Energy Efficient ADC Bit Allocation for Massive MIMO: A Deep-Learning Approach

It is known that adopting Variable-Resolution (VR) ADCs in millimeter-wave (mmWave) Massive Multiple-Input Multiple-Output (MaMIMO) receivers improves Energy Efficiency (EE). However, the effect of imperfect channel state information (CSI) at the receiver is detrimental in achieving the EE. None of the previous works consider imperfect CSI for designing ADC Bit Allocation (BA) for MaMIMO receivers. We propose a deep learning based framework to achieve a near-optimal EE for MaMIMO receivers. The contributions of this paper include a machine learning approach to arrive at a BA that achieves near-optimal EE by training the framework for a combination of perfect and imperfect channels using the conditions derived for capacity maximization. Using simulations, we show that the EE obtained using our proposed approach is very close to that of the brute force both for perfect and imperfect channels. Also, through simulations, we claim a computational complexity advantage using the proposed approach compared to brute force after sufficient learning of the channels presented to the system.