Graph-Embedded Multi-Agent Learning for Smart Reconfigurable THz MIMO-NOMA Networks

Xu, Xiaoxia; Chen, Qimei; Mu, Xidong; Liu, Yuanwei; Jiang, Hao

doi:10.1109/jsac.2021.3126079

Cited by 31 publications

(16 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In [23], the authors compared the RIS-assisted NOMA scheme and the RIS-assisted zero-forcing beamforming (ZFBF) transmission scheme and then identified the best scenarios to adopt NOMA or ZFBF. In contrast to the alternating optimization techniques, the authors in [24] conceived a novel smart reconfigurable terahertz (THz) multiple-input multipleoutput (MIMO)-NOMA framework, where a novel multi-agent deep reinforcement learning algorithm was proposed by exploiting the decentralized partially-observable Markov decision process. As a step further, in [25], both a deep learning approach and a reinforcement learning approach were developed for the RIS-Assisted NOMA networks to maximize the effective throughput of the entire transmission period.…”

Section: A Related Work 1) Ris-enabled Noma Communicationsmentioning

confidence: 99%

“…Since the final relaxed problem without the rank-one constraint is a standard semidefinite program (SDP) [39], it can be efficiently solved by well-known convex optimization tools, such as the CVX [40]. The objective value obtained from problem (25) yields a lower bound of that from problem (22) owing to the relaxation in (24). After the solution is derived, we can get the beamforming vector {w k } via Cholesky decomposition as W k = w k w H k .…”

Section: A Active Beamforming Optimization At the Bsmentioning

confidence: 99%

“…The manipulation for the non-convex constraint (21d) is similar to the methods used in Sec. IV-A, from which they be approximated as (24). Let e m denote an M-dimensional column vector with the m-th element being 1 and all others being 0.…”

Section: B Passive Beamforming Optimization At the Star-rismentioning

confidence: 99%

See 2 more Smart Citations

Queue-Aware STAR-RIS Assisted NOMA Communication Systems

Zhang¹,

Liu²,

Mu³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Simultaneously transmitting and reflecting reconfigurable intelligent surfaces (STAR-RISs) have been receiving great attention nowadays due to the capability of achieving full-space coverage. In this paper, the queue-aware STAR-RIS assisted non-orthogonal multiple access (NOMA) communication system is investigated to ensure the system stability. To tackle the challenge of infinite time periods required for stability, the long-term stability-oriented problem is reformulated as a queue-weighted sum rate (QWSR) maximization problem in each single time slot based on the Lyapunov drift theory. In particular, the rate weight allocated to each user is determined by the length of a data queue, which is maintained at the base station (BS) and pending delivery to each user. Then, the QWSR is maximized by jointly optimizing the NOMA decoding order, the active beamforming coefficients (ABCs) at the BS, and the passive transmission and reflection coefficients (PTRCs) at the STAR-RIS, where three STAR-RIS operating protocols are considered, namely energy splitting (ES), mode switching (MS), and time switching (TS). For ES, to handle the highly-coupled and non-convex problem, the blocked coordinate descent and the successive convex approximation methods are invoked to iteratively and alternatively optimize the problem. Moreover, the proposed iterative algorithm is further extended to a penalty-based two-loop algorithm to solve the binary amplitude constrained problem for MS. For TS, the formulated problem is decomposed into two subproblems, each of which can be solved in a similar manner as introduced for ES. Simulation results show that: i) our proposed STAR-RIS assisted NOMA communication achieves better performance compared with the conventional schemes; ii) the reformulated QWSR maximization problem is proven to ensure the system stability; and iii) the TS protocol achieves superior performance with respect to both the QWSR and the average queue length.

show abstract

Section: A Related Work 1) Ris-enabled Noma Communicationsmentioning

confidence: 99%

Section: A Active Beamforming Optimization At the Bsmentioning

confidence: 99%

See 1 more Smart Citation

Queue-Aware STAR-RIS Assisted NOMA Communication Systems

Zhang¹,

Liu²,

Mu³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Alternatively, the authors of [13] unfolded a power allocation enabled iterative weighted minimum mean squared error (WMMSE) algorithm with a distributed GNN architecture, which achieves higher robustness and generalizability in unseen scenarios. In reconfigurable intelligent surface (RIS) aided terahertz massive MIMO-NOMA networks, the authors of [14] integrated the graph neural network into distributed multi-agent deep reinforcement learning architecture to facilitate information interaction and coordination. Moreover, the authors of [15] learned a distributed heterogeneous GNN over wireless interference graph with a parameter sharing scheme, which enables more efficient scheduling than homogeneous GNNs.…”

Section: A Prior Workmentioning

confidence: 99%

Distributed Auto-Learning GNN for Multi-Cell Cluster-Free NOMA Communications

Xu¹,

Liu²,

Chen³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

A multi-cell cluster-free NOMA framework is proposed, where both intra-cell and inter-cell interference are jointly mitigated via flexible cluster-free successive interference cancellation (SIC) and coordinated beamforming design, respectively. The joint design problem is formulated to maximize the system sum rate while satisfying the SIC decoding requirements and users' data rate constraints. To address this highly complex and coupling non-convex mixed integer nonlinear programming (MINLP), a novel distributed auto-learning graph neural network (AutoGNN) architecture is proposed to alleviate the overwhelming information exchange burdens among base stations (BSs). The proposed AutoGNN can train the GNN model weights whilst automatically learning the optimal GNN architecture, namely the GNN network depth and message embedding sizes, to achieve communication-efficient distributed scheduling. Based on the proposed architecture, a bi-level AutoGNN learning algorithm is further developed to efficiently approximate the hypergradient in model training. It is theoretically proved that the proposed bi-level AutoGNN learning algorithm can converge to a stationary point. Numerical results reveal that: 1) the proposed cluster-free NOMA framework outperforms the conventional clusterbased NOMA framework in the multi-cell scenario; and 2) the proposed AutoGNN architecture significantly reduces the computation and communication overheads compared to the conventional convex optimization-based methods and the conventional GNN with a fixed architecture.

show abstract

“…From the resource allocation perspective, the issue of user clustering, i.e., which THz users are to be grouped together for the implementation of NOMA, has been studied in [12]- [14]. The application of advanced machine learning methods to resource allocation in THz-NOMA networks has also been investigated in [15], where intelligent reflecting surfaces have been used to reconfigure the wireless propagation environment. Unlike these existing works about THz-NOMA, this paper focuses on how to use NOMA as a type of add-on in THz networks.…”

mentioning

confidence: 99%

Joint Beam Management and Power Allocation in THz-NOMA Networks

Ding¹,

Poor²

2022

Preprint

View full text Add to dashboard Cite

This paper investigates how to apply non-orthogonal multiple access (NOMA) as an add-on in terahertz (THz) networks. In particular, prior to the implementation of NOMA, it is assumed that there exists a legacy THz system, where spatial beams have already been configured to serve legacy primary users. The aim of this paper is to study how these pre-configured spatial beams can be used as a type of bandwidth resources, on which additional secondary users are served without degrading the performance of the legacy primary users. A joint beam management and power allocation problem is first formulated as a mixed combinatorial non-convex optimization problem, and then solved by two methods with different performance-complexity tradeoffs, one based on the branch and bound method and the other based on successive convex approximation. Both analytical and simulation results are presented to illustrate the new features of beam-based resource allocation in THz-NOMA networks and also demonstrate that those pre-configured spatial beams can be employed to improve the system throughput and connectivity in a spectrally efficient manner. Index TermsNon-orthogonal multiple access (NOMA), Terahertz (THz), the branch and bound method, successive convex approximation, beam management, power allocation. I. INTRODUCTION Terahertz (THz) communications and non-orthogonal multiple access (NOMA) are two key enabling technologies for the envisioned sixth generation (6G) mobile network [1], [2]. On the one hand, the use of THz communications is promising because a huge amount of bandwidth in the THz spectrum is available for communications [3]-[6]. On the other hand, the use of NOMA transmission can significantly improve spectral efficiency and support massive connectivity, by encouraging intelligent spectrum cooperation among mobile users [7], [8]. The two communication techniques are naturally complementary to each other. For example, using Z. Ding and H. V.

show abstract

Graph-Embedded Multi-Agent Learning for Smart Reconfigurable THz MIMO-NOMA Networks

Cited by 31 publications

References 31 publications

Queue-Aware STAR-RIS Assisted NOMA Communication Systems

Queue-Aware STAR-RIS Assisted NOMA Communication Systems

Distributed Auto-Learning GNN for Multi-Cell Cluster-Free NOMA Communications

Joint Beam Management and Power Allocation in THz-NOMA Networks

Contact Info

Product

Resources

About