Efficient Training of Adaptive MIMO Channel Tracking Algorithms

Abstract:
In this paper, improved training schemes for adaptive decision-directed tracking filters are proposed and applied to time-varying flat fading MIMO channels. In case of periodic training with a limited amount of pilots, it is shown that a trade-off between investing pilots for good initialization and exclusive training of the algorithm leads to a lower BER and a higher spectral efficiency than conventional training. In E. Eitel and J. Speidel, Efficient training of Kalman algorithm for MIMO channel tracking, we introduced a novel aperiodic training scheme for the Kalman filter. We extend this scheme to the RLS and LMS algorithms and develop appropriate metrics for filter divergence detection for each algorithm. The metric thresholds are derived analytically. We show that the proposed training reduces the BER significantly. The impact of feedback delay caused by pilot request and transmission is also investigated. Simulation results show that pilot on request training outperforms periodic training even for large mean pilot delays.