Spatio-Temporal Hybrid Learning-Based Cross-Layer Resource Allocation for Energy-Efficient in 5G IoT Networks

Abstract

Fifth-generation (5G) wireless networks have brought the Internet of Things (IoT) devices to unprecedented levels of difficulty due to the rapid growth and development of the devices in the wireless networks. Massive Machine-Type communications (mMTC) involves the use of thousands of low-power and battery-constrained devices that create heterogeneous and changing traffic. Despite the superior spectral performance, ultra-reliable low-latency communication (URLLC), and network slicing of 5G, the efficient utilization of the resources is a life-threatening problem because of the dynamic channel scenarios, the interference, and varied Quality of Service (QoS) demands. The classic optimization-based resource allocation techniques usually make use of precise mathematical models and make fixed assumptions, and are therefore computationally costly and less responsive to quickly evolving network conditions. In order to overcome these drawbacks, smart cross-layer optimization solutions have been considered as the potential solution. Cross-layer resource allocation is a common parameter that is considered jointly on the parameters of the physical, MAC, and network layers to obtain the global performance: transmit power, subcarrier assignment, scheduling, modulation schemes and queue states. Nevertheless, a majority of current methods lack the ability to represent both spatial correlations (e.g., pattern of interferences, topology) and temporal fluctuations (e.g., bursts in traffic, variations in channels). The paper suggests a spatio-temporal hybrid learning-based cross-layer framework of distributing resources in 5G IoT networks with energy-efficiency. The proposed approach will adaptively optimize resource allocation by combining spatial feature modeling with temporal learning processes and decision policies measured by reinforcement learning and reducing energy use and meeting QoS thresholds. The framework facilitates scalable, data-driven network control, which is much more energy efficient, latency performance and system reliable in dense IoT deployments.