Developing NOMA to Next Generation Multiple Access (NGMA): Future Vision and Research Opportunities

I. INTRODUCTION

• 6G targets include ten-fold greater connectivity density than 5G, a 1 terabit-per-second peak rate, 100-fold higher energy efficiency, 0.1 millisecond latency, and 99.99999% reliability.Most 6G scenarios must optimize multiple objectives simultaneously, making efficient NGMA techniques important.Source: §I · p. 1
• Prior NOMA research focused on static devices and broadband-user data rates, leaving mobility and connectivity, reliability, and latency tradeoffs unaddressed.The introduction also links these gaps to overloaded traffic in heterogeneous ultra-dense networks, where NOMA can provide additional access.Source: §I · p. 1
• The article proposes “One Basic Principle plus Four New”: reconsider SIC, then explore massive connectivity, new physical-layer techniques, applications, and AI-enabled resource allocation.The four directions address QoS requirements including latency and reliability, physical-layer integration, heterogeneous scenarios, and adaptive resource allocation.Source: §I · p. 1

II. BASIC PRINCIPLE: RETHINKING SIC IN NOMA

• CSI-based SIC orders uplink decoding by channel conditions, whereas QoS-based SIC can decode a low-rate IoT sensor before a broadband user.The QoS-based example prioritizes the sensor because strong co-channel interference can make its first-stage achievable rate extremely low.Source: §II · pp. 1–2
• Both conventional CSI- and QoS-based SIC have an outage-probability error floor that cannot be removed merely by increasing transmit power.For CSI-based SIC, at high SNR the strong user’s SINR becomes constant, preventing the first-stage outage probability from reaching zero.Source: §II · p. 2
• Hybrid SIC jointly considers channel conditions and QoS, avoiding the conventional schemes’ outage floor; its decoding order is illustrated against conventional schemes in Fig. 2.It decodes the secondary user first when that user’s channel can guarantee the primary user’s QoS; otherwise, it decodes the primary user first, though further SIC optimization remains open.Source: §II · p. 2, Figure 2 · p. 2

III. NEW REQUIREMENTS: MASSIVE CONNECTIVITY FOR NOMA

• NGMA’s main 6G task is massive connectivity constrained by user-experience requirements for data rate, latency, and reliability.The passage states that NGMA can then maximize spectral efficiency, which also contributes to energy conservation.Source: §III · p. 2
• For massive connectivity, downlink NOMA should use QoS-oriented design, while semi-grant-free transmissions are preferable for uplink NOMA.The passage links these choices to satisfying performance metrics and QoS requirements.Source: §III · p. 2
• High spectral efficiency in 6G also contributes to energy conservation.The passage presents spectral efficiency as an outcome after connectivity is constrained by diverse QoS requirements.Source: §III · p. 2

A. QoS-based NOMA for Downlink Transmission

• Downlink NOMA must address SIC-related high latency and low reliability, especially when many users are served.The passage notes that multiple SIC iterations introduce long waiting times and high outage probabilities.Source: A. QoS-based NOMA for Downlink… · pp. 2–3
• QoS-based clustering generally favors users with different QoS, and placing IoT sensors before broadband users can reduce SIC-caused latency.The passage says there is no single correct answer to whether clusters should group users with the same or different QoS; IoT sensors prefer broadband users with stronger channels than themselves.Source: A. QoS-based NOMA for Downlink… · p. 3
• Downlink power allocation should use the higher of successful-decoding and latency-reliability-based data rates for IoT sensors, but only successful-decoding rates for broadband SIC.Allocated power need only satisfy each user’s QoS requirement to maximize connectivity.Source: A. QoS-based NOMA for Downlink… · p. 3

B. Semi-GF NOMA for Uplink Transmission

• Semi-GF NOMA serves intermittent uplink users by pairing grant-based and grant-free users in a cluster and broadcasting an acceptable received-power threshold.The base station grants an ORB to each grant-based user, evaluates remaining power for grant-free users, then broadcasts the threshold for their transmissions.Source: B. Semi-GF NOMA for Uplink Transmission · p. 3
• For single-ORB access, thresholds may specify minimum transmit power and maximum interference tolerance, with the latter depending on grant-based-user delay tolerance.Delay-sensitive grant-based users are decoded first, so their threshold contains only maximum interference tolerance; for multi-ORB access, user counts are unknown, making threshold design more complex.Source: B. Semi-GF NOMA for Uplink Transmission · p. 3
• In multi-ORB access, power pools and access-class-barring can reduce collisions and improve connectivity, but protocol design remains challenging.The passage reports the highest connectivity with ideal power pools and user barring, while uncontrolled transmit power produces more collisions than traditional grant-based transmission.Source: B. Semi-GF NOMA for Uplink Transmission · p. 3

IV. NEW TECHNIQUES: INTEGRATION OF NOMA WITH EMERGING PHYSICAL LAYER TECHNIQUES

• The section identifies integrating NOMA with OTFS, which targets high mobility, and THz communications, which offer multi-GHz bandwidth, as 6G research challenges.Source: §IV · p. 3

A. OTFS-NOMA

• OTFS was proposed to provide nearly non-fading channels in doubly dispersive conditions where delay and Doppler effects impair fast-mobile communications.Traditional OFDM subcarriers in the delay-Doppler plane are not orthogonal, introducing inter-symbol and inter-carrier interference.Source: A. OTFS-NOMA · p. 3
• OTFS-NOMA benefits fast- and slow-mobile users in latency and spectral efficiency by giving fast users additional ORBs and slow users access to spectrum otherwise allocated to fast users under OTFS-OMA.The slow-mobile user decodes the fast user's signal in the delay-Doppler plane, cancels it, then decodes its own signal in the time-frequency plane.Source: A. OTFS-NOMA · p. 4

B. THz-NOMA

• NOMA is promising for improving spectral efficiency in THz networks when limited-scattering channels are highly correlated, unlike the nearly orthogonal channels expected when independent users are fewer than antennas.Source: B. THz-NOMA · p. 4
• THz-NOMA groups highly correlated users into clusters and assigns them different transmit-power levels for decoding with hybrid SIC.Source: B. THz-NOMA · p. 4
• Location-based clustering can group users in the same geographic area using full CSI only for a cluster-head reference user, while a K-means-based solution maximizes energy efficiency.The cited machine-learning-aided solution also uses hybrid precoding with a sub-connection structure and distributed alternating direction method of multipliers for power allocation.Source: B. THz-NOMA · p. 4

C. Integration of NOMA with Other Techniques

• For integrated sensing and communication, SIC can cancel the decoded communication signal so conventional sensing algorithms can analyze the radar return.Source: C. Integration of NOMA with Other… · p. 4
• For visible-light communication, NOMA can provide high spectral efficiency to address LEDs' small modulation bandwidth, particularly in multi-user cases.The passage also notes VLC's high SNR.Source: C. Integration of NOMA with Other… · p. 4
• For index modulation, NOMA can handle strong inter-user interference by grouping users with similar spatial features, increasing spectral and energy efficiency.Source: C. Integration of NOMA with Other… · p. 4

A. NOMA in Integrated Terrestrial and Aerial Networks

• SAGINs combine terrestrial networks with aerial platforms and satellites to fill single-tier coverage holes and pursue ubiquitous connectivity.Their optimization must address massive connectivity alongside delay, throughput, BER, and power objectives; hybrid-SIC NOMA is described as applicable to heterogeneous QoS in multi-tier SAGINs.Source: A. NOMA in Integrated Terrestrial and… · p. 5

B. NOMA in Reconfigurable-Intelligent-Surface-Enhanced Networks

• In RIS-enabled networks, NOMA can offer higher spectral efficiency and massive connectivity than OMA, while RIS phase shifts and placement can adjust individual users' channel quality.Source: B. NOMA in… · p. 5
• QoS-based SIC can relax the perfect-cascaded-CSI requirement of traditional CSI-based SIC in RIS networks, while hybrid SIC addresses the tradeoff between channel-estimation complexity and spectral efficiency.Source: B. NOMA in… · p. 5

C. NOMA in Robotic Communications/Autonomous Networks

• NOMA can support autonomous robots’ delay-sensitive, collision-tolerant communications, while dynamic clustering with mobile users can enhance both groups’ received SINR.The proposed hybrid-SIC-enabled OTSF-NOMA scheme clusters robots and other mobile users with heterogeneous mobility states.Source: C. NOMA in Robotic… · p. 5

D. NOMA in VR/AR Multi-Layer Video Transmission

• Hybrid-SIC-enabled NOMA is proposed for delay-sensitive VR/AR multilayer video, where fixed-size OMA resource blocks can make users wait for access.The passage says hybrid SIC can cancel the primary users’ outage-probability error floor to provide reliable transmission.Source: D. NOMA in VR/AR Multi-Layer Video… · p. 5

E. NOMA in E-Health Systems

• For e-health, QoS-based SIC detects the delay-sensitive inquired user before decoding the delay-tolerant monitored user, efficiently using communication resources.Hybrid-SIC-enabled NOMA is also described as helpful for massive connectivity under strict QoS and for providing additional access channels for offloading computation to edge devices or servers.Source: E. NOMA in E-Health Systems · pp. 5–6

• Heterogeneous, dynamic 6G scenarios call for new approaches because conventional methods have limited controllable variables and struggle with Markovian problems.Hybrid SIC adds controllable variables through joint CSI- and QoS-based clustering and power allocation.Source: §VI · p. 6
• Machine learning is presented as a competitive candidate for NOMA because AI-based algorithms can address many control variables and temporally correlated events.The paper contrasts long-term network evolution objectives of ML algorithms with conventional policies focused on immediate benefits.Source: §VI · p. 6
• ML can enable interaction between base stations and dynamic environments, with reinforcement learning adapting policies using real-time feedback and historical experience.The section introduces ML’s role in NOMA-based networks.Source: §VI · p. 6

A. Reinforcement Learning for NOMA-based Networks

• Reinforcement learning lets base stations and access points learn from real-time feedback and historical experience to adapt control policies in dynamic environments.The section identifies dynamic user clustering and pairing as an RL application, with re-clustering improving performance in dynamic scenarios.Source: A. Reinforcement Learning for… · pp. 6–7
• RL resource allocation targets long-term benefits across power, caching, computing, and spectrum, but its optimality is not guaranteed and its advantage shrinks in simple static scenarios.RL is suited to Markovian problems, so NOMA problems must be reformulated as Markovian.Source: A. Reinforcement Learning for… · p. 7

B. Deep Learning for NOMA-based Networks

• Deep learning can acquire CSI from training on existing channel models and estimate dynamically fluctuating channels, addressing a challenge in NOMA with MIMO or RIS.Source: B. Deep Learning for NOMA-based Networks · p. 7
• Deep learning is considered for dynamic NOMA resource allocation, signal design, detection, and decoding, but reducing architecture complexity remains open.Resource allocation examples include power allocation, subchannel assignment, and subcarrier assignment.Source: B. Deep Learning for NOMA-based Networks · p. 7

C. Other Machine Learning for NOMA-based Networks

• K-means and expectation maximization can cluster NOMA users using channel-response correlations or learned user structures in fixed and dynamic scenarios.Source: C. Other Machine Learning for… · p. 7
• Federated learning keeps raw training data local while sharing model parameters, and NOMA can reduce aggregation latency by providing same-channel multiple access.Transfer learning can update full or partial neural-network parameters for faster convergence in similar new environments with limited data or training time.Source: C. Other Machine Learning for… · p. 7

VII. CONCLUSION AND OUTLOOK

• The conclusion presents hybrid SIC’s joint CSI- and QoS-based decoding order as removing the conventional outage-probability error floor for joining users.It surveys massive connectivity, physical-layer integration, heterogeneous scenarios, and AI enhancement, and identifies benefits especially for networks with distinct QoS requirements.Source: §VII · p. 7