1 Introduction
-
superficial role of NOMA to address the IoT use-cases and networks
-
the required novel resource management and allocation aspects for NOMA (largely relying on the AI/ML techniques) to assure futher improvements in spectral efficiency and lowering complexity of SIC receivers
-
the ultra-dense network triggered modifications of the legacy cellular RAN paradigm towards the direction of cell-free, and cell-free mMIMO techniques
-
the role of NOMA in combination with the MEC as it is becoming especially important due to (i) ubiquitous access techniques in hybrid terrestrial and non-terrestrial networking, (ii) its importance regarding fulfillment of 6G sustainability goals where energy-efficiency becomes essential driver of design and deployment decisions
-
perspective of smooth alignment with existing OFDMA based 4G and 5G networks.
2 NOMA and IoT in Standards
3 Overview of NOMA
-
NOMA in 5G and beyond—this sub-section deals with current research in the framework of 5G or targeting beyond 5G aspects, e.g. cell-free concept. Concerning 5G, the sub-section focuses on NOMA/Orthogonal multiple access (OMA) coexistence, NOMA in combination with legacy 5G architectures like SD-RAN, as well as gives some indications of similarities and differences in NOMA when combined with the various 5G use-cases.
-
NOMA in 6G—this sub-section provides an overview of mainly the 6G research directions and main components of 6G and especially the Artificial Intelligence / Machine learning (AI/ML)/Deep Learning (DL) aspects in combination with the NOMA techniques
4 NOMA in 5G and Beyond
4.1 NOMA 5G Basics
4.1.1 Coexistence of NOMA and Other Access Techniques
4.1.2 NOMA in Software Defined RAN Networks
4.1.3 NOMA Resource Management and 5G Use-Cases
4.1.4 Beyond 5G NOMA - Cell Free NOMA
4.2 NOMA in 6G
4.2.1 Evolution Towards 6G
-
Zero-energy air interfaces are emerging.
-
Cell-free networks that imply no explicit connection established between IoT devices of the future (so called “Massive Type Device”) and a base station (BS) of a network cell, thus reducing signaling overhead while improving the reliability by enabling multiple BSs to demodulate the incoming transmissions, either separately or jointly.
-
Increase of network heterogeneity—more micro-operators, more new RATs, smaller and denser cells.
-
Infrastructure-less, dynamically formed networks for mission critical usage.
-
The need to support growing number of traffic patterns and service classes.
4.2.2 AI/ML Aspects of NOMA
-
User clustering and power allocation for NOMA in mm-Wave.
-
CSI prediction, support for spectrum sensing, awareness of user localization and throughput prediction.
-
NOMA-based energy-efficient task scheduling for MEC and the total energy consumption with MEC.
-
NOMA for ultra-dense deployment and grant-free access.
-
Computation offloading and subcarrier allocation problem in multi-cell, Multi-carrier (MC) NOMA.
-
Challenges of complex sensing model in cooperative spectrum sensing for non-orthogonal multiple access.
Reference | Type of learning | Deep learning model | Application |
---|---|---|---|
11 | Supervised | Recurrent neural network | Rapid and optimized resource allocation |
14, 13 | Supervised | LSTM | Channel estimation |
15 | Reinforcement | Deep Q-network | subcarrier assignment and power allocation |
16 | Reinforcement | Attention-based neural network | Joint power allocation and channel assignment |
17 | Supervised and | Deep belief network | Power optimization assignment |
18 | Supervised | Deep neural network | Automatic realization of the CSI |
21 | Unsupervised | Deep neural network | Signal constellation design |
25 | Supervised | Deep neural network | Reliability improvement of grant-free access |
26 | Supervised learning | Deep neural network | Optimization for energy-efficient scheduling |
27 | Reinforcement | Deep neural network [73] | Power allocation optimization |
based on Q-learning [71] |
4.2.3 Deep-Learning for NOMA Resource Allocation
-
User clustering (user pairing, cell-free, cluster identification, power control, grant-free mode)
-
Task scheduling (joint with computational resource and power allocation; task offloading, cooperative offloading)
-
Resource management (maximize energy efficiency under QoS/interference and power limitation, subchannel and power allocation, outage capacity maximization, caching)
-
Heterogeneous requirements (diverse service requirements, beamforming, multiple-access selection, power control)
-
NOMA improvements (decoding order, multi-user detection, CSI prediction, multi-access signatures optimization).
-
inputs: channel (coefficients, for pairs of users in NOMA), SINR, demand from users
-
outputs: power control, channel access time, resource allocation requirement
4.2.4 Overview of Recent R &D Projects Dealing with NOMA
5 Sustainablity Aspects of NOMA
5.1 Energy Efficiency in NOMA
Paper | Energy efficiency | Algorithm/Aproach | Complexity result |
---|---|---|---|
[5] | EE(NOMA)>EE(OMA) EE(proposed)>EE(NOMA) | User association Subcarier assignment | \(\theta (NM(M+T_{t})) \theta (MK(MK+NT_{t}^{'}))\) |
[32] | Proposed solution needs only 20–30% less energy consumed compared to random clustering and power control | User clastering and computation resource allocaction transmit power control | Low complexity |
[80] | Only the power level are considered \({P_{h},P_{l}}\) for power control. No other mentions. | Grant-free access methods based on Q-learning is proposed | N/A |
[11] | Basic OMA is always less energy efficient than NOMA.But optimized-OMA can sometimes be more efficient than NOMA. | OMA scheme with optimal time and power allocation (O-OMA) H-NOMA scheme with optimal user selection. | N/A |
Low complexity solution is robust against the increase in number of IoT devices (NOMA clients). | Computation Resource Allocation and SIC Ordering Algorithm. | Optimal solution Low-complexity solution | |
[14] | In DL: Algorithms applied to OMA and NOMA bring similar average cluster energy In UL: OMA case consumes more energy for the same algorithms as in NOMA average transmit energy per user in the UL under OMA scheme is higher than the NOMA schemes, due to orthogonal transmission. | SWIPT enabled NOMA cell with join sub carrier assignment, time-switching and power allocation (J-SA-TS-PA) | J-TS-PA-polynomial complexity \(\theta (2|\ell _{m}|^{2}T_{m,k}(\in ))\) Subcarier allocation- polynomial complexity \(\theta (KM^{2})\) Joint optimization has higher complexity |
[103] | Focus on latency and task completion probability | Closed-form mathematical expressions of the successful computation probability in MEC offloading NOMA system | N/A |
[70] | Energy efficiency of task offloading to MEC with NOMA is always better than reference FDMA system with bandwidth equally divided between K-users. Also as number of users growths the NOMA is always more efficient. | With the obtained optimal power control solutions, the task offloading partitions and time allocation are obtained by the successive convex approximation algorithm | N/A |
[44] | Exhaustive search max-min computational efficiency is lower than for NOMA schemes proposed. Computing efficiency with proposed NOMA scheme is better than OMA and local computing on the UE. | Heuristic search algorithm to obtain optimal policy, including uplink transmit power allocation policy and local computing resource allocation policy | \(\Theta [(\frac{K^{2}+13K}{2})(HMS+NI)]\) |
[96] | There is no detailed energy efficiency analysis provided. | Design efficient iterative algorithm by jointly designing the secrecy rate, local computing bits, and power allocation | N/A |
[51] | Energy consumption of NOMA offloading decreases by increasing the available number of frequency RBs; second, the energy consumption improves by increasing number of users sharing PRBs | Minimizes the energy consumption of MEC users via optimizing the user clustering, computing and communication resource allocation, and transmit power | N/A |
[20] | Proposed NOMA scheme always achieves lower energy consumption than FDMA (OMA) scheme. | Non-convex problem of energy consumption minimization is transformed into a task and power allocation problem, and a subchannel allocation problem, plus the delay constraints are considered. | Proposed algorithms complexity is \(O(n^{3})\) for both algorithms with guaranteed convergence. Where n depicts number of UEs in the center of the cell |
6 Conclusions
-
NOMA performance and complexity: essentially, NOMA introduces a tradeoff between performance (sum rate) and complexity. Analyzing this tradeoff is a worthwhile research topic. [...] energy efficiency of cell free massive MIMO-NOMA could be quantified [76]. The effect of SIC processing delays, SIC decoding error propagation, unpredictable interference caused by grant-free access, and imperfect CSIs should be analyzed in further details [50]
-
Realistic receiver design: realistic receiver needs to take into account practical issues such as active user detection for grant-free transmission, non-ideal channel estimation, time and frequency offset handling and receiver complexity The key issue for grant-free is that how to perform UE identifications [85]. Although spatial domain NOMA is quite effective in improving the spectral efficiency, the use of conventional pilots to acquire channel information causes severe pilot contamination [61]. Possible solutions include blind (pilot-free) data-driven methods [85], channel predictions using non-RF data [8] and the enhancement of pilot design. The use of multi-pilots along with the application of strategies similar to modern random access to decode them is an example of the latter [32].
-
Grant-free schemes: while advanced NOMA has been well researched in configured grant-free schemes, in other grant-free approaches,the global power control, resource allocation and configuration cannot be accomplished efficiently, calling for advancements towards uncoordinated access policies. Open issues for grant-free access are: UL transmission detection, HARQ related procedures, RRC and L1 signaling, link adaptation, switching between orthogonal and NOMA [88]. This poses the further challenge of multi-user interference (MUI), for which the one-dimensional randomness of the power domain yielded thanks to the near-far effect may not be enough. Instead, higher-dimensional randomness including also e.g. code- and spatial-domains should be introduced [61]
-
User clustering and prediction more complicated clustering algorithms that are robust to noises and outliers should be studied, incorporating the optimization of the number of user clusters into mmwave NOMA systems is capable of further improving the sum rate of the mmWave NOMA systems. Future work can also consider more sophisticated on-line and reinforcement learning procedures that update the partition according to the dynamic mmWave NOMA scenarios [25]. The analysis of customer behavior is one of the hardest challenges during the prediction of customer requests [36]
-
Role of AI/ML in NOMA: the efficient design of the core neural networks is still a headache. Most of the current deep neural networks employed in NOMA provides a high computational cost. Therefore, reducing the computational complexity is a significant research issue for the future. Moreover, using a specific neural network, different researchers used different architecture. The number of layers and the quality and quantity of the training dataset is different for a different architecture, however, the optimized system is yet to be discovered [41]
-
MEC offloading and orchestration in TN/NTN huge challenges on MEC systems have been brought by the highly unpredictable task flows arriving at MEC servers, requiring advanced optimization and estimation technologies to balance a trade-off between system performances and energy consumption [30] Recently NOMA and SIC has been indicated by the European Space Agency as one among multiple items to consider for research in the 5G/6G strategic appraoch. New research is required to address the challenges of orchestrating resources in a comprehensive ecosystem where IoT, edge/fog and cloud converge to form a computing continuum [7]. Finally, it is interesting to consider a multi-objective optimization problem in DC-assisted MEC offloading, i.e., minimization of power consumption with an interruption probability constraint and minimization of interruption probability with a power consumption constraint [55]