Data-Driven Optimization of Time and Frequency Resolution for Radar Transmitter Identification

摘要

An entirely new set of criteria for the design of kernels (i.e. generating functions) for time-frequency representations (TFRs) has been recently proposed. The goal of these criteria is to produce kernels (and thus, TFRs) which will enable accurate classification without explicitly defining, a priori, the underlying features that differentiate individual classes. These kernels, which are optimized to discriminate among multiple classes of signals, are referred to as signal class-dependent kernels, or simply class-dependent kernels. Here this technique is applied to the problem of radar transmitter identification. Several modifications to our earlier approach have been incorporated into the processing, and are detailed here. It will be shown that an overall classification rate of 100% can be achieved using our new augmented approach, provided exact time registration of the data is available. In practice, time registration can not be guaranteed. Therefore, the robustness of our technique to data misalignment is also investigated. A measurable performance loss is incurred in this case. A method for mitigating this loss by incorporating our class-dependent methodology within the framework of classification trees is proposed.