Date of Award

31-8-2024

Document Type

Thesis

School

School of Computing

Programme

Ph.D.-Doctoral of Philosophy

First Advisor

Dr.A.Umamakeswari

Keywords

5G, QoS, Resource Allocation, Machine Learning, Deep Learning

Abstract

Wireless network systems must have effective resource allocation, particularly in the context of 5G networks when flexibility is needed to meet a range of network requirements. Resource allocation is essential in cellular network contexts to guarantee equitable access to customers, partners, and cellular service users. Since resource distribution determines network performance, it offers significant advantages when executed well. One of the biggest issues with 5G technology is resource allocation, particularly when it comes to the Quality of Service (QoS) for various applications. Resources in wireless networks include items like channels, power, and spectrum; these must all be apportioned according to user needs. The current cellular networks feature rigid services and limited infrastructure, which makes resource allocation difficult and inefficient. Wireless networks have user-centric resources that must be distributed effectively to provide QoS. As far as efficient resource allocation is concerned, it is imperative to ensure QoS to the user.

The next generation Node B (gNB) employs a number of wireless resource scheduling techniques, including Proportional Fairness (PF), Best Channel Quality Indicator (BCQI), and Round Robin (RR) techniques. Using the Channel State Information (CSI) provided by the terminal and the throughput of the most recent Transmission Time Interval (TTI), the base station uses the PF algorithm to modify scheduling settings. On the other hand, unexpected, transient events could lead to irregularities in throughput statistics or channel conditions, which would cause imbalances and errors in resource distribution.

Building scheduling algorithms thus depends on choosing the right Modulation and Coding Scheme (MCS) to predict network performance in upcoming TTIs based on long-term network performance as a baseline. Based on the Reinforcement Learning (RL) approach, which guarantees MCS selection based on dynamic channel conditions, the Reinforcement Learning based Loop Link Adaptation (ReL-LLA) model is suggested for appropriate MCS selection. The suggested model is built using the gNB side's link adaptation technique, which ensures that the User Equipment's (UE's) required data rate is maintained and modified in response to shifting channel conditions.

The suggested model builds the RL model using the fundamental Deep Deterministic Policy Gradient (DDPG). The computation of the incentive function to guarantee the proposer MCS selection is what makes the suggested model new. Resource Blocks (RBs) are dispersed equally among all UEs in a network thanks to resource allocation techniques at gNB. It has been suggested that PF Scheduling (PFS) approaches be used to strike a balance between performance measurements, especially UE throughput and fairness. In addition, these systems, in contrast to previous scheduling algorithms like RR and BCQI, base their priority function on indications of the historical and present UE channel quality, namely the instantaneous and average data rate.

The priority function of the connected UEs, which is determined at each TTI and is based on the weighted past achievable data rate and the weighted present data rate with the highest MCS index, forms the basis of the suggested adaptive resource scheduler. The Grid Search LSTM (Gr-LSTM) model is the suggested approach for UE priority prediction for the next TTI. Choosing the MCS based on the suggested ReL LLA is one of the Gr-LSTM's input functions. The suggested model employs a greedy strategy to methodically investigate the ideal hyper-parameter in a hierarchical fashion. The restricted number of data samples in real-time may have an impact on the UE priority prediction. The TL approach is applied in the proposed enhanced UE priority prediction model. The Gr-LSTM model executing the target task receives its parameters from the suggested model, which transfers them from the Gr-LSTM model executing the source job.

Download

Share

COinS