Performance of SVD-based fractionally-spaced equalizers in data transmission systems

An algorithm is presented for implementing a complex fractionally spaced equalizer (CFSE), which uses the least-mean-square (LMS) stochastic gradient method and is based on singular value decomposition (SVD). The SVD is used to adaptively prefilter the channel output and reduce the eigenvalue spread of the autocorrelation matrix of the CFSE tap inputs. It is shown that SVD accelerates the convergence of the CFSE, but increases somewhat the steady-state excess mean square error. The computational complexity of the SVD-based CFSE is also discussed. Simulation results are presented.<>