New orthogonal adaptive equaliser based on block SVD-KLT

The least means square (LMS) is a simple yet very effective algorithm. However the algorithm suffers from convergence problems in communication channels with low signal to noise ratio (SNR), subjected to strong multipath fading. The reason for the reduced performance of LMS is due to the various assumptions made in the basic algorithmic development of the algorithm. In channels subjected to strong multipath fading it can not be guaranteed that the input vector is independent and Gaussian as it may consists of correlated interference components or the delayed version of the same input signal. In such cases the covariance matrix comprises large sparse components and is ill conditioned. This is identified by the large eigenvalue spread in the covariance matrix. In order to ameliorate the effects of mutually correlated vector elements, the input signal is transformed into corresponding vector of uncorrelated signal components. In this paper, various self orthogonalising transform schemes have been developed, including block singular value decomposition (SVD) based Karhunen-Loeve transform (KLT) and applied to equalisation problems for communication signals in low SNR and multipath channels. A self-orthogonalising adaptive filter is shown and it consists of a conventional adaptive filter preceded by a preprocessor. The function of the preprocessor is to orthogonalise the input vector so that the input vector components to the adaptive filter are uncorrelated. We propose an approach which employs the block KLT, implemented using SVD which is amenable to systolic implementation. The KLT-SVD preprocessor precedes a conventional LMS filter