Convergence of linear successive interference cancellation in CDMA

Paper in proceeding, 1997

In this paper we consider a matrix-algebraic approach to linear successive interference cancellation (SIC). It has been shown that both single- and multi-stage linear SIC schemes correspond to a one-shot linear matrix filtering. Eigenvalue conditions on the resulting linear filter for convergence of the multistage scheme are considered and the concept of e-convergence is introduced for determining the number of stages necessary for practical convergence. It is observed that the BER for the users does not generally reach its minimum when the scheme converges, hence, optimal performance is achieved after a limited number of cancellation stages, which does not have to be the same for different users.