PaperlayerView exact revision on arXiv

Source-linked AI summary

Evolution of NOMA Toward Next Generation Multiple Access (NGMA) for 6G

Yuanwei Liu, Shuowen Zhang, Xidong Mu, Zhiguo Ding, Robert Schober, Naofal Al-Dhahir, Ekram Hossain, Xuemin Shen

arXiv:2108.04561v2cs.ITeess.SP

TL;DR

The paper addresses how multiple access can support rapidly growing, heterogeneous, and massive wireless connectivity for 6G. It surveys NOMA’s capacity foundations and related techniques, develops a multi-antenna NOMA-based NGMA framework for downlink and uplink, and identifies implementation challenges. Its supported conclusion is that NGMA research can build on NOMA while requiring further work on practical limitations and broader multi-antenna settings.

  • Problem

    6G networks require multiple access schemes that efficiently support massive connectivity, heterogeneous services, high bandwidth efficiency, and low latency.

  • Method

    The paper surveys NOMA capacity limits, candidate NGMA techniques, applications, multi-antenna methods, optimization and machine learning, then proposes a unified multi-antenna NOMA framework for downlink and uplink.

  • Results

    The paper establishes a NOMA-based foundation for NGMA and identifies practical challenges including SIC error propagation, optimization trade-offs, and extensions to general multi-user MIMO.

  • Takeaways & Limitations

    NOMA provides a basis for NGMA research, but practical deployment and broader multi-antenna generalization remain important research directions.

  • Takeaways & Limitations

    In static or quasi-static scenarios, machine-learning optimality cannot generally be theoretically proved or strictly guaranteed, unlike conventional mathematical optimization.

Abstract

from arXiv · show

Due to the explosive growth in the number of wireless devices and diverse wireless services, such as virtual/augmented reality and Internet-of-Everything, next generation wireless networks face unprecedented challenges caused by heterogeneous data traffic, massive connectivity, and ultra-high bandwidth efficiency and ultra-low latency requirements. To address these challenges, advanced multiple access schemes are expected to be developed, namely next generation multiple access (NGMA), which are capable of supporting massive numbers of users in a more resource- and complexity-efficient manner than existing multiple access schemes. As the research on NGMA is in a very early stage, in this paper, we explore the evolution of NGMA with a particular focus on non-orthogonal multiple access (NOMA), i.e., the transition from NOMA to NGMA. In particular, we first review the fundamental capacity limits of NOMA, elaborate on the new requirements for NGMA, and discuss several possible candidate techniques. Moreover, given the high compatibility and flexibility of NOMA, we provide an overview of current research efforts on multi-antenna techniques for NOMA, promising future application scenarios of NOMA, and the interplay between NOMA and other emerging physical layer techniques. Furthermore, we discuss advanced mathematical tools for facilitating the design of NOMA communication systems, including conventional optimization approaches and new machine learning techniques. Next, we propose a unified framework for NGMA based on multiple antennas and NOMA, where both downlink and uplink transmissions are considered, thus setting the foundation for this emerging research area. Finally, several practical implementation challenges for NGMA are highlighted as motivation for future work.

I. INTRODUCTION

NGMA is motivated by rapidly growing connectivity demands and the limitations of orthogonal access. The paper examines the transition from NOMA to NGMA and surveys its capacity foundations, techniques, tools, and unified framework.

  • 13.1 billion mobile users and 29.3 billion Internet-enabled devices were predicted by the end of 2023, intensifying pressure on 4G and 5G connectivity.
  • NGMA aims to connect tremendous numbers of users and devices efficiently and flexibly over shared wireless resources.
  • Orthogonal access has become inefficient because limited spectrum constrains spectral efficiency and the number of supportable users.
  • The paper reviews NOMA capacity limits, NGMA requirements and candidates, multi-antenna NOMA, applications, emerging-technology interactions, and optimization and machine-learning tools.
  • NOMA achieves the capacity region of the general Gaussian MISO/MIMO MAC through successive interference cancellation, whereas the general MISO/MIMO broadcast-channel capacity remains unknown.

B. New Considerations for NGMA Design

NGMA design must address massive access, finite-blocklength transmission, and the shift from orthogonal to non-orthogonal resource sharing. Candidate schemes include power-, code-, and spatial-domain approaches.

  • Massive-access capacity limits must be revisited when the number of users grows with coding blocklength.
  • Short-packet IoT transmission requires finite-blocklength analysis because low latency limits coding blocklength.
  • NGMA trends toward non-orthogonal access because orthogonal schemes are strictly suboptimal and support only limited users over available resources.
  • Power-domain NOMA serves users on shared time, frequency, and code resources by distinguishing them through power, using superposition coding and successive interference cancellation.
  • Code-domain NOMA uses sparse or low-cross-correlation user-specific spreading sequences with iterative message-passing multiuser detection.
  • Spatial-division multiple access serves users on shared time, frequency, and code resources while distinguishing them spatially through multiple antennas.

3) Space Division Multiple Access (SDMA):

Multi-antenna NOMA exploits spatial degrees of freedom through beamformers, while preserving NOMA’s applicability to overloaded and critically loaded systems. Its design must also satisfy SIC-related constraints.

  • Multi-antenna techniques add spatial degrees of freedom that can enhance NGMA performance compared with single-antenna systems.
  • MIMO-NOMA beamformer design controls desired and interference power, thereby affecting users’ signal-to-interference-plus-noise ratios.
  • Beamformer-based MIMO-NOMA: Beamformer-based MIMO-NOMA assigns a linear beamformer to each user and adds constraints to ensure effective successive interference cancellation.
  • Beamformer-based MIMO-NOMA: Beamformer-based MIMO-NOMA applies to underloaded, critically loaded, and overloaded regimes.
  • Beamformer-based MIMO-NOMA: Beamformer-based designs require jointly optimizing decoding order and beamformers, whose complexity grows exponentially with the number of users.

2) Cluster-based MIMO-NOMA:

Cluster-based MIMO-NOMA reduces beamformer and SIC complexity by grouping users with similar spatial characteristics. Research extends this design toward interference management, massive MIMO, high-frequency bands, and applications including UAV communications.

  • Cluster-based MIMO-NOMA: Cluster-based MIMO-NOMA groups users into clusters that share beamformers, reducing the complexity of fully joint user-level design.
  • Cluster-based MIMO-NOMA: Grouping users with similar spatial features can suppress or eliminate inter-cluster interference and use limited spatial degrees of freedom more efficiently.
  • Cluster-based MIMO-NOMA: Zero-forcing beamforming has been used to remove inter-cluster interference and support efficient user-clustering algorithms.
  • Cluster-based MIMO-NOMA: Optimized clustered MIMO-NOMA showed superior sum achievable rates to MIMO-OMA, while other work jointly optimized power allocation and user admission for energy efficiency under QoS constraints.
  • Massive MIMO-NOMA: Massive MIMO-NOMA can improve secrecy rate and energy efficiency over conventional massive MIMO-OMA, but accurate CSI acquisition creates substantial training and feedback overhead.
  • mmWave/THz MIMO-NOMA: NOMA is also studied with mmWave and THz technologies, where beamspace MIMO and hybrid precoding address high-frequency hardware complexity and RF-chain requirements.
  • UAV-BS-enabled NOMA: UAV-enabled NOMA supports heterogeneous QoS and massive connectivity by adjusting channel conditions through UAV positioning and sharing resources among users.

2) NOMA-assisted cellular-connected UAVs:

NOMA-assisted cellular-connected UAVs share spectrum with ground users, creating severe interference that ground-aerial uplink NOMA can help manage through asymmetric channel conditions. The supplied passages also survey NOMA applications in robotics, machine-type communications, and other emerging network scenarios.

  • 2) NOMA-assisted cellular-connected UAVs:: Cellular-connected UAVs act as aerial users sharing spectrum with ground users, avoiding dedicated spectrum allocation but creating severe interference.The shared-spectrum setting is important for alleviating spectrum shortage, but it imposes interference-management requirements.
  • 2) NOMA-assisted cellular-connected UAVs:: Ground-aerial uplink NOMA exploits stronger line-of-sight aerial channels by decoding and subtracting the UAV signal before decoding ground-user signals.Related work applies this framework to cooperative interference handling and trajectory optimization for cellular-connected UAVs.
  • B. NOMA-Enhanced Robotic Communications: Connected robots rely on information exchange with APs and BSs, but their more blockage-prone channels produce rapidly changing conditions compared with aerial links.This makes robotic communications a related but more challenging application scenario.
  • C. Massive and Critical Machine-Type Communication (MC-MTC): Grant-Free random access removes grant acquisition, allowing devices to transmit without waiting for permission and enhancing connectivity relative to grant-based access.GF-NOMA and semi-GF NOMA are discussed as random-access approaches for machine-type communications.
  • C. Massive and Critical Machine-Type Communication (MC-MTC): Multi-transmit-power and DQN-based power-control methods improve reported data-rate performance, but current random-access NOMA work mainly focuses on throughput rather than latency or reliability.The passages identify efficient random-access NOMA concepts for other MTC QoS metrics as an open need.
  • Other 6G application scenarios: NOMA is also discussed for MEC and e-health, where it supports task transmission, computation-result delivery, and coordination among numerous smart devices.These applications are motivated by high speed, low latency, and healthcare connectivity requirements.

2) E-health:

RISs provide low-cost, low-power reconfigurable control of wireless propagation and can be integrated with NOMA to create desired channel orders and additional design degrees of freedom. The supplied passages report stronger NOMA gains than OMA in RIS-assisted settings while also identifying challenges for mobility and channel resolution.

  • A. Reconfigurable Intelligent Surface (RIS)-NOMA: RISs use reconfigurable elements and a smart controller to modify signal propagation without RF chains, reducing hardware cost and power consumption.They can be deployed on structures such as building facades, indoor walls, billboards, and windows.
  • A. Reconfigurable Intelligent Surface (RIS)-NOMA: RIS-aided QoS-based NOMA can order users by QoS requirements and shape their channel conditions to realize the desired decoding order.This addresses cases where a weaker-channel user requires a higher data rate than a stronger-channel user.
  • A. Reconfigurable Intelligent Surface (RIS)-NOMA: RIS-NOMA research extends to dynamic RIS configuration, multi-antenna transmission, joint active-passive beamforming, and lower-complexity user ordering.Exhaustive-search decoding-order optimization may be computationally unacceptable in multi-antenna RIS-NOMA.
  • A. Reconfigurable Intelligent Surface (RIS)-NOMA: NOMA significantly outperforms TDMA and FDMA in maximum weighted sum rate, with larger gains as the number of RIS elements increases.The comparison concerns an RIS-aided 4-user communication system with blocked direct AP-user links and optimized RIS placement.
  • A. Reconfigurable Intelligent Surface (RIS)-NOMA: Asymmetric RIS deployment is preferable for NOMA, whereas symmetric deployment is superior for OMA.The deployment preference differs across the two multiple-access schemes.
  • Mobility-related challenges: High-mobility users create Doppler and resource-occupation challenges, while OTFS performance is limited by delay-Doppler resolution.Improving OTFS resolution can require longer transmission durations and more frequency channels.

C. Integrated Sensing and Communication (ISaC)-NOMA

NOMA is presented as a flexible access technique for 6G scenarios including integrated sensing and communication, coordinated multipoint, full-duplex, and visible-light communications. The section also surveys mathematical optimization tools for designing NOMA resource allocation.

  • C. Integrated Sensing and Communication (ISaC)-NOMA: NOMA can provide additional access channels within the same time/frequency resource block for integrated sensing and communication.This may help address the challenge of combining communication and sensing signals in shared resources.
  • C. Integrated Sensing and Communication (ISaC)-NOMA: NOMA-aided joint radar and multicast-unicast communication significantly outperforms conventional SDMA and TDMA in both radar detection and communication performance.
  • 1) Coordinated Multi-Point (CoMP)-NOMA:: CoMP-NOMA can mitigate inter-cell interference for cell-edge users while enabling flexible resource sharing involving cell-centre users and different cells.
  • 2) Full-Duplex (FD)-NOMA:: FD-NOMA improves resource efficiency by supporting simultaneous downlink and uplink transmission, but requires effective co-channel interference suppression.
  • 3) NOMA for VLC:: NOMA is attractive for indoor VLC because limited users, high transmit SNR, and slowly varying CSI can reduce SIC, increase throughput gains, and lower transmission complexity.
  • Resource allocation for NOMA: NOMA resource allocation is addressed using mathematical optimization methods covering convex problems, non-convex problems, numerical methods, and analytical methods.Convex formulations can generally yield globally optimal solutions with polynomial time complexity.

2) Non-convex optimization methods:

The paper surveys non-convex optimization methods for NOMA resource allocation, emphasizing structure exploitation, iterative approximations, global-search methods, and decomposition strategies. It also positions machine learning as useful for diverse and time-varying NOMA resource-allocation problems.

  • 2) Non-convex optimization methods:: Most NOMA resource-allocation problems are non-convex because of diverse user demands and complicated network structures.
  • Matching theory: Matching theory models NOMA sub-channel allocation as a many-to-many matching game and supports low-complexity algorithms.
  • Successive Convex Approximation (SCA): Successive Convex Approximation finds high-quality suboptimal solutions by iteratively replacing non-convex objectives or constraints with convex tight surrogate functions.
  • Branch-and-Bound (BnB): Branch-and-Bound seeks globally optimal solutions for non-convex problems by enumerating potentially optimal solutions, often with exponential complexity.
  • Monotonic Optimization: Monotonic optimization can find globally optimal solutions efficiently when objectives and constraints are monotonic in the optimization variables.
  • Block Coordinate Descent (BCD): Block Coordinate Descent decomposes coupled joint optimization into smaller subproblems by iteratively optimizing variable blocks while fixing the others.Its convergence and solution quality require evaluation for the specific problem structure.
  • Machine learning for NOMA: Machine learning is surveyed alongside mathematical optimization for NOMA resource allocation, including deep learning applications to CSI acquisition, detection, clustering, and allocation.

2) Reinforcement learning for NOMA:

The paper reviews reinforcement learning and related machine-learning approaches for NOMA resource allocation in dynamic, heterogeneous, and distributed wireless environments. It contrasts optimization and learning methods and distinguishes when DL or DRL is better suited.

  • C. Discussion and Outlook: Conventional optimization can be computationally expensive, system-dependent, and inaccurate in rapidly time-varying wireless environments.
  • 2) Reinforcement learning for NOMA:: Reinforcement learning learns from dynamic or uncertain environments and historical experience by maximizing long-term reward.
  • 2) Reinforcement learning for NOMA:: RL-based NOMA research includes power allocation, throughput maximization, UAV communications, and NOMA-MEC resource allocation.
  • Federated learning for NOMA: Federated learning supports distributed NOMA training by exchanging model parameters rather than sensitive raw data.
  • Machine learning for NOMA: DL and DRL have been applied across NOMA problems, while transfer learning, meta-learning, supervised learning, and unsupervised learning are additional approaches.
  • C. Discussion and Outlook: ML is favored in complex time-varying or unknown environments, whereas mathematical optimization retains advantages in static or quasi-static scenarios because its optimality can be guaranteed.
  • C. Discussion and Outlook: DRL is particularly suitable for Markov decision processes such as UAV trajectory design and long-term resource allocation with varying constraints.

VII. ROAD AHEAD: A MULTI-ANTENNA AND NOMA-BASED UNIFIED FRAMEWORK FOR NGMA

The paper proposes a unified NGMA framework combining multi-antenna transmission and NOMA for downlink and uplink communications. It analyzes SIC requirements and shows that the relative advantage of NOMA versus SDMA depends on channel geometry.

  • VII. ROAD AHEAD: A MULTI-ANTENNA AND NOMA-BASED UNIFIED FRAMEWORK FOR NGMA: The unified NGMA framework combines multi-antenna transmission and NOMA for both downlink and uplink multi-user communication.It includes signal models and implementation principles for each transmission scenario.
  • SIC in MIMO-NOMA: Successful SIC requires the strong user’s decoding rate for the weak signal to be at least the weak user’s corresponding decoding rate.The paper states this as Rw→s ≥ Rw→w.
  • SIC in MIMO-NOMA: In multi-antenna systems, user ordering cannot be determined solely from scalar channel power gains because user channels are vectors or matrices.
  • System regimes: The framework considers underloaded or critically loaded systems with N ≥ K and overloaded systems with N < K.
  • Scenario 1: When two user channels are orthogonal, beamformers can eliminate inter-user interference, making SIC and user ordering unnecessary.
  • Scenario 2: When user channels are highly correlated, SDMA suffers severe inter-user interference, whereas NOMA can remove one user’s interference through SIC and outperform SDMA.

2) Overloaded system:

In overloaded multi-antenna systems, NOMA combines spatial separation and SIC to serve users beyond available spatial degrees of freedom. The unified NGMA framework generalizes SDMA and NOMA through adaptable user grouping and beamforming.

  • 2) Overloaded system:: When N=2 and K=4, SDMA suffers severe inter-user interference, whereas cluster-based NOMA groups correlated channels and uses two beamformers.Users 1 and 3 form one cluster, while users 2 and 4 form another.
  • 2) Overloaded system:: NOMA mitigates inter-cluster interference spatially and intra-cluster interference through SIC, allowing two additional users to be served simultaneously.The strong user in each cluster can receive its intended signal interference-free under the described design.
  • 2) Overloaded system:: The framework groups K users into an optimizable number M of clusters, with binary variables specifying intra-cluster SIC decoding order.M=K yields one user per cluster, while M=1 groups all users together.
  • 2) Overloaded system:: The unified framework contains SDMA, beamformer-based NOMA, and cluster-based NOMA as special cases through different choices of M.SDMA uses M=K, beamformer-based NOMA uses M=1, and cluster-based NOMA uses 1<M<K.
  • 2) Overloaded system:: By combining multiple antennas with NOMA, unified NGMA provides flexible transmission across underloaded, critically loaded, and overloaded regimes instead of switching between fixed schemes.The framework is intended to provide enhanced degrees of freedom across these loading conditions.

C. A Unified Framework for NGMA: Uplink Case

The uplink framework models K single-antenna users transmitting to an N-antenna base station. It represents reception through detection vectors and motivates a unified design after reviewing SDMA- and NOMA-based detection.

  • C. A Unified Framework for NGMA: Uplink Case: In the uplink, K>1 single-antenna users transmit information to an N-antenna base station over their respective channels and powers.The received signal includes the users’ transmitted signals and additive white Gaussian noise.
  • C. A Unified Framework for NGMA: Uplink Case: Unlike downlink users, the uplink base station decodes all K information streams because every received stream is desired.A normalized detection vector is assigned for decoding each stream.
  • C. A Unified Framework for NGMA: Uplink Case: The framework first reviews SDMA-based and NOMA-based uplink signal detection schemes before introducing unified NGMA uplink transmission.The review addresses the benefits and drawbacks of the existing schemes.

1) SDMA (Parallel detection):

Uplink SDMA uses parallel detection, with the base station directly decoding each stream while attempting to mitigate interference through its detection vector. It is best suited to underloaded systems with sufficiently uncorrelated channels.

  • 1) SDMA (Parallel detection):: Uplink SDMA directly decodes each user’s signal in the presence of interference from all other users.Each information stream is detected separately using its own detection vector.
  • 1) SDMA (Parallel detection):: SDMA detection vectors must completely or substantially mitigate inter-user interference, which is possible only when K≤N.This condition favors underloaded systems with sufficient spatial degrees of freedom.
  • 1) SDMA (Parallel detection):: SDMA offers parallel processing and therefore low latency, but its performance degrades in overloaded systems or when user channels are strongly correlated.Interference mitigation becomes ineffective in those conditions.

2) NOMA (Serial detection):

Uplink NOMA uses serial SIC detection at the base station, decoding streams sequentially and subtracting already decoded signals. Its decoding order affects achievable rates and can be optimized for user requirements.

  • 2) NOMA (Serial detection):: The base station detects one uplink signal at a time and subtracts each decoded signal before decoding subsequent streams.Later-decoded signals avoid interference from streams already decoded.
  • 2) NOMA (Serial detection):: Unlike downlink NOMA, uplink NOMA has no SIC condition between users because one receiver decodes all signals, allowing any decoding order.The order still affects each stream’s achievable communication rate.
  • 2) NOMA (Serial detection):: Uplink NOMA is mainly beneficial in overloaded systems, but it can also operate in underloaded or critically loaded systems without lower communication rates than SDMA.Its SIC can further cancel interference that SDMA does not completely mitigate, at the cost of serial-detection latency and higher complexity.

3) A unified NGMA framework for uplink transmission:

The unified uplink NGMA framework combines SDMA and NOMA through layered signal detection, generalizing both schemes while providing additional detection degrees of freedom. The framework also exposes challenges in user grouping, dynamic optimization, and extension to multi-user MIMO systems.

  • Unified uplink NGMA framework: The framework divides K information streams into L layers, decoding selected streams in parallel within each layer and subtracting decoded signals across layers.The parameter satisfies 1 ≤ L ≤ K, allowing intermediate detection structures between fully parallel and fully serial decoding.
  • Unified uplink NGMA framework: For L = 1, all streams are decoded in parallel by treating the other signals as interference, recovering SDMA.This case reduces the proposed rate expression to the SDMA expression in (10).
  • Unified uplink NGMA framework: For L = K, streams are decoded serially with previously decoded interference subtracted, recovering NOMA.This case reduces the proposed rate expression to the NOMA expression in (11).
  • Unified uplink NGMA framework: The proposed framework integrates SDMA and NOMA and provides more degrees of freedom for signal detection.Table VII summarizes its optimization variables and relationships with existing uplink schemes.
  • Challenges and future work: The framework is proposed for both downlink and uplink and is described as generalizing existing schemes while going beyond them.The paper identifies this unified framework as a foundation for further NGMA research.
  • Challenges and future work: User grouping introduces binary variables and makes joint optimization of grouping, beamforming, and resource allocation non-trivial.Matching theory and machine learning are identified as possible tools, especially for dynamic environments with frequent user arrivals and departures.
  • Challenges and future work: The current framework assumes single-antenna users with one intended data stream, leaving general multi-user MIMO extension as an important research direction.Multi-antenna users create more challenging joint transmit and receive beamformer design with higher computational complexity.

A. Modulation and Detection Design

Practical NGMA requires modulation-aware detection, robust mitigation of successive-interference-cancellation errors, and scalable channel estimation. The paper highlights overhead, error propagation, imperfect cancellation, and computational complexity as implementation challenges.

  • Modulation and detection design: Practical NOMA superposition may combine different modulation schemes, so modulation design must ensure theoretically achievable rates can be realized in implementation.Examples include BPSK and QAM signals.
  • Modulation and detection design: Unknown modulation types and orders require additional receiver functionality, motivating machine-learning-based blind modulation detection for superimposed signals.Knowing modulation information at the receiver can otherwise require higher-layer signaling overhead.
  • Error propagation mitigation: SIC error propagation can become severe in overloaded NGMA clusters because an early demodulation error creates residual interference for later users.The error probability of remaining users increases when lower-order symbols are decoded incorrectly.
  • Error propagation mitigation: Perfect SIC is generally unattainable, while more accurate mitigation methods can increase complexity and require better performance-versus-complexity tradeoffs.Existing models may inadequately characterize residual interference with practical modulation schemes.
  • Advanced channel estimation techniques: Accurate CSI affects user grouping, decoding order, signal reconstruction, and SIC performance in NOMA and NGMA.These dependencies make CSI acquisition more important than in OMA.
  • Advanced channel estimation techniques: Large numbers of users and antennas make NGMA channel estimation challenging because conventional near-optimal methods can impose unacceptable signaling overhead and computational complexity.The paper calls for NGMA-specific methods using conventional techniques and machine learning, particularly for mmWave, THz, selective fading, and RIS-assisted systems.
  • Conclusion: The paper surveys NOMA research and proposes a multi-antenna, NOMA-based unified NGMA framework for both downlink and uplink.It also identifies practical implementation challenges for future investigation.
Loading 2108.04561v2…