Structured tensor-based algorithms for delayed exponential fitting

In this paper, we address the problem of subspace-based estimation of the pole (angular-frequency and damping-factor) of a sum of damped and delayed sinusoids. Usually, this approach is based on the decomposition of structured matrices over a block-Vandermonde basis which verifies the shift-invariance property. In the proposed work, we generalize the matrix approach to the structured tensor case in the context of multilinear algebra. We show on synthetic fast-time varying signal, the advantages of this approach.