Robust Multi-Dimensional Harmonic Retrieval Using Iteratively Reweighted HOSVD

Higher-order singular value decomposition (HOSVD) is usually required in R-dimensional (R-D) harmonic retrieval, where R ≥ 3. In this letter, we devise an iteratively reweighted HOSVD technique, which is referred to as IR-HOSVD, for multi-dimensional frequency estimation in the presence of impulsive noise. The main idea is to minimize the ℓ_p-norm residual errors along all the R dimensions, where 1 <p <2. After decomposition, standard subspace techniques can be applied for parameter estimation. Based on the numerical results, IR-HOSVD outperforms several state-of-the-art techniques in terms of root mean square frequency error for different impulsive noise models.