An Efficient Tensor Decomposition Scheme for Large-Scale Spectrum Environment Data Processing

This letter proposes an efficient tensor decomposition scheme, termed tensor flower (TF), for rapid matrix format factorization of large-scale spectrum environment data. TF leverages the divide-and-conquer methodology to break down higher-order tensors into lower-order components to complete matrixized tensor decomposition. This is achieved by representing the tensor as an ordered collection of factor matrices resembling an inflorescence structure. Then, a streamlined algorithm based on alternating least-squares (ALS) is devised to validate the feasibility, while a hierarchical algorithm with adaptive ranks (HAR) is developed to achieve faster TF decomposition. Simulation results demonstrate that TF, as a general-purpose tensor decomposition scheme, can efficiently process large-scale spectrum environment data.