Skip to Main Content
Many current radar-based human detection systems employ some type of Doppler or Fourier-based processing, followed by spectrogram and gait analysis to classify detected targets. However, Fourier-based techniques inherently assume a linear variation in target phase over the aperture, whereas human targets have a highly nonlinear phase history. This mismatch leads to significant loss in SNR and integration gain. In this paper, two novel human-modeling based non-linear phase detectors are presented. The first (ONLP) computes maximum likelihood estimates of unknown parameters of a model of the human torso response, while the second (EnONLP) stores the expected returns of a 12-point model for each combination of model parameter values in a dictionary and uses orthogonal matching pursuit to find the optimal sparse approximation to the data. The performance of ONLP, EnONLP, and conventional STAP is compared and application to target characterization discussed.