Tensors for signal and frequency estimation in subspace-based methods: when they are useful?

Abstract

Tensor modifications of singular spectrum analysis for signal extraction and frequency estimation problems in a noisy sum of exponentially modulated sinusoids are reviewed. Modifications using Higher-Order SVD are considered. Numerical comparisons are carried out. It is shown numerically that for, the signal extraction problem, tensor methods generally perform worse than matrix methods for a single-channel series, but can outperform multi-channel SSA for a series system. For frequency estimation, tensor modifications are generally advantageous