Abstract : Massive multiple input multiple output (MIMO) is a broadly used technique that can provide numerous gains in spectral efficiency. However, the degradation of beamforming performance due to outdated channel state information at the transmitter side (CSIT) induced by the mobility of users has been a significant problem waiting to be solved. It is reported that system performance will decrease 50 percent even in a moderate 30 km/h speed scenario. However, the CSI cannot be simply reconstructed through interpolation in high mobility scenarios due to the limitation of pilot density — the phenomenon is known as “Doppler aliasing”. To address this, we propose a novel nonuniform pilot pattern that can provide more spectrum resolution compared with the uniform pilot currently used in most communication protocols. Meanwhile, we maintain the density of pilots in order not to sacrifice the payload resources. Based on the novel pilot setting, we propose two-channel prediction schemes with compressive sensing and matrix completion methods. Simulation results show our scheme can outperform deep learning-based and auto-regressive-based methods for about 15 percent in terms of average throughput in the simulated channel generated from the COST2100 channel model. To further verify the applicability, we apply our schemes in real channels measured from a channel sounding campaign, the proposed methods also achieve 5 percent gain which validates their superiority over conventional methods.
Index terms : Channel aging, channel prediction, compressive sensing, massive MIMO, matrix completion.