Title: Orthogonal Transforms for Signals on Directed Graphs

Abstract: In this paper we consider the problem of defining transforms for signals on directed graphs, with a specific focus on defective graphs where the corresponding graph operator cannot be diagonalized. Our proposed method is based on the Schur decomposition and leads to a series of embedded invariant subspaces for which orthogonal basis are available. As compared to diffusion wavelets, our method is more flexible in the generation of subspaces, but these subspaces can only be approximately orthogonal.