A multi-protocol energy optimization method for an adaptable wireless MAC system through machine learning

The traditional methods used to optimize energy efficiency in wireless communication systems employ medium access control (MAC) protocols that operate under complex functions that require a high computational cost to optimize a limited number of parameters. Furthermore, the inability of these methods to work under different protocols that involve learning and adapting the device specifications associated with existing problems, such as security, error tolerance, and human interference, makes their implementation in real systems impractical; this limitation mainly applies to networks with high node scalability. This paper presents a novel approach to this problem using machine learning to attain energy savings. The method proposes combining operating information from multiple power consumption control algorithms, CSMA/CA or slotted ALOHA (a variant of ALOHA), benchmark MAC protocols used in WiFi, and LoRaWAN technologies, creating a database of optimized solutions, which serves as a training base for a neural network capable of learning the behavior of all protocols simultaneously and creating a unified self-adaptive energy optimization model that considers multiple physical (PHY) and MAC layer variables for different devices and protocols. The proposed approach simultaneously presents solutions that optimize the energy reduction algorithms for different protocols, approaching or improving the performance of the techniques, saving 97.6% in CPU computation and 113,322,733% of the processing time in the search for the same solutions. The main contribution of this work is the proposal of an adaptable multi-protocol approach based on machine learning, which manages resources in slotted ALOHA and CSMA/CA benchmark protocols for wireless networks. Furthermore, it facilitates multi-objective optimization via machine learning for energy efficiency in real networks. It creates a new intelligent system that promotes efficient communication for multiple MAC protocols and considers the device’s processing capacity limitation. This work also shows that a neural network can approximate and optimize exact functions when the optimal parameters cannot be mapped mathematically.