Target Detection in
|Daniel E. Kreithen
MIT Lincoln Laboratory
244 Wood Street
Lexington, MA 02173-9108
Abstract Most theoretical work in adaptive beamforming has focused on the case where clutter is assumed to be statistically non-varying over the area of interest. In fact, real-world clutter varies statistically to a degree dependent on both the type of terrain and the training method used for estimating quantities needed for adaptive beamforming. The combined effect of training method and varying clutter gives rise to large clutter discretes which sometimes appear in the post-adaptively nulled data. The classic solution to the post-adaptive beamforming target detection problem uses a sliding window CFAR-type algorithm, either in range only, or in range and Doppler. Many variations of the common cell averaging CFAR algorithm, including order-statistic techniques, exist in order to mitigate the effects of the large clutter discretes left in such adaptively beamformed data. All of these CFAR techniques ignore a basic ingredient contained in the data-the deterministic ground clutter azimuth-Doppler relationship.
We propose and illustrate one method of incorporating this deterministic relationship into a detection algorithm. The method involves measuring the strength of the ground clutter in each range/Doppler bin and modifying that quantity by an estimate of the system nulling effect which is applied to that range/Doppler bin during the adaptive beamforming processing. The resulting range/Doppler map has large clutter discretes which are correlated with large clutter discretes contained in the post-adaptively nulled range/Doppler maps. Analysis of this correlation is performed by means of two-dimensional scatterplots. Additionally, we observe that targets tend to be uncorrelated between these two data sets, demonstrating a separation between targets and clutter discretes on the two-dimensional scatterplot. We propose a new detection algorithm based on this observation. We also model both target and clutter distributions, and propose a simple method for approximating theoretically optimal threshold curves.
Direct comments and questions to: firstname.lastname@example.org
© MIT Lincoln Laboratory. All rights reserved.