Hybrid precoding based on tensor decomposition for mmWave 3D-MIMO systems

Millimeter wave (mmWave) communication is a significantly enabling technology in next generation cellular system. Combined with massive number of antennas, the throughput can be greatly improved but the computation complexity and power consumption can also be incredibly high. In this paper, we study the hybrid precoding design in mmWave three-dimensional (3D) massive multiple-input multiple-output (MIMO) systems. To exploit the characteristic of planar antenna arrays, we represent the 3D-MIMO channel response with tensor, and find the null space of interference users with tensor decomposition. The null space can be well approximated by the Kronecker product of azimuth and elevation directional array vectors, and thus the designed analog precoder can eliminate inter-user interference. Combined with the baseband digital precoding, which find the maximal projection direction of the desired channel on the null space, the conventional zero-forcing block-diagonalization (ZF-BD) precoding method is extended to tensor context with constant-modulus constraint of null space elements. Since there are massive antennas and only limited RF chains, the proposed method has larger freedom to suppress interference. Simulation results verify its superiority.