Optimal waveform selection for target classification

This paper studies the design of a set of outgoing radar signals to discriminate between two target classes. We model the reflectivity function of each target by a two-dimensional stochastic process to account for uncertainties and propagation effects. The signals are selected to minimize the expected number of transmissions that are needed to guarantee a given confidence level in the classification decision. We argue that this goal can be achieved by selecting the signals that maximize the Kullback-Liebler information number between the two target classes. We illustrate our approach with a particular model. We show that for this model, the optimal set of waveforms can be designed off-line and depends on both the statistics of the reflectivity functions of the targets in both classes and the observation noise level.