Machine Learning-Inspired Hybrid Precoding for HAP Massive MIMO Systems with Limited RF Chains

Energy efficiency (EE) is the main target of wireless communication nowadays. In this paper, we investigate hybrid precoding (HP) and massive multiple-input multiple-output (MIMO) systems for a high-altitude platform (HAP). The HAP is an emerging solution operating in the stratosphere at an amplitude of up to 20-40 km to provide communication facilities that can achieve the best features of both terrestrial and satellite systems. The existing hybrid beamforming solution on a HAP requires a large number of high-resolution phase shifters (PSs) to realize analog beamforming and radio frequency (RF) chains associated with each antenna and achieve better performance. This leads to enormous power consumption, high costs, and high hardware complexity. To address such issues, one possible solution that has to be tweaked is to minimize the number of PSs and RFs or reduce their power consumption. This study proposes an HP sub-connected low-resolution bit PSs to address these challenges while lowering overall power consumption and achieving EE. To significantly reduce the RF chain in a massive MIMO system, HP is a suitable solution. This study further examined adaptive cross-entropy (ACE), a machine learning-based optimization that optimizes the achievable sum rate and energy efficiency in the Rician fading channel for HAP massive MIMO systems. ACE randomly generates several candidate solutions according to the probability distribution (PD) of the elements in HP. According to their sum rate, it adaptively weights these candidates' HP and improves the PD in HP systems by minimizing the cross-entropy. Furthermore, this work suggests energy consumption analysis performance evaluation to unveil the fact that the proposed technique based on a sub-connected low-bit PS architecture can achieve near-optimum EE and sum rates compared with the previously reported methods.