Improve CFAR Algorithm Based on Closed Loop by Neural Network

Abstract

In the Constant False Alarm Rate (CFAR) processor, several algorithms can be used to decide the target in the Cell under Test (CUT) in the detection process stage at the receiver side. Since all these algorithms are considered an open loop processor, continually their performance accuracy with environmental changes cannot be guaranteed. This paper presents a Closed Loop CFAR (CL-CFAR) processor, as a proposed new CFAR, to guarantee the continuity of their performance. A shift register is used to save the decision of each cell after threshold CUT as a pattern, then a neural network (NN) back propagation is used to recognize this pattern, which represents the state of the lagging window. After that, the output of the NN is back to the return signal classifier, which is responsible for selecting the optimal CFAR, which is used. Where the proposed closed loop CFAR is used for switching between certain CFAR algorithms, the switching is based on the changing the field environment. The results show over perform of the closed loop compared with conventional algorithms. It showed for a single target the probability of detection PD is 90–97% with Pfa from 10-4 to 10-8 by using the selected CA-CFAR. Further, for Multi-target 100% with the same Pfa using selected OSGO-CFAR and for a closed multi-target, the PD is 94–100% with the same Pfa with selected OSSO-CFAR, also for clutter edge situation the PD is 94–98% with the same Pfa with selected OSSO-CFAR. The probability of detection of the proposed closed loop-CFAR is 96% in different and changeable environments.