window. Furthermore, it does not require information on the pdfs of the non-homogeneous noises and the number of received signals containing non-homogeneous noises when the received signal has both homogeneous and non-homogeneous noises.

In the heterogeneous noise environment, GLRT experiences performance degradation due to clutter returns, so that an ASMGLRT was proposed to reject the heterogeneous noise by calculating GLRT variables with homogeneous noise only when at least one virtual path with homogeneous noise only in the received signals exists. However, in the severe clutter environment such as sea clutter, the assumption of receiving one virtual path with homogeneous noise only might not be met. Therefore, we also propose a variable length-adaptive selection method for GLRT (VL-ASMGLRT) to settle the heterogeneous noise problem in the severe clutter environment. The proposed algorithm estimates the noise statistics by selecting homogeneous noise only in the received signal, therefore the noise statistic is a more robust and effective solution to detect a target in the severe clutter environment. The performance of the proposed algorithm is obtained in terms of detection and false alarm probabilities, and is compared to GLRT and ASMGLRT in both homogeneous and heterogeneous noise environments.