PaperlayerView exact revision on arXiv

Source-linked AI summary

A Survey of NOMA: Current Status and Open Research Challenges

Behrooz Makki, Krishna Chitti, Ali Behravan, Mohamed-Slim Alouini

arXiv:1912.10561v1cs.ITeess.SP

TL;DR

The paper examines NOMA’s performance and practicality for dense networks, including why 3GPP did not continue it as a 5G NR work-item. It reviews NOMA against OMA-based approaches and proposes lower-complexity transmission techniques, finding limited comparative gains but opportunities to improve implementation practicality.

  • Problem

    NOMA can outperform OMA with suitable parameters, but receiver, UE pairing, coordination, and other implementation complexities challenge its practicality in dense networks.

  • Method

    The paper reviews 3GPP Release 15 discussions, compares NOMA with MU-MIMO through simulations, and develops lower-complexity UE pairing, receiver, and NOMA-HARQ techniques.

  • Results

    NOMA’s relative performance gain over MU-MIMO and other Rel-15 mechanisms was not consistently worthwhile, while proposed techniques reduce NOMA implementation complexity.

  • Takeaways & Limitations

    NOMA remains a possible beyond-5G technology for dense or ultra-dense networks if its spectral efficiency and implementation practicality are improved.

Abstract

from arXiv · show

Non-orthogonal multiple access (NOMA) has been considered as a study-item in 3GPP for 5G new radio (NR). However, it was decided not to continue with it as a work-item, and to leave it for possible use in beyond 5G. In this paper, we first review the discussions that ended in such decision. Particularly, we present simulation comparisons between the NOMA and multi-user multiple-input-multiple-output (MU-MIMO), where the possible gain of NOMA, compared to MU-MIMO, is negligible. Then, we propose a number of methods to reduce the implementation complexity and delay of both uplink (UL) and downlink (DL) NOMA-based transmission, as different ways to improve its efficiency. Here, particular attention is paid to reducing the receiver complexity, the cost of hybrid automatic repeat request as well as the user pairing complexity. As demonstrated, different smart techniques can be applied to improve the energy efficiency and the end-to-end transmission delay of NOMA-based systems.

I. INTRODUCTION

NOMA co-schedules multiple UEs on shared time, frequency, or code resources and was studied as a candidate for LTE, 5G, and beyond-5G systems. This paper reviews 3GPP’s conclusions, compares WSMA-based NOMA with MU-MIMO, and develops techniques targeting NOMA’s implementation complexity.

  • Motivation: NOMA allows multiple UEs to share the same time, frequency, and/or code resources through non-orthogonal transmission.It has been considered for LTE, 5G, and beyond-5G systems, including several 3GPP applications.
  • Motivation: NOMA schemes differ mainly in UE signature design, using spreading, coding, scrambling, or interleaving distinctness under the superposition principle.Examples include power-domain NOMA, SCMA, PDMA, RSMA, MUSA, IGMA, WSMA, and IDMA.
  • Motivation: With suitable parameter settings, NOMA can outperform OMA but incurs receiver, UE-pairing, and coordination complexity.This trade-off motivates NOMA’s proposed use in dense networks with insufficient orthogonal resources.
  • Contributions: The paper summarizes the 3GPP Release 15 study-item discussions and the decision not to continue NOMA as a work-item.The review is intended to provide guidance for improving NOMA’s practicality.
  • Contributions: Link-level evaluations compare WSMA-based NOMA with MU-MIMO under ideal and non-ideal channel estimation using BLER.The reported relative NOMA gain was not large enough to justify its implementation complexity.
  • Contributions: The paper proposes low-complexity approaches for UE pairing, receiver design, and NOMA-HARQ.These techniques target reduced implementation complexity while supporting improved spectral efficiency and practical adoption.

II. PERFORMANCE ANALYSIS

The performance analysis introduces WSMA as a spreading-based NOMA scheme and examines how signature-sequence correlation and overloading shape its operation. It uses several performance indicators to generate signature sets, with examples evaluated at 100% overloading.

  • A. WSMA-based NOMA: WSMA distinguishes users with non-orthogonal, short, low-cross-correlation, non-sparse spreading sequences.The sequences are based on the Welch bound and are assigned as user-specific signatures.
  • A. WSMA-based NOMA: Each UE is assigned a normalized signature sequence of dimension L, while the signature matrix contains K user vectors.The overloading factor is K/L, and NOMA targets K/L > 1 to support higher user density.
  • A. WSMA-based NOMA: The signature matrix is generated by optimizing a chosen performance indicator, such as total squared correlation, worst-case coherence, or minimum chordal distance.These indicators impose different correlation properties, including equal correlation or selected zero correlations among vectors.
  • A. WSMA-based NOMA: At the Welch-bound equality, several system metrics, including sum-capacity and sum-MSE, are optimized simultaneously.The resulting signature collection is called a Welch bound equality set, although the bound applies to the ensemble rather than each sequence individually.
  • A. WSMA-based NOMA: The illustrative Gramian matrices use K = L, representing 100% overloading, although WSMA is mainly designed for K > L.This setting is used for simplicity when showing correlation properties for different performance indicators.

B. NOMA vs MU-MIMO

The comparison evaluates WSMA-based NOMA against MU-MIMO under ideal and non-ideal channel estimation. NOMA can reduce interference and error floors, but its performance advantage is generally small and trades off against spectral efficiency and implementation complexity.

  • Comparison setup: WSMA repeats each QAM symbol across L resources using a UE-specific spreading sequence, while MU-MIMO separates users primarily through spatial degrees of freedom.The NOMA composite signal includes each user’s QAM symbol, transmit power, fading channel, and spreading sequence.
  • Comparison setup: With WSMA, each UE uses L times more resources for the same QAM-symbol count, creating a trade-off between overloading and spectral efficiency.Increasing K for fixed L can raise sum-rate but may require optimizing conflicting metrics.
  • Ideal channel estimation: Under ideal channel estimation, WSMA outperforms MU-MIMO in BLER for the assumed setup and various MU-MIMO group counts, while heavily loaded MU-MIMO saturates.The simulations use QPSK, 20-byte transport blocks, four BS receive antennas, MMSE detection, and K = 6 or K = 12 users.
  • Ideal channel estimation: Even with ideal channel estimation, NOMA’s relative performance gain over MU-MIMO is negligible and decreases as the number of UEs increases.At higher MU-MIMO group counts, fewer users share each resource element, reducing multiuser interference and narrowing the difference.
  • Non-ideal channel estimation: With non-ideal channel estimation, WSMA and MU-MIMO have similar BLER over wide SNR ranges, with the better scheme depending on K, grouping, SNR, and target BLER.For K = 12, WSMA outperforms MU-MIMO with G = 6 only beyond target BLER 10^-3; for K = 6 and G = 3, the threshold is 10^-2.

C. NOMA for beyond 5G

The 3GPP study compared many NOMA schemes and receivers across extensive link-level cases. Its results found no consistent worthwhile gain over MU-MIMO or Rel-15 mechanisms, motivating NOMA’s deferral to beyond 5G.

  • 3GPP study conclusions: The 3GPP NOMA study included 14 companies, more than 35 link-level cases, and BLER evaluations using generally aligned parameters.The broad evaluation enabled comparisons among different transmission schemes and receivers.
  • 3GPP study conclusions: In ideal conditions, NOMA could be better or worse than MU-MIMO depending on the number of UEs and simulation parameters.The outcome was not uniformly favorable across tested configurations.
  • 3GPP study conclusions: With realistic multipath channel estimation, NOMA’s relative gain decreased, and MU-MIMO could outperform it depending on the parameter settings and channel model.No clear NOMA gain over Rel-15 mechanisms appeared across all studied scenarios.
  • 3GPP study conclusions: The lack of worthwhile and conclusive gains led 3GPP not to continue NOMA as a work-item, while leaving it for beyond 5G use cases involving ultra-dense UEs.The paper presents this decision alongside its own comparisons in Figs. 2–4.

III. REDUCING THE IMPLEMENTATION COMPLEXITY

NOMA implementation complexity grows across UE pairing, signal decoding, and CSI acquisition, especially in dense networks. The paper proposes low-complexity approaches for uplink NOMA that also extend to downlink and other NOMA configurations.

  • NOMA complexity arises in UE pairing, signal decoding, and CSI acquisition, and increases rapidly with the number of UEs in dense networks.
  • The schemes are presented for uplink NOMA but can be extended to downlink transmission and arbitrary numbers of paired UEs.
  • The discussion uses power-domain NOMA with successive interference cancellation receivers, while the approaches also apply to other NOMA transmissions and receivers.

A. HARQ using NOMA

NOMA-HARQ faces poor diversity, error propagation, and potentially multiple retransmissions that increase end-to-end delay. The described process buffers failed signals, delays one retransmission, and uses decoding and interference cancellation to recover both signals.

  • CSI acquisition and UE pairing overhead make NOMA most suitable for fairly static channels without frequency hopping.
  • When both signals fail, the base station buffers them, delays one UE’s retransmission, and requests retransmission from the other UE.
  • Error propagation can make one decoding failure affect subsequent sequential decodings, increasing the probability of multiple HARQ retransmissions and high end-to-end delay.
  • The base station combines retransmitted copies, decodes a signal, and can then use successive interference cancellation to decode the other signal after removing interference.

1) Smart NOMA-HARQ [36]:

Smart NOMA-HARQ uses coordinated decoding and retransmission timing to reduce retransmissions and improve end-to-end throughput while increasing fairness between users.

  • Smart NOMA-HARQ can reduce the number of retransmissions and improve end-to-end throughput.
  • The required retransmissions of the cell-edge UE decrease remarkably, increasing fairness between the UEs.
  • The setup requires the base station to decode all buffered signals in each round and inform UEs of appropriate retransmission times.

2) Dynamic UE Pairing in NOMA-HARQ [37]:

Dynamic UE pairing adapts NOMA transmissions across retransmission rounds when decoding fails. Failed users can reuse another UE’s spectrum resource or pair with a stronger new UE to add virtual diversity.

  • 2) Dynamic UE Pairing in NOMA-HARQ [37]: Dynamic pairing adds virtual diversity by considering different UE pairs in different retransmission rounds according to message decoding status.
  • 2) Dynamic UE Pairing in NOMA-HARQ [37]: A failed UE may reuse the other UE’s spectrum resource during retransmissions.
  • 2) Dynamic UE Pairing in NOMA-HARQ [37]: After UE1 fails, it can be paired with a new UE0 having a better channel to provide a small retransmission boost.

3) Multiple Access Adaptation in Retransmissions [38]:

The proposed adaptive retransmission scheme begins with dedicated OMA resources and permits failed UEs to reuse another UE’s bandwidth, combining NOMA with HARQ. The paper also develops rate-based and coordinated multi-point pairing methods to reduce CSI and pairing overhead.

  • Multiple Access Adaptation in Retransmissions [38]: Failed UEs can reuse another UE’s bandwidth in later retransmission rounds, while successfully decoded UEs continue new transmissions on their own resources.The example has UE2 retransmit over both w1 and w2 after failure, while UE1 sends a new message over w1.
  • Multiple Access Adaptation in Retransmissions [38]: HARQ combines SIC-based decoding with MRC across repeated signal copies, allowing the receiver to decode a retransmitted message using interference-free and interference-affected observations.UE1 is first decoded and removed from one resource, after which UE2’s three copies are combined.
  • Multiple Access Adaptation in Retransmissions [38]: The adaptive scheme exploits network and frequency diversity, increases achievable UE rates, improves fairness, and significantly enhances service availability or network reliability versus conventional OMA.The approach is presented as useful for buffer-limited systems and as balancing receiver complexity against network reliability.
  • Multiple Access Adaptation in Retransmissions [38]: The proposed RTD-HARQ methods are also applicable to other HARQ protocols.
  • Rate-based UE Pairing [40]: Rate-based pairing limits CSI acquisition by first screening UE pairs through rate demands and estimated successful-pairing probabilities, then collecting pilots only for promising pairs.The BS assigns resources when a pair’s estimated success probability exceeds a threshold, estimates channel quality, and selects powers satisfying rate demands.
  • Rate-based UE Pairing [40]: In COMP-NOMA, high-rate backhaul enables SIC and pairing algorithms to be centralized at one BS, while another BS can pair its cell-center UEs without advanced pairing.This applies when interference toward the coordinating BS remains acceptable.

2) UE Pairing in COMP-NOMA:

The COMP-NOMA pairing scheme uses high-rate backhaul to confine SIC-based reception and pairing computation to one base station while supporting pairing across other base stations’ cell-center UEs.

  • UE Pairing in COMP-NOMA: SIC-based reception is needed only at BS1, and the UE pairing algorithm can run only at one base station.NOMA transmission continues when at least one base station finds a suitable pair for UE2.
  • UE Pairing in COMP-NOMA: BS2 can pair each of its cell-center UEs with the existing pair when interference toward BS1 is not high, avoiding advanced pairing algorithms.

C. Receiver Adaptation

The receiver-adaptation scheme avoids unnecessary sequential decoding by using the estimated probability of successful decoding to choose when to continue SIC. It reduces implementation complexity and improves end-to-end throughput, while the paper concludes that NOMA’s BLER gain over MU-MIMO is insufficient in current use cases to justify its complexity.

  • Receiver Adaptation: Sequential decoding can increase end-to-end delay, receiver complexity, and energy consumption compared with OMA, motivating conditional receiver adaptation.
  • Receiver Adaptation: If UE1 is decoded successfully, the BS removes its signal and decodes UE2 interference-free; otherwise, it immediately sends NACKs for both UEs.The failed UE2 signal is buffered for later HARQ processing.
  • Receiver Adaptation: The proposed receiver reduces implementation complexity and improves end-to-end throughput by adapting decoding to the estimated probability of successful decoding.For two paired UEs, the relative performance gain is reported to increase with the number of paired UEs.
  • Conclusions: For current use cases, NOMA’s BLER gain over MU-MIMO was not large enough to convince 3GPP to continue it as a work-item.
Loading 1912.10561v1…