A subspace approach for blind multiuser detection based on second-order nonstationarity

This paper presents a novel technique for blindly starting up an MMSE receiver for a user newly entering a CDMA system. It is assumed that only one user at a time is admitted into the system and that data before and after the onset of the new transmission are available. In general, the two corresponding autocorrelation matrices (before and after the new user is admitted) differ by a low-rank matrix contributed by the autocorrelation of the new user signal. This fact yields a useful property, that the generalized eigenvectors corresponding to the minimum eigenvalue span the noise subspace of the low-rank signal term. Exploiting this subspace relation, the signature of the new user can be identified and thus an MMSE receiver for it can be constructed.