HYPE: CNN Based HYbrid PrEcoding Framework for 5G and Beyond

Deepti Sharma, Kuldeep Marotirao Biradar, Santosh Kumar Vipparthi, Ramesh Babu Battula

Abstract: 5G and beyond (B5G) technologies have emerged with the increasing demand for higher data-rate wireless communication with low latency. The B5G network utilizes hybrid beam-forming (HBF) as a promising solution to provide large bandwidths with directional communication. Conventional HBF techniques are computationally complex and unable to fully exploit the spatial & partial channel state information, which results in very low spectral efficiency. Hence, this paper proposes an optimized framework, HYPE, integrating the convolutional neural network resulting in the reduced complexity for the HBF technique. A two-phase analog shifter is used to maximize the spectral efficiency of the system by resolving the constant modulus constraint. Experimental results justify the enhanced performance of the proposed framework than conventional algorithms. Extensive ablation studies on the proposed work were carried out to analyze efficiency more in detail.