Pilot contamination (PC) is a major impediment of large-scale multi-cell multiple-input multiple-output (MIMO) systems. Hence we propose an optimal pilot design for timedomain channel estimation, which is capable of completely eliminating PC. More specifically, a sophisticated combination of downlink training and ‘scheduled’ uplink training is designed with the aid of the optimal pilot set. Given the optimal pilot set, every user acquires its unique downlink time-domain channel state information (CSI) through downlink training. The estimated downlink CSIs are then embedded in the uplink training. As a result, PC can be completely eliminated, at the cost of a slight increase in training computational complexity. Our simulation results demonstrate the power of the proposed scheme. Most significantly, our scheme imposes a modest training overhead of (L + 3), training-phase durations corresponding to the number of OFDM symbols, where L is the number of cells, which is substantially lower than that imposed by some of the existing PC elimination schemes. Therefore, it imposes a less stringent requirement on the channel’s coherence time. Finally, our scheme does not need any information exchange between base stations.
展开▼
