Optimal radio access technology selection and datarate allocation in 5G and beyond networks
Year: 2025
Type: dissertation
Abstract: The continuous and exponential number of users in mobile communication, which lead to spectrum saturation called for the exploration of 5G spectrum in Extremely High Frequency (EHF) range. This was done to accommodate the excess devices in form of mobile stations, smartphones, IoT devices, laptops, tablets, palmtops, wearables and other network devices attached to the mobile networks, from the breakthrough recorded in the implementation of full IP, 4G Long Term Evolution (LTE) networks. Meanwhile, the migration of 4G LTE to 5G network for seamless operation, brought about the two architectural designs of 5G network; 5G Non-Stand Alone (NSA), and 5G Stand-Alone (SA), as specified in 3GPP release 15. In 5G NSA, the 5G network still leverages on 4G LTEsupport, whilethe5GSAdoesnot. However,itisrecommendedthat5GSAnetworks still run along side 4G SAinMultipleRadioAccessTechnology(Multi-RAT)scenario, for inter-generation handover. Hence, the choice of a particular network RAT between 5G and 4G LTE in terrestrial communication, which many users depend on becomes imperative. Therefore, the need for optimal RAT selection between these two networks. Many criteria which are mainly user-centric have been considered in literature. However, this research considers both network-centric and user-centric factors in both 5G NSA and 5G SA RAT selection scenarios, for efficient quality of experience (QoE) from users’ perspective, and quality of service (QoS) from operators’ point of view. Latitude, Longitude, Received Signal Strength Indicator (RSSI), Bandwidth and Data rate were used as criteria to find the appropriate RAT, since mobile users are handed over from one base station to another, based on factors like; geographical location of user, bandwidth requirement and network coverage. It is an open fact, that users that are most impacted as regards RAT selection are the pedestrians and travelling users. Consequently, the choice of pedestrian in a dense urban region, and mobile user travelling from a sub urban region to urban region in public bus, which have not been used in previous researches in 5G NSA network. Pedestrians and public transport users experience frequent handovers due to mobility; thus, optimising their RAT selection significantly impacts overall network efficiency. Moreover, unlike previous studies, this research effort utilised novel machine learning (ML) techniques, for joint optimisation of RAT selection and data rate allocation, which involved practical methodological rigour of live data validation, and cross-validation metrics, of high significance to users, and quantified improvements to operators. In the pedestrian scenario with single 5G NSA base station, RAT selection was formulated as a classification process, and further implemented with tree based classification MLalgorithms: Decision Tree (DT), Extra Tree (XTREE), Random Forest (RF), Gradient Boosting (GB), and eXtreme Gradient Boosting (XGBoost) to select the appropriate RAT. Standard classification evaluation metrics, revealed measure of accuracy of DT at 91.82%, RF at 87.64%, XTREE at 86.75%, GB at 91.86%, and XGBoost at 93.86%, respectively. These algorithms were cross-validated for future data, where DT, RF, XTREE, GB and XGBoost recorded 0.909, 0.912, 0.918, 0.918, 0.929 cross-validation score respectively. Observation showed XGBoost has highest performance value, hence adopted as the algorithm for RAT selection. However, in the 5G NSA multiple base station scenario transversed by a mobile user in public transport bus, RAT selection was carried out by highly rated mobilty algorithms: Support Vector Machine (SVM), Deep Neural Network (DNN), and eXtreme Gradient Boosting (XGBoost) as benchmark, to select the appropriate RAT between 4G and 5G RATs. Similarly, evaluation of results with standard classification metrics showed SVM, DNN and XGBoost at 0.8100, 0.9915, and 0.9955 respectively. Results were further cross-validated, where SVM, DNN, and XGBoost showed score values of 0.7293, 0.9906, and 0.9857 respectively. This further justify XGBoost as the suitable ML model, based on its overall outstanding accuracy performance and cross-validation score. Furthermore, in preparation for the implementation of 5G SA in terrestrial communication terrain, operators and vendors have also begun to equip their products with ML functionalities. Hence, due to challenges related to 5G SA in area of coverage due to high frequency, it is proposed that the 5G SA implementation is run independently with 4G SAnetworks for inter-RAT fall-over. Therefore, to address the significant challenge in selecting the most suitable RAT for efficient users’ experience, this research work proposes a novel solution, that leverages on ML based technique and weighted user association method, to address the complexities associated with these heterogeneous networks. The proposed solution considers factors such as; transmit power of users, location of users, available bandwidth on base stations, number of existing users on base stations, and pathloss between users and base stations, from both users’ and networks’ perspective. These aforementioned factors are then used as constraints, in an optimisation problem solved by a novel optimiser, to achieve maximised data rate allocation and appropriate RAT selection for users. In contrast to previous researches, the proposed method jointly optimises RAT selection and data rate allocation using ML integrable optimiser. Evaluation of the MILP solver’s results with many constraints, revealed a 4% performance increment over Heuristic method’s results, with limited constrains and known for its swiftness. The proposed solution was validated with percentage error and standard deviation metrics, whichfurtherestablishtheefficacyoftheproposedsolution, inimprovingnetwork performance, users’ satisfaction, and resource utilization.
Author: Salau, Nurudeen Oladehinbo
Institution: University of the West of Scotland
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7104881789
Cites:
Cited by:
Related to: