Mixed-numerology for radio access network slicing

Yang, Bowen (2021) Mixed-numerology for radio access network slicing. PhD thesis, University of Glasgow.

Full text available as:
[thumbnail of 2021YangPhD.pdf] PDF
Download (2MB)

Abstract

Network slicing is a sustainable solution to support the various service types in future networks. In general, network slicing is composed of core network slicing and radio access network (RAN) slicing. The former can be realized by allocating dedicated virtualized core network functionalities to specific slices. Similarly, RAN slicing includes the virtualization and allocation of the limited RAN resources. From the physical layer perspective, supporting RAN slicing implies the need of unique radio-frequency (RF) and baseband (BB) configurations, i.e., numerology, for each slice to fulfil its quality of service requirements. To support such a heterogeneous mixed-numerology (MN) system, the transceiver architecture and widely used signal processing algorithms in the traditional single-service system need to be significantly changed. A clear understanding of mixed-numerology signals multiplexing and isolation is of importance to enable spectrum and computation efficient RAN slicing. Meanwhile, an effective channel estimation is the guarantee of performing almost all receiver signal processing. Fundamental channel estimation investigations also constitute a crucial piece of MN study.

This thesis aims to systematically investigate the OFDM-based MN wireless communication systems in terms of system modeling, channel equalization/ estimation, and power allocation. First, a comprehensive mixed-numerology framework with two numerologies is proposed and characterized by physical layer parameters. According to the BB and RF configurations imparities among numerologies, four scenarios are categorized and elaborated on the configuration relationships of different numerologies. System models considering the most generic scenario are established for both uplink and downlink transmissions. Two theorems are proposed as the basis of MN algorithms design, which generalize the original circular convolution property of the discrete Fourier transform. The proposed theorems verifies the feasibility of the one-tap channel equalization in MN systems. However, they also indicate that both BB and RF configuration differences result in inter-numerology-interference (INI). Besides, severe signal distortion may occur when the transmitter and receiver numerologies are different. Therefore, a pre-coding algorithm is designed by utilizing the theorems to compensate the system degradation resulting from the signal distortion. INI cancellation algorithms are proposed based on collaboration detection scheme and joint numerologies signal models for downlink and uplink, respectively. Numerical results shows that the proposed algorithms are able to significantly improve the system performance.

Another objective of this thesis is to verify the effectiveness of the existing channel estimation algorithms and to develop new ones in the presence of MN. To achieve these goals, three channel estimation methods, i.e., least-square linear interpolation, least-square ‘sinc’ interpolation, and minimum mean square error ‘sinc’ interpolation are implemented and theoretically analyzed in both single-user and multi-user scenarios. The analysis reveals that the pilot signal to noise ratio, pilot distance, and position of pilot signals jointly affect the channel estimation. In particular, a signal distortion factor caused by the RF configuration difference is spotted to seriously affect the channel estimation performance, whose values are mainly decided by the degree of configuration mismatch. On the other hand, INI also degrades the channel estimation in the MN system. The existence of interference-free subcarriers is demonstrated based on the derived closed-form expression of the INI. Pilot design principles in terms of pilot signal placement are developed according to the analyses. Numerical results shows that minimum mean square error based channel estimation has the best performance and robustness to the configuration mismatch. In addition, the proposed pilot design principles could produce comparable channel estimation results with the legacy OFDM systems where no INI and signal distortion exist.

The two problems associated with the MN system, i.e., signal distortion and INI, could negatively affect the power distribution of the received MN signals, and the system performance in terms of spectrum efficiency may be seriously degraded. Consequently, it becomes outstandingly important to introduce an efficient subcarrier-level power allocation scheme in such kinds of systems to counter the performance degradation caused by the configuration mismatch. As such, this thesis makes the attempt to extend the two-numerology model to contain ‘M’ different numerologies. Based on the model, closed-form expressions of desired signal, interference, and noise are derived. The derivation shows that interference generated from different numeroloies are linearly superimposed in the frequency domain. The distribution of signal-to-interference-plus-noiseratio (SINR) is analyzed theoretically. An iterative convex approximation power allocation algorithm is proposed by applying the derived SINR. Results show that the power allocation algorithm contributes to remarkable spectrum efficiency improvement compare to the other schemes, and an extra subband filtering process could bring about even higher performance.

The work presented in this thesis provides guidance for multi-numerology system design in terms of parameter selection, and the frame structure and algorithms design. Moreover, it presents a solution as to how the radio access network slicing can be underpinned in the physical layer in a spectrum efficient way.

Item Type: Thesis (PhD)
Qualification Level: Doctoral
Colleges/Schools: College of Science and Engineering > School of Engineering
Supervisor's Name: Zhang, Dr. Lei, Imran, Professor Muhammad Ali and Sun, Dr. Yao
Date of Award: 2021
Depositing User: Theses Team
Unique ID: glathesis:2021-82683
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 04 Feb 2022 15:16
Last Modified: 08 Apr 2022 16:52
Thesis DOI: 10.5525/gla.thesis.82683
URI: https://theses.gla.ac.uk/id/eprint/82683
Related URLs:

Actions (login required)

View Item View Item

Downloads

Downloads per month over past year