Many time critical data processors use simple and computationally inexpensive, but highly suboptimal, prefilters to separate relevant from irrelevant data. Many of these approaches were traditionally driven by lack of processing resources. For example, radars that operate in dense or high clutter environments must separate legitimate potentially maneuvering target reflections from background noise. A common technique for separating clutter from target reflections is to preprocess the radar data by finding sequences of returns across multiple scans or detection frames that appear to come from a moving, but non-accelerating, target. Radar measurements that pass these multiple scan tests are then passed to the main tracking algorithm; those that do not are identified as clutter. We propose to develop new radar signal preprocessing, and clutter identification and rejection algorithms that increase the number of nonclutter measurements available for local active sensor tracking of maneuvering targets, while reducing the probability of false detections. We aim to utilize modern FPGA's to implement neural network processing elements to process multiple parameters of raw radar detections in a way that greatly reduces the effects of a wide beamwidth, high sidelobe antennas (i.e. E2C AEW radar). Additionally, we will leverage the curve fitting capabilities of neural networks to determine if a non-correlated new hit could possibly obey the maximum dynamic turn radius at the target velocity and treat it as a maneuvering target
