On robustifying some second order blind source separation methods for nonstationary time series (Q2442675)

The author discusses some blind source separation (BSS) methods for time series with nonstationary variances. The BSS method is usually formulated as \(\mathbf{x}_t=\mathbf{A}\mathbf{s}_t\), where \(\mathbf{s}=(\mathbf{s}_t)\), \(t=0,\pm 1,\pm 2,\dots\), is a \(p\)-variate time series which satisfies some general assumptions, \(\mathbf{x}_1\), \(\dots\), \(\mathbf{x}_T\) is an observed \(p\)-variate time series and \(\mathbf{A}\) is a full rank \(p\times p\) matrix. The following models are discussed. The second order source separation model is defined in Section 2 and two estimation methods are considered in Section 3. In Section 4, the nonstationary source separation model is defined. Three estimation methods for this type of model are given in Section 5. Their robust versions are presented in the next section. At the end of the paper, some simulation results are presented and discussed.