Hierarchical Clustering based Dynamic Subarrays for Hybrid Beamforming Massive MU-MIMO

Abstract: Hybrid Beamforming in multiple input multiple output systems has surged in the past few years as the prime precoding architecture for massive MIMO systems since it addresses the systems performance and complexity/power consumption tradeoff. Most of the literature addresses HBF MIMO with a fixed antenna subarray, however it is possible to further increase the spectral efficiency of the system by dynamically configuring the HBF MIMO subarray according the channel state information of the users. In this study it is shown that finding the subarray configuration which maximizes the largest singular values of the channel covariance matrix is a good near-optimal approach to find the mapping which provides the best system performance. With this discovery, two dynamic mapping algorithms, based on hierarchical clustering analysis, are presented. Since both algorithms provide marginally identical performance, the least computational expensive one was chosen to run an extensive study to understand under which conditions the dynamic subarray should be used and how the channel characteristics and the system parameters can affect the potential gain.To implement this new dynamic subarray architecture, it is necessary to add a switch network to the HBF MIMO. Therefore, to reduce the complexity of the network, and consequently its insertion losses, some constrained variants of the dynamic mapping algorithm are proposed.Simulation results show that in some particularly challenging scenarios the dynamic subarray can provide up to 100% performance increase, when compared with the fixed architecture, in the low SNR region and for high SNR more than 50% sum rate increase. Results also indicate that in the long term, for various scenarios, the dynamic subarray is expected to show significant performance gain.