GSC realisations using the two-dimensional transform-domain LMS algorithm

The authors propose two new realisations of the broadband generalised sidelobe canceller (GSC) which are based on a two-dimensional (2D) extension of the transform-domain LMS algorithm. In the first realisation, the tap-input matrix of the GSC (following the blocking network) is mapped into the transform-domain via a 2D unitary image transform. Following this step, the transformed data are weighted and summed to produce the GSC output and the processor weights are updated via the self-orthogonalising LMS algorithm in the transform-domain. The second realisation uses a separable 2D transform and provides a narrowband decomposition of the GSC. The computational complexity of these new GSC realisations is analysed and shown to be comparable to that of the fast FLMS-GSC recently proposed by Chen and Fang (see J. Acoust. Sec. Am., vol.91, p.3354, 1992). However, as demonstrated by computer simulations, the new GSC realisations have a much faster convergence rate than the FLMS-GSC when the eigenvalue spread of the tap-input correlation matrix is large