Two generalized greatest of selection CFAR algorithms

The conventional cell-averaging (CA) CFAR detector gives a minimum loss of detection power in homogeneous background, but it exhibits intolerable rise in false alarm probability during abrupt transitions of clutter power levels and significant degradation in detection probability in the presence of interfering targets. The ordered statistics (OS) CFAR processor may resolve multiple targets quite well, but it lacks effectiveness in preventing excessive false alarms during clutter power transitions and its sample sorting time is too long. Therefore, some algorithms such as OSGO and GOSGO are proposed. The advantage of applying greatest of selection logic lies in controlling the rise of false alarms at clutter edges. If several ordered samples are linearly combined to give the noise power estimation for the leading or lagging sub-window, better detection performance can be made than with a single ordered sample as considered in OSGO. In this paper, two new greatest of CFAR schemes (TMGO and QBWGO) are proposed. They split the reference window into two sub-windows and uses TM or QBW methods to create two local noise power level estimations, the greatest value of them is used to set an adaptive threshold. It is shown that the detection performance of TMGO and QBWGO is superior to that of OSGO both in homogeneous background and multiple targets situations