1: \begin{abstract}

2: Factor analysis (FA) or principal component analysis (PCA) models

3: the covariance matrix of the observed data as $\mathbf{R}=\mathbf{S}\mathbf{S}^{T}+\boldsymbol{\Sigma}$,

4: where $\mathbf{S}\mathbf{S}^{T}$ is the low-rank covariance matrix

5: of the factors (aka latent variables) and $\boldsymbol{\Sigma}$ is

6: the diagonal matrix of the noise. When the noise is anisotropic (aka

7: nonuniform in the signal processing literature and heteroscedastic

8: in the statistical literature), the diagonal elements of $\boldsymbol{\Sigma}$

9: cannot be assumed to be identical and they must be estimated jointly

10: with the elements of $\mathbf{S}\mathbf{S}^{T}$. The problem of estimating

11: $\mathbf{S}\mathbf{S}^{T}$ and $\boldsymbol{\Sigma}$ in the above

12: covariance model is the central theme of the present paper. After

13: stating this problem in a more formal way, we review the main existing

14: algorithms for solving it. We then go on to show that these algorithms

15: have reliability issues (such as lack of convergence or convergence

16: to infeasible solutions) and therefore they may not be the best possible

17: choice for practical applications. Next we explain how to modify one

18: of these algorithms to improve its convergence properties and we also

19: introduce a new method that we call FAAN (Factor Analysis for Anisotropic

20: Noise). FAAN is a coordinate descent algorithm that iteratively maximizes

21: the normal likelihood function, which is easy to implement in a numerically

22: efficient manner and has excellent convergence properties as illustrated

23: by the numerical examples presented in the paper. Out of the many

24: possible applications of FAAN we focus on the following two: direction-of-arrival

25: (DOA) estimation using array signal processing techniques and portfolio

26: selection for financial asset management. 

27: \end{abstract}

28: