Agenda

Signal Processing Seminar

• Thursday, 8 March 2018
• 13:00-15:00
• HB 17.150

Time-Varying System Theory - Can it be connected to Graph Signal Processing?

Long time ago, I was developing time-varying system theory. The rows of an "arbitrary" matrix can be viewed as impulse responses of a time-varying system. Next, there is a notion of causality, which relates to upper triangular matrices. And there is a "shift operator" which provides connections between the rows. From these ingredients, it turned out that we can develop a state-space theory where the matrix is implemented by a series of "nodes" that communicate to each other via "states". The ARMA graph filtering work of Elvin e.a. results in similar expressions, where the shift operator is the Laplacian. An open question is if this can be connected to the TV system theory? If so, we know how to do realization theory (given the responses, find a minimal realization, i.e. minimize the number of communication links) and approximation theory (given a realization, find one of lower complexity that has approximately the same response). The talk won't give the answers, but I hope it can start a discussion.

Overview of Signal Processing Seminar