Deep learning-based energy efficiency and power consumption modeling for optical massive MIMO systems

The fifth generation (5G) wireless communication system is considered a promising and recent research. Massive Multiple-Input Multiple-Output (MIMO) system has an influential role in improving game-changing enhancements in area throughput and energy efficiency (EE). EE refers to one of the easiest and most cost-effective ways to combat climate change, reduce energy costs for consumers, and improve the competitiveness of businesses. Deep Learning (DL) can significantly improve area throughput and EE. It plays a crucial role in the 5G wireless communication systems. Optical systems are not far from this system, which include the optical components which serve more accurately.To assess the overall power usage in up-link and down-link communications, a power dissipated model is introduced. The proposed model incorporates the overall power used by the base station (BS) power amplifier and circuit components as well as single antenna user equipment (UE). In this paper, EE and power consumption of massive MIMO systems are calculated based on Convolutional Neural Network hybrid with Long Short-Term Memory cell (CNNLSTM). This model is proposed to overcome the high complexity and over fitting by replacing the inner dense connections with convolution layers resulting in improved model performance. There are different linear processing schemes applied for detection and precoding, as Minimum Mean Squared Error (MMSE), Zero-Forcing (ZF), and Maximum Ratio Transmission/Maximum Ratio Combining (MRT/MRC). These schemes are applied to train our proposed CNNLSTM.It is observed the results are improved by 12.8% when using ZF (perfect CSI) and the system outperforms other schemes by 10%, 10.44% and 12.05% when using MRT, ZF (imperfect CSI), and MMSE, respectively, for the EE performance. The obtained results also reveal that an improvement of 7.5% is achieved when using MRT. It outperforms other schemes by 6.5%, 5% and 5%, respectively, when using ZF (perfect CSI), ZF (imperfect CSI), and MMSE for average power consumption per antenna using the CNNLSTM model. When using MRT, an improvement of 7.5% is achieved in the area throughput performance, and it outperformed the other schemes, ZF (perfect CSI), ZF (imperfect CSI) and MMSE, by 5.2%, 5% and 5.2%, respectively.