Mean-level detection in the frequency domain

Conventional mean-level detection (cell averaging CFAR) is applied to frequency-domain data obtained by using the discrete Fourier transform (DFT) as a bank of Doppler filters. It is shown that the scalloping that results from use of the DFT not only results in a loss in SNR, but also causes a limiting, or compression, of the detection probability. This compression is due to the contamination of the mean-level estimate by the presence of significant signal components in all filters, which occurs when the Doppler frequency is not exactly the centre of a Doppler filter. Applying a window (e.g. Hamming) to the data does not alleviate the compression problem. However, it is shown that zero padding the data, which is the conventional approach to avoiding scalloping loss, also prevents compression of the detection probability when a technique called odd-even processing (OEP) is used. Results given show that using zero padding and OEP leads to only a small loss compared with optimum mean-level detection applied in the case when the Doppler frequency is assumed known