Signal modeling and detection using cone classes

A new signal model-the cone classes-is presented. These models include classical models such as subspaces but are more general and potentially more useful than some existing signal models. Examples of cone classes include time-frequency concentrated classes and subspaces with bounded mismatch. The maximum likelihood detector for a cone class of signals in the presence of Gaussian noise is derived, and a simple algorithm is suggested as a possible detector implementation. The detector is examined in the specific case of subspaces with bounded mismatch. It is shown that there are conditions under which this detector has a higher detection probability for fixed false alarm than that of a comparable subspace detector and energy detector.