Analysis of adaptive least squares filtering in massive MIMO

This paper considers an adaptive beamforming algorithm for a massive MIMO system with multiple cells. The pilot contamination problem arises in multi-cell systems owing to transmission of the same pilots from users (or mobile stations) in different cells. The focus of this paper is to study the impact of different training sequences on pilot contamination in Massive MIMO systems. Specifically, we consider an adaptive beam-forming salgorithm which has been previously applied in MIMO interference networks. This algorithm uses bidirectional training in which training sequences are sent from current beamformers to adapt the mobile station receive filters and then the training sequences using mobile station filters as beamformers, are sent in reverse direction to adapt the beamformers at base station side. The adaptation of both transmit and receive filters is done using the least squares objective function. The adaptive beamforming algorithm shows improvement in performance in terms of average sum rate if the random training sequences are transmitted from users in different cells. Numerical results are presented to corroborate the mathematical analysis.