Machine Learning Based Small Cell ON/OFF for Energy Efficiency in 5G Heterogeneous Networks

Machine Learning (ML) predictions are envisioned to influence the implementation of 5G networks. 5G heterogeneous networks (Hetnets) are characterized by the combined operation of small cells or small base stations and a main base station in every macrocell coverage area of the cellular network. A vital performance indicator is the network energy efficiency that reflects the system power consumption and the user throughput. The small cells being additional hardware, play a key role in increasing the overall power consumed which can be curbed by incorporating a strategy to turn off the appropriate small cells without affecting the quality of service. User throughput prediction is another key engineering task in mobile communications. In this paper, we propose an ML based small cell ON/OFF algorithm in 5G Hetnets to address the challenges of throughput reduction and optimum power consumption in the small cell switching process. The proposed algorithm is executed as follows: The network parameters or predictors that affect the system throughput are carefully identified. Through multiple linear regression, the correlation between the predictors and the throughput is evaluated. Based on this analysis the ML model is trained to predict the throughput as well as the required number of active small cells for any set of input parameters. The largest RSRP (Received Signal Reference Power) criteria select the small cells that are to be kept ON. Simulation results justify an improvement in throughput and energy efficiency of the proposed scheme by 17% as compared to the benchmark schemes.