Massive MIMO is expected to enable much higher throughput and energy efficiency compared to traditional MIMO systems. It is considered to be a potential key technology in future 5G standards. Despite the promise, there are still open problems that limit the potential of massive MIMO. Among them, this thesis focuses on some of the challenges related to the acquisition of Channel State Information (CSI) in both Time-Division Duplex (TDD) mode and Frequency Division Duplex (FDD) mode.
In channel estimation phase of TDD mode, the pilot contamination effect constitutes a bottleneck for overall performance. We present novel approaches that tackle this problem by exploiting second-order statistics of the user channels. We demonstrate analytically that in the large-number-of-antennas regime, the pilot contamination effect completely vanishes under a certain condition on the channel covariance. This condition states that the support of multipath angle-of-arrival (AoA) of interference is non-overlapping with the AoA support of the desired channel. This condition is tightly related to the low-rankness property of channel covariance. Furthermore, we show that such a low-rank property is not inherently related to ULA. It can be generalized to non-uniform array, and more surprisingly, to two-dimensional distributed arrays.
Although the proposed MMSE-based estimator leads to full pilot decontamination under the non-overlapping condition in angular domain, in practice this condition is unlikely to hold at all times, owing to the random user location and scattering effects. To this end, we propose novel robust channel estimation schemes that combine the merits of MMSE estimator and the known amplitude based projection method. Asymptotic analysis shows that the proposed methods require weaker conditions compared to the known methods to achieve full decontamination.
Finally, we tackle the CSI feedback problem for massive MIMO operating in FDD mode by novel cooperative feedback mechanisms which are enabled by Device-to-Device (D2D) communications. The exchange of local CSI among users allows to construct more informative forms of feedback based on this shared knowledge. For a given feedback overhead, the sum-rate performance is assessed and the gains of our novel methods compared to a conventional massive MIMO setup without D2D are shown.