Projection-based degree of distinguishability in switching linear systems

In this paper, we consider the problem of identifying the active mode of a switching linear system from data sequences of a finite length. The results combine elements from canonical correlation analysis and subspace projection methods. In addition to providing insight into the geometric meaning of the problem, the results turn out to be of practical relevance whenever mode identification is addressed in the presence of noisy-corrupted measurements. In particular, the results can be used to provide a characterization of the regions of the space where mode identification occurs independently of a prescribed noise level.