Introduction
The design multiple access schemes is of interest in the cellular systems design. Here, the goal is to provide multiple user equipments (UEs) with radio resources in a spectrum-, cost- and complexity-efficient manner. In 1G–3G, frequency division multiple access (FDMA), TDMA (T: time) and CDMA (C: code) schemes have been introduced, respectively. Then, Long-Term Evolution (LTE) and LTE-Advanced developed orthogonal frequency division multiple access (OFDMA) and single-carrier (SC)-FDMA as orthogonal multiple access (OMA) schemes. Also, 5G new radio (NR) utilizes OFDMA waveform in both uplink (UL) and downlink (DL) transmission. Such orthogonal designs have the benefit that there is no mutual interference among UEs, leading to high system performance with simple receivers.
In the last few years, non-orthogonal multiple access (NOMA) has received considerable attention as a candidate multiple access technique for LTE, 5G and beyond 5G systems. With NOMA, multiple UEs are co-scheduled and share the same radio resources in time, frequency and/or code. Particularly, 3GPP has considered NOMA in different applications. For instance, NOMA has been introduced as an extension of the network-assisted interference cancellation and suppression (NAICS) for inter-cell interference (ICI) mitigation in LTE Release 12 [1] as well as a study-item of LTE Release 13, under the name of DL multi-user superposition transmission (DMST) [2].
Different schemes have been proposed for NOMA including, power domain NOMA [3], SCMA (SC: sparse code) [4], [5], PDMA (PD: pattern division) [6], RSMA (RS: resource spread) [7], multi-user shared access (MUSA) [8], IGMA (IG: interleave-grid) [9], Welch-bound equality spread multiple access (WSMA) [10], [11], IDMA (ID: interleave-division) [12], NCMA (NC: non-orthogonal coded) [13], ACMA (AC: asynchronous coded) [14], low code rate spreading (LCRS) [15], non-orthogonal coded access (NOCA) [16], low code rate and signature based shared access (LSSA) [17] as well as UGMA (UG: user grouped) [18]. These techniques follow the superposition principle and, along with differences in bit- and symbol-level NOMA implementation, the main difference among them is the UEs’ signature design which is based on spreading, coding, scrambling, or interleaving distinctness.
Various fundamental results have been presented to determine the ultimate performance of NOMA in both DL [3], [19], [20], [21], [22] and UL [21], [22], [23], [24], to incorporate the typical data transmission methods such as hybrid automatic repeat request (HARQ) to the cases using NOMA [25], [26], [27], to develop low-complexity UE pairing schemes [28], [29], [30], and to reduce the receiver complexity [4], [31], [32]. As shown in these works, with proper parameter settings, NOMA has the potential to outperform the existing OMA techniques at the cost of receiver, UE pairing and coordination complexity. For these reasons, NOMA has been suggested as a possibility for data transmission in dense networks with a large number of UEs requesting for access such that there are not enough orthogonal resources to serve them in an OMA-based fashion. Particularly, in 2018, 3GPP considered a study-item to evaluate the benefits of NOMA and provide guidelines on whether NR should support (at least) UL NOMA, in addition to the OMA [33], [34]. However, due to the reasons that we explain in the following, it was decided not to continue with NOMA as a work-item, and to leave it for possible use in beyond 5G.
In this paper, we study the performance of NOMA in UL systems (in the meantime, most of the proposed schemes of Section III are applicable/easy-to-extend for DL transmission). The contributions of the paper are threefold:
We summarize the final conclusions presented in 3GPP Release 15 study-item on NOMA. Particularly, we present the discussions leading to the conclusion of not continuing with NOMA as a work-item. Such conclusions provide guidelines for the researchers on how to improve the practicality of NOMA.
We present link-level evaluation results to compare the performance of WSMA-based NOMA and multi-user multiple-input-multiple-output (MU-MIMO) in different conditions. Here, the results are presented for the cases with both ideal and non-ideal channel estimation. As we show, the relative performance gain of WSMA-based NOMA compared to MU-MIMO, in terms of block error rate (BLER), is not that large to motivate its implementation complexity.
We demonstrate different techniques to reduce the implementation complexity of NOMA-based systems. Here, we concentrate on developing low-complexity schemes for UE pairing, receiver design and NOMA-HARQ, where simple methods can be applied to reduce the implementation complexity of NOMA remarkably. These results are interesting for academia because each of the proposed schemes can be extended and studied analytically in a separate technical paper.
There are a number of survey papers on NOMA [35], [36], [37], [38], [39] in which the performance of power-domain NOMA [35], [36], [37], [38], cognitive radio inspired NOMA [36], code-domain NOMA [38] and signature-based NOMA [39] has been reviewed, and different aspects of MIMO transmission [35], [36], [38], user pairing [37] and receiver design [39] in NOMA have been studied. As opposed, in this paper, we mainly concentrate on the 3GPP discussions on NOMA, comparison between MU-MIMO and WSMA-based NOMA as well as introducing methods to reduce the implementation complexity of NOMA.
As we demonstrate, different techniques can be applied to reduce the implementation complexity of NOMA. Moreover, there is a need to improve the spectral efficiency and the practicality of implementation, in order to have NOMA adopted by the industry.
Performance Analysis
In this section, we first present the principles of WSMA as an attractive spreading-based NOMA technique. Then, we compare the performance of WSMA NOMA with MU-MIMO and summarize the final conclusions presented in 3GPP Release 15 study-item on NOMA.
A. WSMA-Based NOMA
WSMA is a spreading-based NOMA scheme [40]. Here, the key feature is to use non-orthogonal short spreading sequences with relatively low cross-correlation for distinguishing multiple users, and the spreading sequences are non-sparse. The WSMA spreading sequences are based on the Welch bound [10], [11], the details of which are explained in the following.
Let us consider \begin{equation*} \text {TSC} = \sum \limits ^{K}_{i=1} \sum \limits ^{K}_{j=1} |\mathbf {s}_{i}^{\text {H}} \mathbf {s}_{j}|^{2},\tag{1}\end{equation*}
\begin{equation*} \text {TSC}\ge \frac {K^{2}}{L}.\tag{2}\end{equation*}
Other PIs that may be considered for the SS generation include the worst-case matrix coherence given as
At times, it may be required to have zero correlation between few vectors of the constituent SSs in the set. In that case, optimizing \begin{align*} |\mathbf {S}^{H}\mathbf {S}|_{\text {TSC}}=&\begin{bmatrix} 1 & \rho _{12} & \rho _{13} & \rho _{14} \\ \rho _{12} & 1 & \rho _{23} & \rho _{24} \\ \rho _{13} & \rho _{23} & 1 & \rho _{34} \\ \rho _{14} & \rho _{24} & \rho _{34} & 1 \end{bmatrix},\tag{3}\\ |\mathbf {S}^{H}\mathbf {S}|_{\mu }=&\begin{bmatrix} 1 & \rho & \rho & \rho \\ \rho & 1 & \rho & \rho \\ \rho & \rho & 1 & \rho \\ \rho & \rho & \rho & 1 \end{bmatrix},\tag{4}\\ |\mathbf {S}^{H}\mathbf {S}|_{\textrm {d}_{\textrm {cord}}}=&\begin{bmatrix} 1 & 0 & \rho _{13} & \rho _{14} \\ 0 & 1 & \rho _{23} & \rho _{24} \\ \rho _{13} & \rho _{23} & 1 & 0 \\ \rho _{14} & \rho _{24} & 0 & 1 \end{bmatrix}.\tag{5}\end{align*}
B. NOMA vs MU-MIMO
A generalized block diagram of the baseband transmitter for NOMA implementation is shown in Fig. 1. The information bits of a UEk are channel coded and then digitally modulated. For a bit-level NOMA implementation, the channel coded bits may be scrambled by a UE specific scrambling sequence and then digitally modulated. Symbol-level UE specific NOMA block appears after the quadrature amplitude modulation (QAM)-modulation block. Using WSMA, each incoming QAM-symbol
With \begin{equation*} \mathbf {y} = \sum \limits ^{K}_{i=1} \mathbf {h}_{k} \odot \mathbf {s}_{k} \sqrt {p_{k}} q_{k} + \mathbf {z},\tag{6}\end{equation*}
With WSMA, each UE uses
The baseline system for comparison with NOMA could be an OMA setup when there are
In another case, a baseline system for comparison could be based on MU-MIMO [46], [47]. In this case, both the NOMA and the MU-MIMO systems could be compared for the same number of users per RE. In addition to the frequency domain, the space domain provides additional degrees of freedom (DoF) to the BS. With multiple receive antennas at the BS, a joint space-frequency multiuser detector may be employed. The MU-MIMO system relies only on the spatial separation while NOMA has additional frequency domain, the assumed spreading domain, for UE separation. The same multiuser detector, with a little or no modification in implementation, may be used for both NOMA and MU-MIMO. For the MU-MIMO, an additional UE grouping and scheduling each group over orthogonal REs must also be considered for a fair comparison. Since increasing UE density is one of the NOMA objectives, a comparison for the maximum number of admissible UEs at a given target BLER may also be verified while comparing NOMA and MU-MIMO.
Considering ideal channel estimation, Figs. 2 and 3 show the link-level performance comparison of WSMA with MU-MIMO when the modulation is QPSK and the transport block size (TBS) is 20 bytes. Also, we concentrate on the massive machine type communications (mMTC) scenario of Rel-15 and the inter-BS distance is set to 1732m. The carrier frequency is 700 MHz and we assume that the channels follow the Tapped Delay Line (TDL-C) model [48, Sec. 7.7.2] with the UE moving at 3kmph. Note that TDL-C channel model may be used for simplified evaluation of non-line-of-sight (NLOS) communication. The considered channel model for link-level simulations suffices the NOMA setup which usually targets a high user density coupled with low mobility and small delay spread values. The channel’s desired rms delay spread is 30ns. Thus, with the considered speed of the UEs, they experience a flat fading channel. Also,
There are four receive antennas at the BS and each UE is equipped with a single antenna. It is assumed that each UE is transmitting with a unit power value over its allocated 6 physical resource blocks (PRBs) and 12 data OFDM symbols. The detector at the BS is MMSE (M: minimum) based. With the spread length 4, the codebook is based on the PI TSC. Here, the spread length refers to the number of resources over which a single transmit symbol is repeated in a weighted manner. That is, for our considered WSMA setup, it refers to
From Figs. 2 and 3, it can be observed that, for the assumed setup and various values of
With NOMA, on the other hand, due to symbol repetition by the low correlation spreading, the energy per resource element (RE) on an average is reduced (note: the SS are unit norm). This ensures that users’ signals perceive a lower MAI. This, however, comes at a possible reduced spectral efficiency, since each NOMA UE will consume
Figure 4 compares WSMA and MU-MIMO for both
Note that, in the simulations, we have concentrated on WSMA-based NOMA, as an efficient method supporting high user density/throughput with low implementation complexity and little or no modification in the existing OMA-based receivers, e.g., [10], [11], [40]. Then, as shown in [50, Fig. 2], with different receivers and synchronized transmission, the achievable BLER of the WSMA, the MUSA, the SCMA and the RSMA schemes are (almost) the same for a broad range of SNRs. Thus, with fairly high accuracy, the qualitative conclusions of Figs. 2 and 3 hold for these types of NOMA as well. Also, to further improve the receiver’s performance, power variation at the UEs could be implemented on top of the spreading based mechanism, e.g., [44], [51]. This will possibly create sufficient variation in the effective channel conditions to enable signal separation. However, for a given NOMA scheme, it has been observed in, e.g., [44, Fig. 15], that the power variation does not offer significant return in gain over the equal power case. Finally, it should be noted that, as opposed to the BSs, in practice the UEs have lower capability/accuracy in power allocation.
C. NOMA for Beyond 5G
During the NOMA Study in 3GPP for 5G NR, a large number of link- and system-level simulations of transmission schemes and corresponding receivers were carried out [33]. In both the link- and the system-level simulations, three scenarios, namely, mMTC, ultra reliable low latency communications (URLLC) and enhanced mobile broadband (eMBB), have been considered with a broad range of parameter settings. Particularly, 14 different companies, each with its own NOMA scheme, provided link-level results and studied the BLER for more than 35 cases. The link-level parameters were generally well aligned among companies, which enabled easy comparison between different methods. Moreover, all NOMA schemes, including those supported by Rel-15, performed similarly at link-level in key conditions. Then, 8 companies provided system-level simulation results, where in total 37 different sets of NOMA versus baseline results were provided. As opposed to the cases with link-level simulations, widely different parameter sets were used in the system-level simulations, with different baselines, making comparisons intractable. Here, the results have been presented for both synchronous and asynchronous operation models, while the main focus was on the synchronous operation.
According to the results presented during the 3GPP study-item, in ideal conditions, NOMA can be better or worse than MU-MIMO, depending on the number of UEs and simulation parameters. With realistic channel estimation in multipath, however, the relative performance gain of NOMA decreases, and MU-MIMO may outperform NOMA, depending on the parameter settings/channel model.
In a more general point of view, different types of NOMA techniques, including RSMA, IDMA, PDMA, UGMA, MUSA, SCMA, LCRS, WSMA, NOCA, NCMA, IGMA, ACMA, and LSSA, have been considered in the link-level simulations, and their performance have been compared with different existing Rel-15 techniques. However, considering these techniques and different parameter settings/simulation conditions, 1) no specific NOMA technique showed considerably better performance, compared to other schemes, and 2) no clear gain from NOMA over Rel-15 mechanisms were observed in all studied scenarios (see [33, Ch. 8] for details of link-level simulation results). Moreover, in a large number of conditions, the system-level simulations from different companies, mainly concentrating on SCMA, MUSA, RSMA, and PDMA, showed no conclusive gain over Rel-15 techniques (see [33, Ch. 9] for details of system-level simulation results). Also, for all URLLC, eMBB and mMTC scenarios and different NOMA approaches, considerable performance degradation is observed in the cases with non-ideal channel estimation, while the effect of channel estimation is more visible in the URLLC scenario.
In summary, in harmony with our results presented in Figs. 2–4, it was hard to find worthwhile NOMA gains. Moreover, as reported by different companies, the relative performance gain of NOMA, compared to exisiting OMA schemes, is the cost of considerable increment in implementation complexity [33]. These were the main reasons that 3GPP decided not to continue with NOMA as a work-item, and leave it for beyond 5G where new use-cases with ultra-dense UEs may be motivating for NOMA.
Reducing the Implementation Complexity
According to the discussions in 3GPP on NOMA, along with the low performance gain of NOMA, compared to existing Rel-15 techniques, one of the key challenges of NOMA is the implementation complexity, in different terms of UE pairing, signal decoding, CSI acquisition, etc. This is specially because NOMA is useful in dense networks where the implementation complexity of the system increases rapidly with the number of UEs. This is the motivation for this section, in which we propose different techniques to reduce the implementation complexity of NOMA. These results are interesting because 1) they provide guidelines to use NOMA with relatively low complexity. Also, 2) each of the proposed schemes, which have been filed in patent applications, can be studied analytically by academia in a separate paper.
For generality and in harmony with the discussions in 3GPP, we present the proposed schemes for UL NOMA. However, as explained in the following, it is straightforward to extend our proposed approaches to the cases with DL transmission. We consider the cases with pairing a cell-center and a cell-edge UE, i.e., UE1 and UE2 in Fig. 5, respectively, with
UL NOMA. UEs with different channels qualities are paired and the BS performs SIC to decode the signals sequentially.
A. HARQ Using NOMA
Due to the CSI acquisition and UE pairing overhead, NOMA is of most interest in fairly static channels with no frequency hopping where channels remain constant for a number of packet transmissions. As a result, the network suffers from poor diversity. Also, NOMA is faced with error propagation problem where, if the receiver fails to decode a signal, its interference affects the decoding probability of all remaining signals which should be decoded sequentially. For these reasons, there may be a high probability for requiring multiple HARQ retransmissions leading to high end-to-end (E2E) packet transmission delay [25], [26], [27]. The following schemes develop NOMA-HARQ protocols with low implementation complexity.
1) Smart NOMA-HARQ [52]
Our proposed retransmission process is explained in Fig. 6. Assume that in Slot 1 the BS can not decode correctly the signals of the UEs, i.e.,
Reducing the expected number of retransmissions in NOMA. If the BS fails to decode both signals, it asks for retransmission from only one of the UEs, while the other UE delays the retransmission. The retransmission gives the chance to decode the retransmitted signal. Then, removing the interference, the BS can decode the other failed signal interference-free and with no need for retransmission.
In this way, NOMA gives an opportunity to reduce the number of retransmissions, and improve the E2E throughput. Also, the fairness between the UEs increases because the required number of retransmissions of the cell-edge UE, i.e., UE2 in Fig. 6, decreases remarkably. The keys to enable such a setup are that 1) the BS should decode all buffered signals in each round and 2) it should inform the UEs about the appropriate retransmission times.
Finally, note that for DL transmission the proposed scheme is adapted as follows. With DL transmission, if none of the UEs can decode their signals correctly, the BS first retransmits the message of the cell-edge UE only, while the cell-center UE receives new messages and buffers the undecoded signals. The cell-edge UE uses typical decoding schemes to decode its own message based on all accumulated signals. On the other hand, following the SIC-based decoding approach, the cell-center UE first decodes and removes the message of the cell-edge UE based on all accumulated retransmitted signals. Then, it tries decoding all of its own received (new and undecoded) signals with no need for retransmissions.
2) Dynamic UE Pairing in NOMA-HARQ [53]
Here, the objective is to improve the performance gain of NOMA-HARQ by adding virtual diversity into the network. In our proposed setup, depending on the message decoding status, different pairs of UEs may be considered for data transmission in different retransmission rounds. As an example, considering Fig. 5, assume that UE1 and UE2 with
Finally, to apply the proposed scheme in DL transmission, the BS can consider a set of predefined UE pairing configurations for each UE. Then, depending on the UEs message decoding conditions, the BS switches to different pairing configurations in different retransmission rounds. Also, with the considered UE pairing and the number of retransmission round, the BS adapts the transmission powers, rates, as well as beamforming and informs the UEs about the considered pairing configuration. The UEs, on the other hand, adapt their decoding scheme based on the instantaneous pairing configuration such that all received copies of each signal are used for message decoding.
3) Multiple Access Adaptation in Retransmissions [54]
Our proposed scheme can be well explained in Fig. 7. In our proposed setup, each UE starts data transmission in its own dedicated bandwidth in an OMA-based fashion. Then, if a UE’s message is not correctly decoded in a time slot, in the following retransmission rounds it is allowed to reuse the bandwidth of the other UE as well. Let us denote the resource block at time
Multiple access adaptation in different (re)transmission rounds. If the signal of an UE is not correctly decoded, it has the chance to reuse the spectrum resource of the other UE during retransmissions.
In this way, compared to the cases with conventional OMA techniques, using the adaptive multiple access scheme, along with HARQ, makes it possible to exploit the network/frequency diversity and increase the UEs’ achievable rates. Moreover, our proposed scheme satisfies the tradeoff between the receiver complexity and the network reliability, and, compared to the state-of-the-art OMA-based systems, improves the service availability/the network reliability significantly. Finally, the proposed scheme improves the fairness between UEs and is useful in buffer-limited systems. Also, note that, while we presented the proposed NOMA-HARQ schemes for RTD (Type II) HARQ [55], [56], the same approaches are applicable for other HARQ protocols as well.
B. Simplifying the UE Pairing
With NOMA, optimal UE pairing becomes challenging as the number of UEs increases, because it leads to huge CSI acquisition and feedback overhead as well as running complex optimization algorithms [28], [29], [30]. For these reasons, we present low-complexity UE pairing schemes as follows.
1) Rate-Based UE Pairing [57]
Consider a dense network with
To limit the CSI requirement, in [57], we propose that the UE pairing is performed only based on the UEs rate demands and the probability of successful pairing. The proposed scheme is based on the following procedure:
Step 1: The BS asks all UEs to send their rate demands.
Step 2: Receiving the UEs’ rate demands and without knowing the instantaneous CSI, the BS finds the probability that two specific UEs can be successfully served through NOMA-based data transmission (see [57] for the detail procedure of finding these probabilities).
Step 3: If the probability of successful pairing for two specific UEs, i.e., the probability that the BS can correctly decode their signals, exceeds some predefined threshold, the BS assigns resources for UL transmission and asks those UEs to send pilots sequentially.
Step 4: Using the received pilots from those paired UEs, the BS estimates the channel qualities in that specific resources, decides if the UEs can be paired and determines the appropriate power level of each UE such that their rate demands can be satisfied.
Step 5: The BS informs the paired UEs about the power levels to use and sends synchronization signals such that their transmit timings are synchronized.
In this way, with our proposed scheme the CSI is acquired only if the BS estimates a high probability for successful UE pairing. This reduces the CSI overhead considerably, particularly in dense networks and/or in the cases with multiple antennas at the UEs. Finally, as we show in [57], to have the maximum number of successful paired UEs, the BS can initially consider the pairs with the highest and lowest rate demands. For instance, assume
Finally, note that the same approach can be applied for DL transmission, except that Step 1 is not required, because the BS already knows the size of the buffered data for each UE. Also, in Step 5, instead of asking the UEs to adapt their transmit powers, the BS informs the UEs to adapt their decoding scheme depending on the considered pairing method.
2) UE Pairing in CoMP-NOMA
The high-rate reliable backhaul links give the chance to simplify the UE pairing in coordinated multi-point (CoMP) networks using NOMA, e.g., [58]. The idea can be well presented in Fig. 8. Here, depending on the UEs positions and rate demands, they may be served with different multiple access schemes. For instance, in Point A (resp. C) of Fig. 8 where the channel
UE pairing in CoMP-NOMA. If a cell-edge UE can be successfully paired with a cell-center UE of one BS, it can be paired with each of the cell-center UEs of other BSs with SIC-based receiver only at one BS.
The advantages of the proposed scheme are: 1) SIC-based receiver is used only in BS1. Also, 2) UE pairing algorithm can be run only in one of the BSs. That is, NOMA-based data transmission is used as long as at least one of the BSs can find a good pair for UE2. Finally, 3) pairing (UE1, UE2), BS2 can consider each of its own cell-center UEs to be paired with them as long as the interference to BS1 is not high. That is, BS2 does not need to run advanced UE pairing algorithms.
C. Receiver Adaptation
Compared to OMA-based systems, the sequential decoding process of the BS may lead to large E2E transmission delay, as well as high receiver complexity/energy consumption [4], [31], [32]. Therefore, it is beneficial to use the sequential decoding only if there is high probability for successful decoding. This is the motivation for the scheme proposed in the following.
Considering Fig. 9, if the signal of UE1 is correctly decoded, the BS continues in the typical SIC-based receiver scheme to first remove the signal of UE1 and then decode the signal of UE2 interference-free (see Slot 1). On the other hand, if the BS fails to decode the signal of UE1, it does not continue message decoding and immediately sends NACKs to both UEs without decoding the UE2’s signal. This is motivated by the fact that with NOMA the transmission parameters, e.g., rate, power, of UE2 are designed based on the assumption that the BS can decode and remove the message of UE1 and, as a result, it decodes the message of UE2 interference-free. Then, with an unsuccessful decoding of the UE1’s message, the BS needs to decode the message of UE2, with poor UE2-BS link quality, in the presence of UE1’s interfering signal and, with high probability, it fails to decode the message of UE2 correctly, while it increases the E2E transmission delay. For instance, let us denote the decoding delay for decoding a codeword of length
Adapting the decoding scheme based on the estimated successful decoding probability. If the message of UE1 is correctly decoded in Slot 1, the BS continues in the SIC-based receiver scheme to first remove the signal of UE1 and then decode the message of UE2 interference-free (see Slot 1). On the other hand, if the BS fails to decode the message of UE1, it immediately sends NACKs for both UEs without decoding the message of UE2. Also, the BS buffers the undecoded signals for process in the next rounds of HARQ (see Slot 2). Then, depending on the UEs message decoding status at the BS, in each time slot the UEs’ data transmission is synchronized correspondingly.
In this way, the proposed setup reduces the implementation complexity considerably and improves the E2E throughput because the decoding scheme is adapted depending on the estimated probability of successful decoding. Particularly, an interested reader may follow the same method as in [59] to study the E2E performance gain of the proposed scheme analytically. Also, while we presented the setup for the cases with two UEs, it can be shown that the relative performance gain of the proposed scheme increases with the number of paired UEs.
Finally, the proposed approach can be well applied in DL transmission where, if the cell-center UE fails to decode a signal, it stops decoding the following signals and informs the BS immediately. Here, it is interesting to note that, as opposed to the UEs, energy consumption at the BS may not be a problem. Therefore, with an UL transmission the main gain of the proposed scheme is in E2E transmission delay reduction, while it is useful in improving the energy efficiency of the cell-center UE during DL transmission.
Conclusion
In this paper, we studied the challenges and advantages of NOMA as a candidate technology in dense networks. As we showed through simulations and in harmony with the discussions in the 3GPP Release 15 study-item on NOMA, NOMA may or may not outperform the typical OMA-based schemes such as MU-MIMO, in terms of BLER. However, for the current use-case scenarios of interest, the relative performance gain of NOMA was not so much such that it could not convince the 3GPP to continue with it as a work-item. On the other hand, the unique properties of NOMA give the chance to develop different techniques reducing its implementation complexity, which may make it more suitable for practical implementation. Therefore, there is a need to improve the spectral efficiency and the practicality of implementation, in order to have NOMA adopted by the industry.