Fast Reduced Rank Equalizer for HSDPA Systems Based on Lanczos Algorithm

In this paper we propose a Lanczos structure for multi-stage Wiener filtering (MSWF) based space-time equalization for high data-rate WCDMA downlink. An efficient algorithm is derived that replaces the MSWF forward recursion with the Lanczos algorithm. The proposed Lanczos-based implementation provides simultaneously fast convergence and low complexity. Moreover, it does not assume any structure for the received signal covariance matrix. The proposed method is compared in simulations to symbol-level adaptive MSWF and sample matrix inversion based equalization using Cholesky decomposition. Symbol-level adaptive equalization suffers from large eigenvalue spread of the symbol-level covariance matrix. On the other hand, Cholesky-decomposition is computationally complex. The results indicate that significant savings in complexity can be gained using the proposed method without compromising the performance, even in 16-bit fixed point arithmetic