Resource Management for Device - to - Device Communication in Hetrogeneous Network Using AI

Abstract

The broad accessibility of high-speed Internet connectivity enables users to engage with data-intensive applications like online gaming, live streaming, and various others. Faster and more reliable Internet connections facilitate the seamless functioning of applications that require large amounts of data. Moreover, it heightened the need for bandwidth accompanied by reduced latency. When the demand for bandwidth exceeds the available spectrum, it can result in congestion and reduce spectral efficiency. So, there is a need to enhance the current cellular network infrastructure. Researchers are investigating and highlighting the potential advantages of device-to device (D2D) communication to enhance spectral efficiency. It provides direct communication, enabling devices to exchange data without depend ing on the cellular infrastructure, which enhance spectrum efficiency and network capacity. Apart from these advantages, D2D communication faces several issues, i.e., initial device discovery, resource allocation (ReA), and interference management. ReA is the major issue in underly D2D com munication, where the efficient allocation of power, spectrum, and channel resources enhances fairness, energy efficiency, and system throughput. In ReA, the devices face interference like the D2D receiver faces inter ference from the cellular users (CUs) and CU experiences interference from the D2D receiver. Hence, there is a requirement for an efficient ReA in a D2D communication environment (DCE). Further, optimising the transmit power of D2D users (DUs) is required to mitigate interference in the DCE. The research community employed various schemes based on game theory, heuristic algorithms, and graph theory for ReA and PC in DCE. However, the existing system faces computation overhead issues when more devices are in the dynamically changing communication environment. Here, we proposed an AI-based ReA and PC scheme that improves the system sum rate Srate while reducing the computation overhead. ❼ Firstly, we propose a novel scheme for efficient ReA in DCE. Firstly, we analyzed the Hungarian algorithm, which performed well for ReA in DCE; it takes the weight matrix as an input and gives the optimum cost value. Nevertheless, it faces issues when the matrix has the same consecutive weights. To tackle this issue, we proposed an autoencoder model, which regenerates the weight matrix (data rate matrix). Later, the regenerated matrix is fed into the Hungarian algorithm, which gives optimum cost with efficient ReA. ❼ The second scheme is an AI-driven PC scheme for DCE. In this scheme, we proposed a cognitive radio (CR) and deep neural net work (DNN) approach for transmitting the PC of DUs. Initially, we applied the CR concept to select the best CUs (secondary users) to mitigate the interference effect in DCE. Then, we proposed a DNN model, which takes the channel gain (CG) matrix as an input and op timizes the transmit power of DUs with minimum computation time. ❼ The performance of the proposed scheme is evaluated using Srate, com putation time, accuracy, loss, optimizers, and learning rates (LRs). The simulation of the proposed AI-driven ReA and PC schemes was per formed on MATLAB 2022a and Google Colab environment. For that, the simulation parameter is considered using the 3rd Generation Partnership Project (3GPP) guidelines.