Pilot Contamination and Mitigation Techniques in Massive MIMO Systems

Abstract: A multi-antenna base station (BS) can spatially multiplex a few terminals over the same bandwidth, a technique known as multi-user, multiple-input multiple-output (MU-MIMO). A new idea in cellular MU-MIMO is the use of a large excess of BS antennas to serve several single-antenna terminals simultaneously. This so-called "massive MIMO" promises attractive gains in spectral efficiency with time-division duplex operation. Within a cell, the BS estimates the channel from mutually orthogonal reverse-link pilot sequences to formulate a receiver for the reverse link and (assuming reciprocity) a precoder for the forward link. The channel coherence is typically constrained in time as well as frequency, leading to a trade-off between the resources spent on pilots and those available for data symbols. This pilot overhead can be reduced by reusing pilot sequences in nearby cells, however this potentially introduces interference in the channel estimation phase, the so-called "pilot contamination" effect. In this thesis, we study the impact of pilot contamination in realistic environments and investigate schemes to mitigate it. We evaluate the mean squared error (MSE) of channel estimates in case of a plain-vanilla least-squares (LS) estimator and a minimum MSE (MMSE) estimator that exploits prior knowledge of second-order channel statistics. Next, we introduce a pilot open-loop power control (pilot OLPC) scheme to improve the SINR-fairness of received pilot signals at the BS. We evaluate the effect of relaxing the pilot reuse factor and also implement a soft pilot reuse (SPR) scheme to distribute pilot sequences efficiently. To study the trade-off between pilot and data symbols, we evaluate the achievable rate in forward link with maximum-ratio and zero-forcing precoding at the BS. We evaluate an inter-cell coordination scheme that exploits prior knowledge of all cross-channel covariance matrices to reuse pilots among spatially well-separated terminals. We simulate a 21-cell MU-MIMO setup with up to 100-antenna BSs and up to 24 single-antenna terminals per cell in an outdoor urban macro environment. We find that pilot reuse 1 causes severe impairment of the channel estimates, which can be improved with pilot OLPC. Pilot reuse 1/3 effectively mitigates pilot contamination, and can improve the achievable rate in the forward link. SPR also mitigates contamination but with a smaller increase in pilot overhead. Inter-cell coordinated pilot allocation, implemented using a greedy approach, provides gains over random allocation only for the initial few pilots. In general, maximum ratio precoding is more robust against pilot contamination than zero-forcing.