New detection manifolds for radar signal processing

This paper examines direct data domain (DDD) moving target detection for multi-channel space-based radar (SBR). DDD techniques differ from data-adaptive processing methods in that they do not rely on statistically stationary and homogeneous training data in order to estimate and null out clutter or interference and thereby reveal potential targets; instead, they operate on each range-Doppler cell independently, after any necessary preprocessing to compensate for platform motion and/or array calibration. Prior work examined a maximum-likelihood angle-of-arrival (AOA) technique and the associated target power estimator to detect slow moving targets. In this paper, we extend that methodology and propose novel detection manifolds for optimally partitioning the two-dimensional statistical space. The two dimensions are formed by the passive AOA likelihood function and the estimated target power. A computer simulation of a space radar system and associated geometry is used to assess the characteristics and performance of the new concept. We show that the two-dimensional probability densities corresponding to the target absent and target present hypotheses (H0 and H1, respectively) occupy regions in this space that are not optimally separated by independent thresholds. This immediately leads to the novel detection manifolds (or curved boundaries between detections and non-detections) that improve the detection probability while also reducing the false alarm rate.