Fractionally-spaced equalizers based on singular value decomposition

Investigation of the eigenstructure of the autocorrelation matrix of the received signal in a complex or real fractionally-spaced equalizer indicates that this matrix has a certain number of nontrivial eigenvalues. By using singular value decomposition, it is shown that the excess mean-square error in least-squares adaptive algorithms is proportional to the number of nontrivial eigenvalues, instead of the number of taps. This leads to methods for effectively reducing the misadjustment error.<
>