Tensor-MODE for multi-dimensional harmonic retrieval with coherent sources

We devise a new method of direction estimation (MODE) based multi-dimensional harmonic retrieval algorithm, which is referred to as tensor-MODE (T-MODE). It is a R-dimensional generalization of the conventional MODE and higher-order singular value decomposition is performed on the tensor observations. Asymptotic efficiency is achieved for the proposed estimator even when the source signals are highly correlated or coherent. Furthermore, the computational cost is reasonable, because it consists of polynomial rooting, followed by a combinatorial search, whereas multi-dimensional optimization is not required.