Stochastic gradient-based decorrelation algorithm with applications to multicarrier CDMA

A stochastic gradient-based decorrelation algorithm is suggested for separation of an unknown linear mixture of signals. It is shown that while the decorrelation algorithm is similar in cost to the LMS algorithm, its rate of convergence is significantly faster, making it more attractive for signal separation. Analysis of the decorrelator algorithm shows that the faster speed of convergence is a consequence of the eigenevalue spread associated with the decorrelation problem, which is smaller than the spread associated with the corresponding mean square problem. Operation of the algorithm is illustrated as an adaptive multiuser detector in a multiple carrier CDMA system.