On Uplink Non-Orthogonal Multiple Access for 5G:Opportunities and Challenges

ZTE Corporation, South Keji Road, 55, Shenzhen 508118, China

I. INTRODUCTION

Orthogonal multiple access schemes have been used in cellular systems since the first generation, where physical resources are divided into time-frequency units and allocated to different users without overlapping. Frequency division multiple access (FDMA) and time division multiple access (TDMA) are two typical examples of orthogonal multiple access schemes.

In 3G cellular systems, code division multiple access (CDMA) is applied to facilitate non-orthogonal multiple access (NOMA) in uplink transmission [1]. Uplink signals from different users are spread using user-specific scrambling codes and superpose with each other in shared physical resource. Though multiple user interference is introduced by non-orthogonal transmission, the quality of service can be guaranteed the spreading factor is large. However, long spreading factor implies low data rate, which is only acceptable for voice/text services in 3rdgeneration(3G) wireless systems but not so suitable for wideband services in Long Term Evolution(LTE)/5thgeneration (5G) wireless systems.Further- more, intricate power control is needed to combat the near-far issue and maintain a same level of signal to interference and noise ratio (SINR) across users to guarantee the performance of cell edge users [2].

Another example of NOMA is multi-user superposition transmission (MUST) adopted in Long Term Evolution-Advanced (LTE-A) as a complementary multiple access technique [3][4][5]. Multiple users share the same physical resource in downlink to increase the system capacity. Diversity in power domain due to near-far effect is exploited in MUST, where 2 users with an appropriate power gap are paired for scheduling. Advanced receiver at the near user can cancel the interference of its paired far user. The following general principles are applied in MUST: 1) Far users should be assigned with low code rate and low modulation order, 2) Gray- mapping property of composite constellation should be ensured at transmitter via certain processing. By following the above principles, interference cancellation of far users can be achieved even without code-word level interference cancellation at user equipment (UE). Nevertheless, MUST is only applicable for grant-based systems where dynamic users pairing and scheduling is one of the key designs in base station (BS). Moreover, grant signaling related to MUST has to be sent to the paired users, so that each user gets necessary information to correctly decode its data from received downlink data packet. The mentioned necessary information includes: near/far user role, resource allocation, modulation order and code rate and so on. Since user pairing and corresponding grant signaling changes per transmission time interval (TTI) (1ms),this grant-based NOMA is more suitable for enhanced mobile broadband (eMBB) with large amount of downlink data to be sent. The signal overhead is negligible compared with data packet size. On the contrary, signaling overhead in grant-based NOMA becomes a major concern in massive machine type communication (mMTC) scenarios [6][7]. A large number of users with sporadic small data need to be served in uplink simultaneously, where grant-based NOMA is not the best choice. To solve this problem, grant-free NOMA has been studied as a promising solution, which is more suitable and efficient for systems with massive connections but infrequent small data packets.

In this paper, grant-free NOMA for uplink transmission is discussed. The paper is organized as follows. In Section II, scenarios and design targets of NOMA are elaborated. Candidates of NOMA schemes are generally summarized and preliminary comparison among a subset of schemes are presented in Section III. In Section IV, design aspects for grantfree NOMA are discussed, especially for those realistic issues which may be different from the traditional grant-based transmissions. The conclusions are provided in Section V.

II. SCENARIOS AND DESIGN TARGETS

In grant-free, as the name implies, UE can autonomously transmit packets without the need to send the scheduling request and waiting for the dynamic grant. The benefits of such scheduling request-free and grant-free are reduced signaling overhead, reduced UE power consumption, reduced latency, etc. Grant-free can be either orthogonal resource based, or non-orthogonal multiple access based. In the former,even though the resources themselves are orthogonal, different users may select the same resource, thus causing collision occasionally.Whenever such collision occurs, the link performance would be significantly degraded. Hence, the resource utilization of grantfree orthogonal based is not high. Grant-free NOMA is capable of handling more number of overlapped or collided users, without severe loss of performance, due to the transmitter side processing and the advanced receiver.Grant-free NOMA is a generic technology that can bring benefits to mMTC, URLLC, eMBB small data, 2-step RACH, etc. This grant-free transmission can be realized at different levels: 1) UE’s resources are pre-configured and periodically allocated, and each time when a packet arrives, the UE would choose the nearest allowable time-frequency resource for the uplink transmission, which is called semi-persistent-scheduling (SPS) based grant-free; 2)UE can randomly select a resource at any time for uplink transmission, leading to contention-based transmission. At both levels, base station (gNB) should perform blind detection either for UE identification or activation. SPS-based grant-free transmission fits for periodic traffic such as VoIP, or when the traffic load is not high such as URLLC. However, due to low resource utilization, SPS based grant free cannot achieve high spectral efficiency or high connection density for sporadic traffic, such as mMTC, eMBB small data, and 2-step RACH.

2.1 Ultra reliable low latency communication (URLLC)

The design targets of URLLC are low latency,high reliability and efficient resource utilization. Periodic or infrequent event-triggered traffic is typical in URLLC use cases. And due to the requirement of high reliability, data transmission is mainly operated in “RRC-connected” mode, which means that received signals from different UEs are time synchronized within cyclic prefix and the frequency offset is small. Here the resource includes time-frequency resources and multiple access (MA)signatures (i.e., spreading code, interleaver/scrambling pattern, demodulation reference signals (DMRS), preamble). When NOMA is applied for URLLC, the time- frequency resources can be shared by different user equipments (UEs), but the MA signatures (at least DMRS) should be uniquely pre-allocated without collision, as shown in figure 1, in order to achieve reliable user detection and accurate channel estimation. To support low latency, grant-free transmission with autonomous repetition can be adopted for URLLC where the resources are semi-statically configured by radio resource control (RRC) signaling (SPS-based grant-free). When aforementioned RRC configuration is completed, there is no need for dynamic grant or layer 1 (L1) activation/deactivation signaling in following transmission.

From the receiver side, on the one hand,advanced interference cancellation schemes may be needed to ensure high reliability in the presence of physical resource collision. On the other hand, the complexity should be moderate due to the low latency requirement.

2.2 mMTC

Different from URLLC, the design targets of mMTC are mainly massive connections with very low cost, low power consumption, and extended coverage.

The uplink traffic of mMTC tends to be sporadic with small packets. If traditional grant-based orthogonal transmission is applied for mMTC, contention-based random access has to be carried out for each uplink small data packet before the packet is to be transmitted.The number of simultaneously connected UEs is generally limited by the capacity of 5G Node B (gNB), base station. To support massive connections with limited resources,UE has to release its connection session and transfer back to inactive or idle state after the transmission is completed, without having no new packet to be sent. Apparently, signaling overhead sometimes may require even more time-frequency resources than the data packet, which leads to highly inefficient resource utilization. To solve this problem, grant-free transmission is proposed due to following advantages: 1) significantly reduced signaling overhead and transmission latency with minimum scheduling procedures, 2) low power consumption at UE side with reduced signaling processing, 3) low cost/complexity UE transceiver design relaxed requirements for time/frequency synchronization.

Based on the above analysis, it is beneficial to operate mMTC UL transmission in RRC inactive or RRC idle state. In this case, the resources and MA signatures of different UEs will not be uniquely allocated, and UE can randomly pick resources from the configured resource pool (fully grant-free). Collision of both physical resource and MA signature are inevitable, and the collision probability is non-negligible for mMTC scenario since the number of potential UEs that are simultaneously accessing the system would be quite large. For example, as shown in figure 2, the number of arrival packet per TTI is shown with different assumptions of inter-arrival time, considering the requirement of 1 million/km2connection density. Assuming that each UE’s packet occupies 1 physical resource block (PRB) and 1 ms resource, there will be more than 15% percentage that more than 6 packets are multiplexed on the same physical resource. The number of packets per TTI would be even larger considering retransmission and low code rate, which implies high overloading requirement in NOMA schemes.

From the viewpoint of receiver, UE detection and channel estimation has to be carried out in a “blind” manner. Moreover, with limited control signaling, synchronization within cyclic prefix cannot easily be maintained.Open-loop power control is another challenge for multi-user decoding and interference management. Compared with URLLC, latency requirement for mMTC is not so strict. Thus receiver complexity issue can be relatively less critical as long as multiple users can be decoded within the time slot(s) for one-shot transmission. Advanced receiver that is robust to channel uncertainties and physical resource or MA signature collision, e.g. successive interference cancellation (SIC) or maximum likelihood (ML)-type receiver, should be at high priority.

2.3 eMBB small packet

Grant-free NOMA can also benefit eMBB scenario with small packet transmission. The design targets would be low latency, low power consumption and signaling overhead reduction. The traffic of eMBB can be either periodic (e.g. voice services) or event-triggered infrequent traffic (e.g. irregular data uploading).Therefore, grant-free NOMA in eMBB can be designed either similar to URLLC, i.e., operated in RRC connected transmission, or similar to mMTC, i.e., RRC inactive transmission.For eMBB use cases, cell-edge user throughput is mainly the bottleneck of overall network performance. By enabling spreading-based NOMA and contention-based resource sharing, the operating point of cell-edge users can be lower, leading to less inter-site interferences and the improved spectral efficiency

2.4 Two-step RACH

RACH process, as an intrinsic contention-based transmission, can also be enhanced by grant-free NOMA. The design target is to improve the capacity of random access (i.e.similar to those for mMTC) while achieving accurate timing-advance (TA) estimation. Traditional four steps in RACH can be simplified to two steps shown in figure 3, where a oneshot transmission with preamble and data is transmitted together.

Fig. 2 Probability for the number of arrival packets per TTI

Fig. 3 Illustration of (a) traditional four-step RACH and (b) two-step RACH

With NOMA, spreading/interleaving/scrambling is applied at transmitter side.With advanced receiver, superposed two-step RACH signal (including preamble and data)from multiple UEs can be decoded, even in the presence of asynchronization and collision.And this can significantly increase the transmission efficiency of two-step RACH. Twostep RACH procedure starts from RRC idle and the UE identification is carried in the data part. Once this data is successfully decoded,the gNB would send a response to the UE.

III. CANDIDATE SCHEMES FOR UL NOMA

3.1 Candidate schemes

In grant-free NOMA, the code rate and modulation order of each user is not high. The target is to multiplex more number of users and to achieve higher sum spectral efficiency than grant-free orthogonal resource based transmission. A good system design of non-orthogonal multiple access needs to consider at least the following three aspects: 1) transmission scheme; 2) receiver implementation; 3) resource configuration and scheduling if applied.Transmission scheme is important to make non-orthogonal transmission more feasible.Transmitter side processing of grant-free NOMA is mainly to keep the per-UE spectral efficiency low, while introducing good characteristics of transmit signals to facilitate multi-user interference cancellation at the receiver side. There are different ways to keep the bit rate low and to distinguish different UEs, e.g., multiple access (MA) signatures.MA signatures can be spreading sequence/code, interleaver/scrambler pattern, or even preamble, demodulation reference signal. They may be operated at modulation symbol level,or at coded bit level or at both. In figure 4 we

Fig. 4 General structure of transmitter side processing for non-orthogonal multiple access schemes

show a general structure of transmitter side for non-orthogonal multiple access schemes.Non-orthogonal multiple access schemes may involve channel coding, interleaver/scrambling, bit-to-symbol mapping, or spreading.

Quite a number of schemes were proposed in 3GPP TR38.802 [8], targeting various aspects of the aforementioned scenarios. In Table 1, a few schemes are listed for example [9]-[18]. These schemes can be roughly categorized into three types: interleaver/scrambling based schemes, low code rate based schemes and spreading based schemes.Interleaver based schemes are usually operated at bit level, where inter-user interference is alleviated via the bit-level repetition and random permutation. Spreading based schemes are normally operated at symbol level, where the low inter-user interference is achieved by using low cross- correlation sequences which is called full-length spreading, or using low density codes as named sparse-sequence based spreading.

Receiver algorithms are usually standard transparent and up to manufacturers’ implementation. However in the case of non-orthogonal transmission, the receiver bears much higher burden than orthogonal transmission, in terms of inter-user interference cancellation. Therefore, accurate modelling of advanced receiver is crucial for the performance evaluation and the assessment of implementation complexity. The choice of receiver algorithms is highly coupled with the transmission schemes. For instance, minimum mean squared error and successive interference cancellation (MMSE-SIC) is often used for short sequence type of signatures with low cross-correlation where matrix inversion operation is more feasible, as shown in [12][19].In figure 5 the MMSE-SIC receiver is shown.After channel decoding, only signals passing cyclic redundancy check are applied for interference cancellation. For the module of transmitted signal regeneration, the successfully decoded bits are re-encoded, re-modulated and re-spread. Matched filter (MF) operation is more suitable for long sequence type of signatures and its computational complexity is very low [20]. To improve the performance of MF type receiver, MF-SIC receiver can be applied.Because MF cannot suppress the multiple user interference efficiently, there is certain performance loss when the number of users is large or near-far effect is significant [9]. Iterative MF-SIC is a different type receiver which contains elementary signal estimator (ESE)and maximum a posteriori probability (MAP)detector [9][10][21]. Because joint detection is decoupled into several single user detections with the information updating of the expectation and variance of the interferences, the detection complexity of ESE or MAP detector for iterative MF-SIC is relatively small, and its complexity only linearly increases with the number of users. While several iterative detections are needed to achieve good performance,decoding for each user is needed in each iterative detection. Therefore, the total computational complexity and signal processing latency may be high. MPA type of receivers can achieve the best performance in theory,however, the detection complexity exponentially increases with the modulation order and the number of users. Although some methods were proposed to reduce the detection complexity [22][23], those algorithms have not be well verified yet.

Needless to say, receiver design constitutes a big portion of hardware implementation of a non-orthogonal system. Realistic receiver needs to take into account of practical issues such as active user detection for grant-free transmission，non-ideal channel estimation,time and frequency offset handling, and receiver complexity.

Within the spreading sequence/code family, the sequence/code can be relatively long or short. For the long spreading type such as pseudo-random noise (PN) sequence, it is easy to maintain low cross-correlation between different users, and the size of sequence pool (2^N where N is the spreading length)can be large enough to achieve low collision probability in the case of random selection of the spreading sequences. For short sequence type, different properties can be imposed,for example, full-length spreading with low cross-correlation, or sparse spreading with low density. High overloading factor can be achieved, e.g. through the optimization of complex-valued codebook. As the number of user increases, the complexity of receiver is increased, the processing delay gets longer,and the performance is degraded due to the error propagation in successive interference cancellation. Long spreading in time domain can be useful for extreme coverage case, while short spreading code can be readily combined with multiple antennas to further decrease thecross-correlation and thus enhance the overloading capability.

Table I Categorization of the non-orthogonal multiple access schemes

Fig. 5 MMSE-SIC receiver

Table II Examples of NOMA schemes with short full-length spreading

F ig. 6 Cross-correlation properties of different full-length spreading schemes

3.2 Preliminary results

Preliminary performance comparisons are shown in this sub- section with several examples of short full-length spreading with low cross-correlation are listed in Table 2.

Given the suggested values of spreading length L and size of sequence pool K for each scheme, the cumulative distribution function (CDF) of cross-correlation is depicted in figure 6. It can be found that, the longer the spreading length, the smaller the overall cross-correlation can be achieved.

Figure 7 illustrate the performance comparison among MUSA, NCMA and NOCA at 200%overloading, with the ideal channel estimation.It is observed that NOCA with longer sequence and lower cross-correlation has the lowest block error rate (BLER), due to the higher frequency diversity gain. However, NOCA performs the worst for the realistic channel estimation, as shown in figure 8. It means that longer sequence suffers more severe error propagation of channel estimates since more UEs are multiplexed on the same resources given the same overloading factor. For MUSA and NCMA with the same spreading length and different pool sizes, there is no significant difference on the BLER performances.

Different from MUSA, NCMA and NOCA,additional random scrambling is performed in GOCA after symbol-level full- length spreading, aiming to lower the inter-user interferences. The performance between MUSA and GOCA are compared as in figure 9. Given the same spreading length and overloading factor, it is seen that scrambling provides little improvement in the BLER performance,assuming the ideal UE identification, e.g., the scrambling sequence of each UE is known to gNB. Note that in reality, blind detection of scrambling sequences would be quite difficult and thus lead to significant performance degradation.

It is noticed that, even for the simulation of realistic channel estimation, different UEs’DMRS resources are preconfigured without DMRS collision. So from the simulation results, it seems that without DMRS collision,different spreading sequences behave more or less the same in terms of BLER performance.This implies that how to avoid/solve RS collision and perform blind detection is more crucial to the design of grant-free NOMA.

IV. DESIGN ASPECTS

4.1 Grant-free

As mentioned in Section III, how to support grant-free is quite important for NOMA evaluations. The key issue for grant-free is that how to perform UE identifications. For uplink transmission in RRC-connected state, orthogonal DMRS can be preconfigured by gNB and used for both UE identification and channel estimation, if the traffic is not quite heavy such as URLLC or eMBB scenarios. Similarly, preamble (e.g. ZC- sequence) can be used for channel estimation and user detection for UL transmission of UE in RRC-inactive state such as mMTC or two-step RACH scenarios.However, in the latter case, UE is required to perform random selection within a sequence pool of preamble, and therefore collision between different UEs may happen. Preamble collision would lead to the following outcome:

1) if the SNRs of multiple users are similar,it is very likely that none of the users can be correctly decoded due to the strong cross-interference;

2) if one of the users has much higher SNR than others’, it is possible that only this user’s signal can be successfully decoded if the summation of the interference from the other users is negligible.

The collision probability, which is the probability that multiple users select the same preamble sequence, determines the number of superposed users. For example, assuming that the pool size is N and M UEs each randomly choose a preamble sequence from the pool, the collision probability can be calculated as:

In light of this, the size of the sequence pool N should be large enough to support multiple UEs to share the same resources. However, note that the size of the pool is also constrained by the length of preamble sequence which should not be too long in order to keep the overhead acceptable. Besides, the complexity of blind multi-user detection linearly increases as the pool size grows.

F ig. 7 Performance comparison between MUSA, NCMA and NOCA with ideal channel estimation

fig. 8 Performance comparison between MUSA, NCMA and NOCA with realistic channel estimation

Fig. 9 Performance evaluation between short full-length spreading with (GOCA)and without (MUSA) scrambling

Another way to support grant-free and blind multi-user detection (MUD) is to use data symbol itself, where full use of the available time/frequency resources is possible [24][25]. Here blind MUD means that when two UEs select the same physical resource, it is still possible to decode the UE with higher SINR based on roughly channel estimation.UE ID can be explicitly included in the data so that UE identification can be achieved once the data is successfully decoded. The decoded data can be further utilized to refine the channel estimation. The error propagation is minimized by code-word level interference cancellation and it is then possible to also decode the UE with lower SINR. The pros of data-only solution are: 1) the overhead for preamble or DMRS can be saved; 2) high overloading can be achieved since there is no preamble/DMRS collision. The cons of data only are 1) complexity is significant since all possible transmission hypotheses should be tested; 2) it is non-trivial to perform hybrid automatic repeat request (HARQ) retransmission and combing.

4.2 Open issues

There are still many open issues for grantfree NOMA, regardless of specific transmitter or receiver schemes, especially on the related procedures, such as

● UL transmission detection

● HARQ related procedures

● RRC and L1 signalling

■ Resource configuration

■ Link adaptation

■ Power control

● Switching between orthogonal and non-orthogonal multiple access

Grant-free NOMA is feasible only when the UL transmission of different UEs sharing the same resources can be correctly detected and identified. For UL grant-free transmission, L1 signalling may not always be needed for activation, and in this case, some of the parameters such as time/frequency resources, RS parameter, modulation and coding scheme (MCS)/transport block size (TBS)value, and power control related parameters should be configure via RRC signalling. In addition, HARQ related procedures including how many HARQ processes are supported,acknowledgement (ACK)/ Negative Acknowledgement (NACK) feedback scheme, and combining scheme of retransmissions should have proper design to further enhance the system performance. Considering the possibility that for extremely high overloaded traffic and extremely light traffic, NOMA may not be reliable or efficient, fallback mechanism from grant-free NOMA to grant-based orthogonal multiple access (OMA) should be allowed.

Another open issue is how to conduct appropriate performance evaluation and analysis for grant-free NOMA through link and system level simulation. At least the following aspect should be taken into account in the simulations:

● Traffic model and deployment scenarios of eMBB (small packet), URLLC and mMTC.

● Device power consumption.

● Coverage (link budget).

● Latency and signalling overhead.

● BLER reliability, capacity and system load.

Realistic modeling of transmitter (Tx)/receiver (Rx) impairments, for instance, potential peak-to-average-power ratio (PAPR) issue,channel estimation error, power control accuracy, resource collision, etc. should be considered. For system-level simulation, proper physical layer abstraction models for link-tosystem (L2S) mapping are still under developing, in order to take into account the blind UE detection and realistic Tx/Rx impairments.

V. CONCLUSIONS

In this paper, we discuss non-orthogonal multiple access (NOMA) technique for uplink data transmission. The use cases and design targets of NOMA for different scenarios are analyzed.Preliminary comparisons between several NOMA schemes are presented. From the performance evaluation, it seems that reference signal/data collision is crucial to the design of grant-free NOMA. Further study is needed for design aspects of grant-free NOMA, especially for the realistic issues such as blind detection, realistic channel estimation with possible RS collision and Tx/Rx impairments.

[1] W. Jiangzhou and N. Tung-Sang, Advances in 3G Mobile Enhanced Technologies for Wireless Communications, Artech House Publishers, Mar.2002.

[2] X. Li, T. Jiang, S. Cui, J. An and Q. Zhang, “Cooperative communications based on rateless network coding in distributed MIMO systems,”in IEEE Wireless Communications, vol. 17, no. 3,2010, pp. 60-67.

[3] S. Sesia, I. Toufik, and M. Baker, LTE-The UMTS Long Term Evolution: From Theory to Practice,Wiley Publishing, 2009.

[4] S. Parkvall, A. Furuska andr, and E. Dahlman,“Evolution of LTE toward IMT-advanced,” IEEE Communications Magazine, vol. 49, no. 2, 2011,pp. 84–91.

[5] 3GPP TR 36.859, Study on downlink multiuser superposition transmission (MUST).

[6] C. Bockelmann, N. Pratas, H. Nikopour, K. Au,T. Svensson, Č. Stefanović, P. Popovski, A. Dekorsy, “Massive Machine-type Communications in 5G: Physical and MAC-layer solutions,” IEEE Communications Magazine, vol.54, no.9, 2016,pp.59-65.

[7] Y. Wang, B. Ren, S. Sun, S. Kang and X. Yue,“Analysis of non-orthogonal multiple access for 5G,” in China Communications, vol. 13, 2016,pp. 52-66

[8] 3GPP, TR38.802, “Study on New Radio Access Technology: Physical Layer Aspects (Release 14)”, 2017.

[9] 3GPP, R1-1609892, “On the Performance of Interleaved-based Multiple Access Schemes,”InterDigital Communications.

[10] 3GPP, R1-163992, “Non-orthogonal multiple access candidate for NR,” Samsung.

[11] 3GPP, R1-1610918, “Link-level evaluation results of UL NOMA schemes,” Intel Corporation.

[12] 3GPP, R1-1608953, “Link-level performance evaluation for MUSA,” ZTE, ZTE Microelectronics.

[13] 3GPP, R1-1609223, “Further evaluation results of NCMA in UL LLS,” LG Electronics.

[14] 3GPP, R1-167249, “Non-orthogonal coded access (NOCA),” Nokia, Alcatel-Lucent Shanghai Bell.

[15] 3GPP, R1-1609332, “LLS results for GOCA scheme,” MediaTek Inc.

[16] 3GPP, R1-163510, “Candidate NR Multiple Access Schemes,” Qualcomm Incorporated.

[17] 3GPP, R1-164037, “LLS results for uplink multiple access,” Huawei, HiSilicon.

[18] 3GPP, R1-1608755, “LLS results of PDMA with realistic channel estimation,” CATT.

[19] Z. Yuan, G. Yu, W. Li, Y. Yuan, and X. Wang,“Multi-user shared access for internet of things,”in IEEE 83rd Vehicular Technology Conference,VTC Spring., vol. 1, 2016, pp. 1–5.

[20] 3GPP, R1-164689, “RSMA and SCMA comparison,” Qualcomm Incorporated.

[21] L. Ping, “Interleave-division multiple access and chip-by-chip iterative multi-user detection,”IEEE Communications Magazine, vol. 43, no. 6,2005, pp. S19-S23.

[22] Y-S. Cheng, M. J. Neely, and K. M. Chugg, “Iterative Message Passing Algorithm for Bipartite Maximum Weighted Matching,” IEEE International Symposium on Information Theory, July 2006, pp.1934 - 1938.

[23] Y. Wu, S. Zhang, Y. Chen, “Iterative multiuser receiver in sparse code multiple access systems,”IEEE International Conference on Communications, 2015, pp.2918-2923.

[24] 3GPP, R1-166404, “Receiver Details and Link Performance for MUSA,” ZTE, ZTE Microelectronics.

[25] Z. Yuan, C. Yan, Y. Yuan, W. Li, “Blind Multiple User Detection for Grant-free MUSA without Reference Signal”, IEEE 86th Vehicular Technology Conference, 2017, pp 1-5.