SMART POWER ALLOCATION IN MOBILE SYSTEMS USING DEEP LEARNING: AN EMERGING FIELD SURVEY

Abstract

The rapid expansion of mobile connectivity, driven by the widespread use of smartphones, IoT devices, and advanced applications like UAVs, driverless cars, and smart cities, has put wireless networks under previously unheard-of pressure. It is anticipated that fifth-generation (5G) and soon-to-be sixth-generation (6G) systems would provide exceptionally high data speeds, very low latency, extensive device connections, and enhanced energy efficiency. Meeting these requirements calls for intelligent and adaptive resource management, with power allocation playing a central role in ensuring spectral efficiency, energy sustainability, and quality of service (QoS). Conventional techniques, including water-filling, game theory, and convex optimization, have been widely used but face limitations when applied to dynamic, heterogeneous, and large-scale environments. Recent years have seen the rise of deep learning (DL) as a potent technique that can handle the high-dimensional, time-varying, and nonlinear characteristics of mobile communication systems. DL-based techniques allow for distributed, data-driven, real-time power allocation through the use of models including deep neural networks, convolutional and recurrent architectures, reinforcement learning, and federated learning. These approaches demonstrate strong adaptability across diverse scenarios, including 5G/6G networks, IoT ecosystems, vehicular communications, and UAV systems. Together, they represent a transformative step toward achieving more efficient, scalable, and intelligent wireless communication.