An alternative realization of the SVD-based FSE

An investigation is conduced of the noise threshold performance of the singular-value-decomposition (SVD)-based least-mean-square (LMS) adaptive algorithm when the autocorrelation matrix of the input signal is not truly rank deficient and when low-rank approximation techniques are used to prefilter the channel output adaptively. It is shown that SVD improves the input SNR (signal-to-noise ratio) and reduces the eigenvalue spread of the autocorrelation matrix, but these improvements fall short of the amount required to significantly improve the convergence characteristics of the LMS fractionally spaced equalizer operating in low SNRs. The results show some improvement in the initial convergence rate of the algorithm, while the excess mean-squared error (MSE) remains essentially unchanged