Chughtai, Mohsan Niaz (2024) Bandwidth and Power Saving Optimisation of Fronthaul for Beyond-5G Mobile Networks. Doctor of Philosophy (PhD) thesis, University of Kent,. (doi:10.22024/UniKent/01.02.107324) (KAR id:107324)
PDF
Language: English
This work is licensed under a Creative Commons Attribution 4.0 International License.
|
|
Download this file (PDF/11MB) |
Preview |
Official URL: https://doi.org/10.22024/UniKent/01.02.107324 |
Abstract
In the fronthaul segment of beyond-5G mobile networks, the bit rate variability leads to varying end-to-end latency, affecting user throughputs managed by coordinated multipoint beamforming (CoMP), which in turn adds to variability in fronthaul bit rates. To address this, a resource allocation scheme is needed to improve both bandwidth utilisation and power efficiency in fibre optic link components. A model is also required to predict percentile delays within the fronthaul segment to evaluate the impact of variable end-to-end latency on user throughputs.
The fronthaul segment in beyond-5G mobile networks is envisioned to be equipped with FlexE. For efficient bandwidth utilisation in the fronthaul, traffic prediction for slot allocation in FlexE calendars is required. If CoMP is used, the allocation of users to remote units (RUs) with a fronthaul path of lower latency to central/distributed unit (CU/DU) will increase the achievable user bit rate. In the thesis research, the use of multi-agent deep reinforcement
(DRL) learning for optimal slot allocations in a FlexE-enabled fronthaul, for traffic generated through CoMP, and for offloading users among different RUs is explored. It is shown via simulations, that the DRL agent can learn to predict input traffic patterns and allocate slots with the granularity of 5 Gbps in the FlexE calendar, resulting in bandwidth enhancement of up to 11.6 % in comparison to ARIMA-based predictions. A separate DRL agent can also offload UEs from paths of higher latency to lower latency paths, resulting in user throughput enhancement by up to 7.3%.
A model for end-to-end latency for a fronthaul network is also developed in the thesis research and used as an input parameter in a pre-developed model for CoMP-based beamforming in the thesis research. The model is adaptable to different network configurations, traffic parameters and can be extended to predict percentile delays higher than 99.9%. A comparison with simulation results indicates that the model tends to over-predict 99.9 percentile delays by varying degrees (8% to around 150%) as network load fluctuates from 0.1 to 0.9, thus serving as an upper bound on percentile delays, without any under-predictions.
The increasing bit rate demands placed on the fronthaul from high and variable user rates will make the consideration of its power consumption an important issue. DRL-based traffic predictions and slot allocations in a FlexE-enabled fronthaul link can reduce the bit rate, leading to lower transmit signal powers and reduced power consumption in the transmitter of a low-cost, short DMT-type fibre optic fronthaul link. Using DRL predictions and simulations of a DMT modulation EAM-based optical fibre link, it is shown that decreasing bit rates and modulation levels, the transmit signal power can be reduced ranging from 22.3% to 34.6% within a fixed bandwidth of 34 GHz and 18 GHz. Such a transmitter could function as a bandwidth variable transponder in a Flexible Ethernet fronthaul.
Item Type: | Thesis (Doctor of Philosophy (PhD)) |
---|---|
Thesis advisor: | Assimakopoulos, Philippos |
Thesis advisor: | Gomes, Nathan |
DOI/Identification number: | 10.22024/UniKent/01.02.107324 |
Uncontrolled keywords: | 5G-mobile networks, Fronthaul, Reinforcement learning, Resource allocation, FlexE, Discrete Multitone Transmission, Power efficiency, Percentile delay, Latency |
Subjects: | T Technology |
Divisions: | Divisions > Division of Computing, Engineering and Mathematical Sciences > School of Engineering and Digital Arts |
Funders: | University of Kent (https://ror.org/00xkeyj56) |
SWORD Depositor: | System Moodle |
Depositing User: | System Moodle |
Date Deposited: | 25 Sep 2024 07:14 UTC |
Last Modified: | 26 Sep 2024 12:50 UTC |
Resource URI: | https://kar.kent.ac.uk/id/eprint/107324 (The current URI for this page, for reference purposes) |
- Link to SensusAccess
- Export to:
- RefWorks
- EPrints3 XML
- BibTeX
- CSV
- Depositors only (login required):