Optimization of multicarrier-division duplex wireless systems

Abstract

Owing to the potentials of enabling to double spectral efficiency (SE) and reduce network latency, inband full duplex (IBFD) has drawn a lot of research so as to substitute the conventional half duplex (HD) of time-division duplex (TDD) and frequency-division duplex (FDD). However, the selfinterference (SI) problem has hindered IBFD from practical deployment. Inspired by the principles of full duplex (FD), in this thesis, a multicarrier-division duplex (MDD) scheme is introduced, which is capable of combining the advantages of IBFD and HD, while simultaneously circumventing their drawbacks. Firstly, in order to make MDD feasible for the operation in large-scale multiple-input multiple-output (MIMO) systems, the thesis commences with addressing the SI problem in propagation-domain with considering the SI cancellation (SIC) requirement of practical analog-to-digital converter (ADC) at receiver. Then, the channel estimation (CE) in MDD MIMO systems is proposed by exploiting the reciprocity and correlation existing between the uplink (UL) and downlink (DL) subchannels. Secondly, the potential of MDD with resource allocation (RA) is first demonstrated, when an unfair greedy algorithm is applied for RA in the multiuser single-input single-output (MU-SISO) systems. Then, a suboptimal algorithm is proposed for MDD millimeter-wave (mmWave) MIMO systems to jointly maximize the sum-rate and achieve the proportional fairness among DL and UL mobile stations (MSs). Two mainstream hybrid precoding strategies are evaluated in the proposed RA scheme and the impact of insufficient SIC on RA is also studied. Thirdly, upon taking the advantages of the flexible time-frequency resource scheduling provided by MDD, two types of frame structures are designed to relieve the channel aging problem in high-mobility communication scenarios. Correspondingly, two Wiener-filtering based predictors (WPs) are introduced under the proposed frame structures for comparing the performance between MDD and TDD, when both CE and residual SI errors are invoked. Moreover, the closed-form expressions for approximating the lower bounded average sum rates of both MDD and TDD systems are derived, when the zero-forcing (ZF) precoding and maximal ratio combining (MRC) are respectively assumed for DL transmission and UL detection. Following the above studies in the context of cellular systems, the synergies between MDD and cellfree massive MIMO (CF-mMIMO) networks are focused. Firstly, a distributed MDD-CF scheme is introduced to enable the FD-style operation but with reduced inter-AP (access point) interference (IAI) and inter-MS interference (IMI). Then, two optimization cases of MDD-CF systems are analyzed by considering simultaneously AP-selection, power- and subcarrier-allocation, under the constraints of individual MSs’ quality of services (QoSs). Specifically, in the first optimization case considering one coherence time (CT) interval, a quadratic transform with successive convex approximation (QTSCA) algorithm is proposed to achieve the SE maximization. By contrast, in the second optimization case on the basis of one radio frame, a two-phase CT (TPCT) interval is designed for MDD-CF systems to guarantee the robust performance over time-varying channels. Correspondingly, a twostep iterative optimization algorithm aided by bisection method is proposed for SE maximization. Our studies show that the proposed QT-SCA algorithm is capable of converging and achieving reliable performance within a few of iterations. However, its complexity increases exponentially with the sizes of CF networks, determined by the numbers of APs, UL/DL users, subcarriers, etc. With this regard and to attain the dynamic power allocation at reduced overhead, a heterogeneous graph neural network (HGNN) is specifically designed for CF networks, which is named as CF-HGNN. The CF-HGNN consists of the adaptive node embedding layer, message passing layer, attention layer and the downstream power allocation layer. Our studies show that CF-HGNN is scalable to the MDDCF networks with various numbers of nodes and subcarriers. Furthermore, when assisted by the proposed user clustering, the CF-HGNN trained based on a small CF network can be applied to the large-scale MDD-CF networks, which may cover large area, have a big number of subcarriers and/or simultaneously support a big number of nodes.

More information

Identifiers

ORCID for Bohan Li: orcid.org/0000-0001-7686-8605

ORCID for Lie-Liang Yang: orcid.org/0000-0002-2032-9327

ORCID for Rob Maunder: orcid.org/0000-0002-7944-2615

Catalogue record

Export record