Litcius/Paper detail

Linear System Identification Based on a Third-Order Tensor Decomposition

Jacob Benesty, Constantin Paleologu, Silviu Ciochină

2023IEEE Signal Processing Letters19 citationsDOI

Abstract

A wide variety of system identification problems can be efficiently addressed based on the Kronecker product decomposition of the impulse response, together with low-rank approximations. Such an approach solves the original system identification problem using a combination of two shorter filters. In this paper, targeting a higher dimensionality reduction, we develop a solution based on a third-order tensor decomposition. In addition, the problem of approximating the rank of a tensor is avoided thanks to the control of a matrix rank. Then, an iterative Wiener filter is developed, which outperforms both the conventional benchmark and the previously developed counterpart that exploits the second-order decomposition.

Topics & Concepts

Matrix decompositionWiener filterKronecker productMathematicsFinite impulse responseAlgorithmTensor (intrinsic definition)Dimensionality reductionKronecker deltaSystem identificationMathematical optimizationRank (graph theory)Tensor productComputer scienceArtificial intelligenceData miningPure mathematicsPhysicsEigenvalues and eigenvectorsCombinatoricsMeasure (data warehouse)Quantum mechanicsTensor decomposition and applicationsAdvanced Adaptive Filtering Techniques
Linear System Identification Based on a Third-Order Tensor Decomposition | Litcius