Toward the Development of Radar Scene Analyzer for Cognitive Radar

In this paper, we propose a target signature search algorithm to address the radar scene analyzer in explicit terms for cognitive radar reception. A cognitive radar receiver encompasses two main blocks: the radar scene analyzer (RSA) and the Bayesian target tracker (BTT). The BTT needs to have prior knowledge of what it is looking for, such as terrain conditions, and the potential targets in that terrain. This information is provided by the RSA with the aid of other external resources. More formally, the statistical information about the environment and the target are central to the realization of the BTT. However, modeling the statistics of unknown target returns is a particularly challenging task. The objective of the proposed RSA structure is to tackle this challenge by extracting useful information about the target from the environment based on the texture modeling of sea clutter. Specifically, we formulate a weak target as an unknown input embedded in the sea surface, whose dynamics are closely coupled by those of the clutter. The mapping that governs the unknown system's dynamics is assumed to be smooth. The observables from the environment are then predicted one-step ahead with a bank of echo state networks (ESNs). The unknown target's signature is extracted from the ESN prediction error and then refined in two adaptive filtering stages. Performance of the resulting method is evaluated using the posterior Cramer-Rao lower bound (PCRB) on some controlled simulations. Finally, the intended application is presented on live recorded sea returns collected by the McMaster Intelligent PIXel Processing (IPIX) radar. Experiments show that the algorithm can accurately extract the target template from the environment.