Modeling and optimization of multiple access for 5G networks and beyond

Abstract

5G has evolved from a set of requirements to a fully specified cellular communication standard in the last five years. 5G’s design goals followed the trend in LTE standards and added support to new services alongside voice and data communications improvements. Naturally, these new services have different quality of service requirements than voice and data applications. To support a myriad of different services with distinct requirements, the design of 5G specifications must be highly flexible, supporting multiple numerologies, frame structures, and random access procedures. These changes demanded a radical redesign of several technologies used in previous networks, including rethinking multiple access. The new services supported by 5G are grouped into three sets of applications based on their requirements: enhanced Mobile Broadband (eMBB), massive Machine Type Communication (mMTC), and Ultra-Reliable Low-Latency Communications (URLLC). While eMBB services mainly refer to the enhancement of current data applications, both mMTC and URLLC address applications not supported by past cellular standards. At the time of writing this thesis, the specification of non-standalone 5G, basically 5G operating on top of legacy LTE networks, is finished. The specification for standalone 5G should be finalized in the coming year. Nonetheless, the development of new cellular standards is a gradual process, and there is much work to be done to fulfill the requirements stipulated for 5G networks. Therefore, the main topic of this work is to propose novel mathematical models to allow the theoretical evaluation of the performance of 5G networks with regards to the multiple access and propose novel optimization methods to guarantee the performance needed for the 5G service. In this setting, the second chapter of this thesis is concerned with optimizing sparse code multiple access (SCMA). SCMA allows the network to schedule more users than orthogonal resources, emerging as an enabler of massive connectivity. We formulate two optimization problems: one seeking to maximize the sum rate of the network and the other the fairness. We formally prove the complexity of both problems and propose two sub-optimal optimization algorithms and compare their performances with available algorithms from the literature. Moreover, we analyze the impact of outdated channel state information on the algorithms performance. The third chapter of this thesis focuses on the modelling of SCMA enabled grant-free access networks. We model this system within the stochastic geometry framework and derive closedform analytical expressions for the area spectral density and the probability of transmission success. Finally, we compare SCMA’s performance and scalability with an orthogonal multiple access system. In the fourth chapter, we focus our efforts on the multiple access aspects of the uplink URLLC service. We consider a grant-free access system operating on the millimetre-wave spectrum with base stations equipped with massive antenna arrays to perform conjugate beamforming to separate the signal from distinct users. We use a spatiotemporal stochastic geometric model to derive closed-form expressions for the system’s reliability and latent access failure probability. Finally, we investigate the impact of the number of base station antennas on the performance and evaluate the suitability of this scheme to satisfy the stringent URLLC quality of service requirements. The fifth chapter of this thesis concerns the distributed link adaptation problem of the uplink transmissions in a grant-free mMTC system. As mMTC applications are power limited, we model the problem as a cross-layer average power minimization under user-specific delay constraints. We formulate the problem as a partially observable stochastic game and propose three different distributed algorithms based on reinforcement learning. We evaluate the performance of the algorithms concerning the network’s average power consumption and delay. Furthermore, we compare the algorithms’ performance with a baseline solution based on a power-boosting reactive hybrid automatic repeat request protocol. We conclude the chapter by analyzing the tradeoffs involving the algorithms’ performance and the signalling overhead they require.

Date	29 Oct 2021
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Georges Kaddoum (Supervisor)