Preconditioned and Rank-Flexible Block Conjugate Gradient Implementations of Mimo Wiener Decision Feedback Equalizers

In this paper, we present two extensions of the block conjugate gradient (BCG) algorithm, a method which exploits the concept of block Krylov subspaces. First, we extend the BCG algorithm such that it is more flexible concerning the dimension of the block Krylov subspace. Second, a computationally efficient preconditioned BCG (PBCG) algorithm is introduced which turns out to outperform the standard BCG algorithm concerning the complexity-performance ratio. Hence, we provide a powerful implementation for reduced-rank signal processing in the minimum mean square error (MMSE) sense. Simulation results show the gain in rank-flexibility and convergence speed