ML-Aided Dynamic Clustering and Classification of UEs as VBs in D2D Communication Networks

With the next generation of mobile devices and streaming services such as Virtual Reality, Augmented Reality, and Meta being available worldwide, the network's data rate, latency, and connectivity must be improved. Even though 5G provides the user with the required Quality of Service (QoS) in terms of data rate and reduced delays by transmitting signals in the higher frequencies (called New Radio (NR)), it faces a lot of attenuation, leading to a short communication range. However, to meet goals set by utilising 6G requirements, many technological advancements and components must be incorporated into the network. The dense disposition of small cells will help reduce the network traffic in hotspot areas and increase the coverage and spectral efficiency. Nonetheless, the current deployment of 5G Base Stations (BSs) and small cells is static and cannot move around even though they are deployed in hot spot areas, leading to high operational costs. Furthermore, more than these static deployments of base stations can be required in an unpredictable scenario of extreme crowd movement. To overcome these issues, Device to Device (D2D) Communication with the dynamic deployment of Virtual Base stations (VBSs) can be called upon, which can be achieved by using User Equipment (UE) such as phones or laptops to mimic the functions of a Base Station (BS). Therefore, in this paper, a User Equipment based Virtual Base Station (UE-VBS) is studied, which will act as a secondary base station and, in turn, help alleviate the traffic load in the network. Specifically, as one UE cannot relieve the entire network traffic load, the network area is split into different clusters by using an unsupervised Machine Learning (ML) clustering technique(i.e., K-Means with Mean Shift Clustering), and a single UE is selected to act as a VBS for that cluster with the utilisation of supervised ML classification techniques (i.e., Decision Trees, Logistic Regression, Linear Discriminant Analysis And Quadratic Discriminant Analysis, Linear Support Vector). In our work, we utilise the K-means along with mean shift clustering techniques to cluster simulated network areas accurately. Also, we use and compare different classification machine learning techniques to predict/classify whether user equipment can be employed as a VBS and become UE-VBs. Our simulation study reveals that the Decision Tree algorithm achieves the highest accuracy in categorising the eligible UEs as UE-VBs.