A memory direction descent search near optimum multiuser detection for DS/CDMA communications

In this paper a memory direction descent search (MDDS) near optimum multiuser detector for DS/CDMA communications is proposed. Based on an explanation of searching the global minima of the quadratic function of maximum likelihood (ML) optimum multiuser detection along the gradient direction for the parallel interference cancellation (PIC) multiuser detection, we modify the gradient direction in each stage as a new direction in the Krylov subspace which is spanned by the previous search directions. Directions in the Krylov subspace partially remember the search directions of previous stages, so searches along them have larger probabilities to avoid converging to the local minima. An easy iterative method is given to determine the memory factors. Simulations show a large performance gain is obtained over conventional PIC detector with few additional computational complexity