Waveform design by Task-Specific Information

We propose a new waveform design metric called Task-Specific Information (TSI). As its name suggests, TSI is an information-theoretic metric formulated for a specific task or set of tasks. For example, in this paper we consider the problem of correctly classifying a linear system from a set of known alternatives. The TSI metric directly quantifies information about the hypotheses while previous information-based approaches quantify mutual information related to potential system realizations. Thus the TSI framework is a more direct measure of task-related performance. In addition, for the problem statement considered in this paper it is possible to find a globally optimum solution through gradient-based search techniques. We compare the performance of TSI and conventional waveform designs.