FFT filter bank based majority and summation CFAR detectors: a comparative study

The FFT filter bank with CFAR (constant false alarm rate) signal detection is an efficient method for detecting narrowband signals in noise. A common technique for improving detection performance involves the summation of the power spectral information over L successive signal data blocks. This L-block summation detector basically amounts to a form of noncoherent time integration. An alternative approach for processing multiple data blocks is the J-out-of-L detector. While the J-out-of-L detector is known to be sub-optimal for an additive white Gaussian noise channel, it has a more robust false alarm rate performance in the presence of impulsive noise. Consequently, a thorough understanding of the relative performance of the L-block summation and J-out-of-L detectors is useful for selecting the best detector for a given application. The paper presents a comparative performance analysis for Gaussian noise. It shows that: (1) the best performing of the L J-out-of-L detectors is the ([L/2]+1)-out-of-L detector called the L-block majority detector ([x] = integer part of x); (2) the L-block majority detector can approach within 1 dB of the performance of the L-block summation detector.