Blind separation of FHSS signals using PARAFAC analysis and quadrilinear least squares

This paper considers the problem of blindly separating multiple-frequency-hopped spread-spectrum (FHSS) signals using an antenna array, without knowledge of hopping patterns or directions of arrival (DOAs). We propose to use PARAllel FACtor (PARAFAC) analysis of four-way data generated by capitalizing on both spatial and temporal shift invariance. As a preprocessing step, we may identify a hop-free subset of the data by discarding high-entropy spectral slices from the spectrogram. The analysis step is implemented using quadrilinear alternating least squares (QALS). We discuss identifiability and performance issues, and compare our algorithm with some earlier algebraic approaches. We also discuss robustness issues when QALS is applied to the raw data without first detecting a hop-free data set.