Neuro-Adaptive Query-Driven Radar Beamforming

Conventional phased-array processing methods are based on the availability of “signal-free” training data from ranges neighboring each range bin of interest to avoid target suppression by the adaptive beamformer. In complex terrain, clutter inhomogeneity in range precludes such training. In this talk, a new adaptive beamformer is presented integrating a neural network (NN) classifier with blind source separation (BSS). The resulting neuro-adaptive adaptive beamformer is well-suited to inhomogeneous environments since it uses only data from the range bin of interest itself. Moreover, the system can be queried to only output time series corresponding to objects corresponding to specific classes from each range bin of interest. The performance of the beamformer is demonstrated using urban phased-array radar data which can be queried to output time series corresponding to small unmanned aerial systems, birds, pedestrians, or motor vehicles.