Iterative multi-user detection and time-variant channel estimation in a multi-carrier (MC) code division multiple access (CDMA) uplink requires high computational complexity. This is mainly due to the linear minimum mean square error (LMMSE) filters that are used for multi-user detection and time-variant channel estimation. Krylov subspace methods allow for an efficient implementation of the LMMSE filter. We show that a suitable chosen starting value, exploiting the iterative receiver structure, allows for a further speedup of the Krylov method. We achieve a complexity reduction by more than one order of magnitude. The Krylov subspace method allows a parallelization of the computations of the multi-user detector, while keeping the receiver performance constant. Numerical simulation results for a fully loaded system with K =64users are presented. 1.

We consider the uplink of a time-variant multiple-inputmultiple -output (MIMO) multi-user (MU) multi-carrier (MC) code division multiple access (CDMA) system. The linear minimum mean square error (LMMSE) filters for channel estimation and multi-user detection at the receiver side require too high complexity to be implementable in a system operating in time-variant channels. We develop a low-complexity implementation of an LMMSE filter based on the Krylov subspace method. We are able to reduce the computational complexity by one order of magnitude. Furthermore, in a system with K users having N T transmit antennas, parallelization of the computations of the multi-user detector into KN T branches is achieved as well as considerable storage reduction. We discuss more specifically a fully loaded system (the number of subcarriers N is equal to the number of users K) with N T =2 transmit antennas per user, NR =4 receive antennas and K = N = 64.

Iterative multi-user detection and time-variant channel estimation in a multi-carrier (MC) code division multiple access (CDMA) uplink requires high computational complexity. This is mainly due to the linear minimum mean square error (LMMSE) filters that are used for multi-user detection and time-variant channel estimation. Krylov subspace methods allow for an efficient implementation of the LMMSE filter. We show that a suitable chosen starting value, exploiting the iterative receiver structure, allows for a further speedup of the Krylov method. We achieve a complexity reduction by more than one order of magnitude. The Krylov subspace method allows a parallelization of the computations of the multi-user detector, while keeping the receiver performance constant. Numerical simulation results for a fully loaded system with K =64users are presented.