Deep Kernel Learning-Based Channel Estimation in Ultra-Massive MIMO Communications at 0.06-10 THz

Abstract

The abundant spectrum resources at the millimeter wave (mm-wave) and Terahertz band (0.06-10 THz) are promising to enable a new paradigm shift in wireless communications to satisfy the demand for higher data rates in the beyond 5G era. Due to limitations of very high propagation attenuations and molecular absorptions at such high frequencies, ultra-massive MIMO (UM MIMO) communications are needed to compensate for the distance limitations, propagation losses, and to achieve higher spatial diversity. However, existing channel estimation algorithms for massive MIMO solutions require excessive computation time and demonstrate high processing complexities, whereas the problem is escalated more when the antenna array size continues to grow in the UM MIMO system. In this paper, a channel estimation method based on deep kernel learning is proposed to address the issues with estimation efficiency and accuracy in the UM MIMO communication system. This nonlinear estimation algorithm is shown to be more efficient compared to classic linear estimation approaches. A numerical comparison is conducted through simulations to show the performance of the proposed solution.