A Survey Of Non-Orthogonal Multiple Access For 5G

3TABLE INOMA SCHEMES PROPOSED FOR THE R EL -14 3GPP NR S TUDY I TEM .NOMA schemes234Power-domainNOMA [11]-[58]SCMA [69]-[90]MUSA [91]PDMA [107] [108]5LSSA [121]67891011RSMA [122]-[124]IGMA [125]IDMA [126]NCMA [127] [128]NOCA [129]GOCA [130]12LDS-SVE [137]131415FDS [138]LCRS [138]RDMA kiaLGENokiaMTKFujitsuIntelIntelMTKFull NamePower-domainnon-orthogonal multiple accessSparse code multiple accessMulti-user shared accessPattern division multiple accessLow code rate and signature basedshared accessResource spread multiple accessInterleave-grid multiple accessInterleaver division multiple accessnon-orthogonal coded multiple accessNon-orthogonal coded accessGroup orthogonal coded accessLow density spreading-signaturevector extensionFrequency domain spreadingLow code rate spreadingRepetition division multiple accessI IntroductionII Basic principles and advantages of NOMAII-A Channel capacity comparison of OMA and NOMAII-B Advantages of NOMA compared with OMAIII Dominant NOMA solutionsIII-A Power-Domain NOMAIII-A1 Basic NOMA relying on a SIC receiverIII-A2 NOMA in MIMO systemsIII-A3 Cooperative NOMAIII-A4 NOMA in CoMPIII-A5 Application of power-domain NOMAIII-B Code-Domain NOMAIII-A1 LDS-CDMAIII-A2 LDS-OFDMIII-A3 SCMAIII-A4 MUSAIII-A5 SAMAIII-C Other NOMA schemesIII-D User grouping and resource allocationIII-E Comparison of NOMA solutionsIII-F Performance evaluations and transmission experimentsIV Challenges, opportunities, and future research trendsV ConclusionsFig. 2.The structure of this article.different users’ signals at the receivers in the time domain.In CDMA, multiple users can share the same time-frequencyresources, while the transmitted symbols of different usersmay be mapped to orthogonal spreading sequences, such asWalsh-Hadamard codes. Hence, a low-complexity decorrelation receiver can be used for multi-user detection (MUD).OFDMA can be regarded as a smart integration of FDMA andTDMA, where the radio resources are orthogonally partitionedin the time-frequency grid. Theoretically, as a benefit oforthogonal resource allocation, there is no interference amongusers in OMA systems, hence low-complexity detectors havinga linearly increasing complexity as a function of the number ofusers can be used to separate the different users’ signals [115].However, the maximum number of supportable users is rigidlyUplink (UL)Downlink ULrestricted by the number of orthogonal resources available inconventional OMA schemes, which becomes a hard limit whenmassive connectivity is required for 5G. Additionally, it hasbeen theoretically shown that OMA cannot always achieve themaximum attainable sum-rate of multi-user wireless systems,while NOMA is capable of achieving the multi-user capacitywith the aid of time-sharing or rate-splitting if necessary [116],which will be detailed in the following subsection.In order to circumvent the above limitation of OMA,NOMA has been recently investigated as a design alternative.The key distinguishing feature of NOMA is that of supportinga higher number of users than the number of orthogonalresource slots, which is achieved with the aid of sophisticated non-orthogonal resource allocation. More explicitly,carefully selected inter-user interference cancellation has tobe employed at the cost of an increased receiver complexity.Furthermore, it is worth bearing in mind that even if weopt for OMA schemes, the time-domain signals are smearedby their convolution with the dispersive channel impulseresponse (CIR). In addition, a historic concept of NOMAis constituted by the family of CDMA systems relying onnon-orthogonal spreading sequences. For example, we havemany more m-sequences than the number of chips Nc ina specific m-sequence. Hence an m-sequence-based CDMAsystem is copable of supporting significantly more than Ncusers, albeit at the cost of imposing gradually increased interuser interference, as the number of users is increased. Thisincreased interference can only be efficiently mitigated withthe aid of powerful multi-user detectors. Hence, the NOMAconcept is appealing. The family of NOMA schemes can bebasically divided into two categories: power-domain NOMAand code-domain NOMA. In power-domain NOMA, differentusers are assigned different power levels according to theirchannel quality, while the same time-frequency-code resourcesare shared among multiple users. At the receiver side, powerdomain NOMA exploits the users’ power-difference, in orderto distinguish different users based on successive interference cancellation (SIC). Code-domain NOMA is similar to

4R2 (b/s)p1 h1A3.462p2 h22.20N 0 10 dB N 0 10 dBCNOMAB0.9302OMAR1 (b/s)0.93 2.20 3.46(a)R2 (b/s)p1 h1A1.000.80p2 h22where W is the bandwidth, Pi is the transmitted power,and N0 is the power spectral density of Gaussiannoise. More particularly, according to the capacity analysis found in the pioneering contribution of Tse andViswanath [116], Fig. 3 and Fig. 4 from [5] portraysthe channel capacity comparison of OMA and NOMA,where a pair of users communicating with a base station(BS) over an AWGN channel is considered as an examplewithout loss of generality. Fig. 3 show that the uplinkof NOMA is indeed capable of achieving the capacityregion, while OMA is suboptimal in general, except atone specific point. However, at this optimal point, ratefairness is not maintained, since the rate of the low-poweruser is much lower than that of the higher-power user,when the difference of the received powers of the twousers is high. Note that the results for the simple two-usercase can be extended to the general case of an arbitrarynumber of users [116]. Explicitly, it is shown in [116]that there are exactly K! corner points when the K-userscenario is considered and the K-user NOMA system iscapable of achieving the same optimal sum rate at all ofthese K! corner points.In the downlink, the boundary of the capacity region isgiven by the rate tuples [118]: N 0 20 dB2N 0 0 dBNOMAOMA0.067C03.705.67B6.61 6.66R1 (b/s)(b)Fig. 3. Channel capacity comparison of OMA and NOMA in the uplinkAWGN channel: (a) Symmetric channel; (b) Asymmetric channel [5] c IEEE.R2 (b/s)p1 h13.46p2 h222N 0 10 dBN 0 10 dBNOMAOMA3.46R1 (b/s)(a)R2 (b/s)p1 h11p2 h20.82 Rk W log 1 N 0 20 dB2AWGN channel: In the uplink of an AWGN channelsupporting K users (K can be larger than 2), the capacityof the multiple access channel can be formulated as [117] KPPKiX i 1 (1)Ri W log 1 N0 W ,i 1N 0 0 dBPk hk N0 W 2KP!Pj , (2) 2 hk j k 10.6NOMA0.4OMAwhich is valid for all possible splits P 0.2KPPk of totalk 1A. Channel capacity comparison of OMA and NOMApower at the BS. The optimal points can be achievedby NOMA with the aid of superposition coding at thetransmitter and SIC at each of the receivers [116]. Moreparticularly, Fig. 4 showed that the boundary of the ratepairs of NOMA is in general beyond that of OMA inasymmetric channels.Fading channels: In fading channels, the sum capacityin the uplink - provided that the channel state information(CSI) is only known at the receiver - can be representedas!)(PK2k 1 hk Pave, (3)Csum E log 1 N0From the perspective of information theory, for the capacityof multiple access systems operating both in additive whiteGaussian noise (AWGN) and fading scenarios, we have thefollowing results for OMA and NOMA (applicable to bothpower-domain NOMA and code-domain NOMA):where we assume that each user has the same averagepower Pave . Hence OMA remains suboptimal in theuplink, while NOMA relying on MUD is optimal [116].MIMO-NOMA and MIMO-OMA: The versatileNOMA concept can be also extended to MIMO scenarios,01234567R1 (b/s)(b)Fig. 4. Channel capacity comparison of OMA and NOMA in the downlinkAWGN channel: (a) Symmetric channel; (b) Asymmetric channel [5] c IEEE. CDMA or multi-carrier CDMA (MC-CDMA), except for itspreference for using low-density sequences or non-orthogonalsequences having a low cross-correlation.

5where the BS has M antennas and each user is equippedwith N antennas. Additionally, multiple users can berandomly grouped into M clusters with two users in eachcluster. It has been shown in [119] that MIMO-NOMAperforms better than MIMO-OMA in terms of its sumchannel capacity (except for transmission to a single userin MIMO systems), i.e., for any rate pair achieved byMIMO-OMA schemes, there is a specific power split forwhich MIMO-NOMA is capable of achieving rate pairsthat are strictly higher.B. Advantages of NOMA compared to OMAWe can see from the capacity analysis that it is feasiblefor NOMA to achieve a higher transmission rate than OMA.Specifically, the main advantages of NOMA compared to theclassical OMA can be summarized as follows: Improved spectral efficiency and cell-edge throughput: The time-frequency resources are shared nonorthogonally among users both in the power-domainNOMA and in the code-domain of NOMA. As describedabove, in the uplink of AWGN channels, although bothOMA and NOMA are capable of achieving the maximum attainable sum capacity, NOMA supports a moreequitable user fairness. Additionally, the capacity boundof NOMA is higher than that of OMA in the downlinkof AWGN channels. In multi-path fading channels subjected to inter-symbol-interference (ISI), although OMAis indeed capable of achieving the maximum attainablesum capacity in the downlink, NOMA relying on MUDis optimal, while OMA remains suboptimal, if the CSI isonly known at the downlink receiver. Massive connectivity: Non-orthogonal resource allocation in NOMA indicates that the number of supportableusers/devices is not strictly limited by the number of orthogonal resources available. Therefore, NOMA is capable of significantly increasing the number of simultaneousconnections in rank-deficient scenarios, hence it has thepotential to support massive connectivity. Of course, itshould be noted that some practical implementation issuesin NOMA systems, such as its hardware imperfectionsand computational complexity, may hinder the realizationof massive connectivity, which will be detailed in SectionIV. Low transmission latency and signaling cost: In conventional OMA relying on access-grant requests, a userfirst has to send a scheduling request to the base station(BS). Then, upon receiving this request, the BS schedulesthe user’s uplink transmission by responding with a clearto-send signal in the downlink channel. Thus, a hightransmission latency and a high signaling overhead willbe imposed, which becomes unacceptable in the case ofmassive 5G-style connectivity. Specifically, the accessgrant procedure in LTE takes about 15.5 ms before thedata is transmitted [120]. In this way, the radical requirement of maintaining a user delay below 1 ms cannot bereadily satisfied [112]. By contrast, dynamic schedulingis not required in some of the uplink NOMA schemes.To elaborate a little further, in the uplink of a SCMAsystem, grant-free multiple access can be realized forusers associated with pre-configured resources defined inthe time- and frequency-domain, such as the codebooks,as well as the pilots. By contrast, at the receiver blinddetection and compressive sensing (CS) techniques can beused for performing joint activity and data detection [91].Hence again, beneficial grant-free uplink transmission canbe realized in NOMA, which is capable of significantlyreducing both the transmission latency and the signalingoverhead. Note that in some NOMA schemes using SICreceivers, the SIC process may impose extra latency.Therefore, the number of users relying on SIC shouldnot be excessive, and advanced MIMO techniques can beinvoked for serving more users, as discussed in SectionIII. Relaxed channel feedback: The requirement of channelfeedback will be relaxed in power-domain NOMA, because the CSI feedback is only used for power allocation.Hence there is no need for accurate instantaneous CSIknowledge. Therefore, regardless whether fixed or mobileusers are supported, having a limited-accuracy outdatedchannel feedback associated with a certain maximum inaccuracy and delay will not severely impair the attainablesystem performance, as long as the channel does notchange rapidly.Given the above prominent advantages, NOMA has beenactively investigated, with a views for employment in 5G asa promising solution. In the next section, we will discuss andcompare the dominant NOMA solutions.III. D OMINANT NOMA S OLUTIONSIn this section, we will discuss the families of prominant NOMA schemes by dividing them into two categories,namely power-domain and code-domain NOMA. Their designprinciples and key features will be highlighted, respectively.We will also provide their comparison in terms of theirspectral efficiency, system performance, receiver complexity,etc. At the end of this section, performance evaluations andtransmission experiments of NOMA will be discussed.A. Power-Domain NOMAIn this subsection, we will discuss the first category ofNOMA, namely, power-domain NOMA. In [14]–[17], theconcept and key features of power-domain NOMA have beendescribed in detail. In contrast to the multiple access schemesrelying on the time-, frequency-, code-domain or on theircombinations, NOMA can be realized in a recently emergednew domain, namely in the power domain. At the transmitter,different signals generated by different users are directlysuperimposed on each other after classic channel coding andmodulation. Multiple users share the same time-frequencyresources, and then are detected at the receivers by MUDalgorithms such as SIC. In this way, the spectral efficiency canbe enhanced at the cost of an increased receiver complexitycompared to conventional OMA. Additionally, it is widelyrecognized based on information theory that non-orthogonal

6multiplexing using superposition coding at the transmitter andSIC at the receiver not only outperforms classic orthogonalmultiplexing, but it is also optimal from the perspectiveof achieving the capacity region of the downlink broadcastchannels [116].Some practical considerations for power-domain NOMA,such as multi-user power allocation, signalling overhead, SICerror propagation and user mobility, were discussed in [14].To achieve a further enhancement of its spectral efficiency,the authors of [14]–[25] invoked a combination of NOMAwith MIMO techniques. Particularly, the capacity comparisonbetween MIMO-NOMA and MIMO-OMA has been investigated in [18] [19], and the superiority of MIMO-NOMAover MIMO-OMA in terms of both sum channel capacityand ergodic sum capacity was shown analytically. Furthermore, in [20] [21], the potential gains of MIMO-NOMAwere shown based on both link-level as well as on systemlevel simulations and using a NOMA test-bed developed. Ahardware SIC receiver was used for taking into account therealistic hardware impairments quantified in terms of the errorvector magnitude (EVM), the number of quantization bits inthe analog/digital (A/D) converter, etc. The simulation resultsand the measurements obtained showed that in the variet ofconfigurations considered, the cell throughput achieved byNOMA is about 30% higher than that of OFDMA. Furthermore, some open implementation issues were also discussedin [20] [21], including the granularity of the multi-user powerallocation both in time and frequency, as well as the signaling overhead, feedback enhancements and receiver designwere detailed. Additionally, the receiver design was discussedin [26]–[28]. A novel NOMA transmitter and receiver designwas proposed in [26], where the signals of multiple usersare jointly modulated at the transmitter side and detected atthe receiver side. In this scheme, the desired signal of thecell center user can be directly detected without detecting thesignal of the cell edge user, i.e., without SIC processing. Thus,a low complexity is achieved. Furthermore, the associatedsimulation results have shown that compared to the ideal SIC,the downlink NOMA link-level performance depends both onthe actual receiver design and on the difference in the powerratio split between the cell edge user and cell center user.Besides, the design and performance of the SIC receiver fordownlink NOMA combined with 2-by-2 open-loop SU-MIMObased on LTE TM3 (Transmission mode 3) were investigatedin [27], where different receiver weight generation schemeswere introduced both before SIC and after SIC according tothe transmission rank combination between the users. Thelink-level simulation results showed that the codeword levelSIC achieves higher performance than the symbol level SICand in fact approaches the performance of ideal SIC. Theimpact of applying the SIC receiver for cell-edge users indownlink NOMA using SU-MIMO was investigated in [28].The simulation results showed that there is an improvement ofthe NOMA gains over OMA in conjunction with SIC processing for the cell-edge users. Furthermore, in order to increasethe attainable performance of the SIC receiver, cooperativeNOMA transmission has been proposed in [29] [30]. A rangeof investigations related to multi-cell NOMA schemes werecarried out in [31]–[33]. Moreover, since having an increasednumber of cell-edge users typically degrades the efficiency ofcoordinated multi-point (CoMP) transmissions, this limitationwas circumvented by a promising NOMA solution proposedfor a CoMP system in [34]. Additionally, the performance ofNOMA techniques supporting randomly distributed users wasevaluated in [35]. These simulation results demonstrated thatthe outage performance of NOMA substantially depended bothon the users’ target data rates and on their allocated power.In [36]–[45], the system-level performance of power-domainNOMA was evaluated and the associated simulation resultsshowed that both the overall cell throughput and the celledge user throughput, as well as the degree of proportionalrate-fairness of NOMA were superior to those of OMA.Furthermore, the impact of the residual interference imposedby realistic imperfect channel estimation on the achievablethroughput performance was investigated in [46]–[48]. Onthe one hand, the channel estimation error results in residualinterference in the SIC process, which hence reduces theachievable user throughput. On the other hand, the channelestimation error causes error in the transmission rate controlfor the respective users, which may result in decoding errorsnot only at the destination user terminal but also at otheruser terminals owing to the error propagation imposed by theSIC process. A simple transmission rate back-off algorithmwas considered in [46] [47], and the impact of the channelestimation error was effectively mitigated. Simulation resultsshowed that NOMA achieves beneficial user throughput gainsover OMA in a scenario subject to channel estimation errors,which is similar to the case associated with perfect channelestimation.Let us now elaborate on the power-domain NOMA techniques in this subsection. Firstly, the basic principle of powerdomain NOMA relying on a SIC receiver will be discussed.Then, a promising extension relying on integrating NOMAwith MIMOs will be discussed for the sake of increasing itsattainable spectral efficiency. Another compelling extensionto a cooperative NOMA transmission scheme will also bepresented. Finally, the networking aspects of NOMA solutionswill be discussed.1) Basic NOMA relying on a SIC receiver: Firstly, weconsider the family of single antenna systems relying on asingle BS and K users.In the downlink, the total power allocated to all K usersis limited to P , and the BS transmits the signal xi to theith user subjected to the power-scaling coefficient pi . In otherwords, the signals destined for different users are weightedby different power-scaling coefficients and then they aresuperimposed at the BS according to:x KX pi xi ,(4)i 1where E[ x

able orthogonal resources in OMA.Another problem is that, despite the use of orthogonal time-, frequency- or code-domain resources, the channel-induced impairments almost invariably destroy their orthogonality. More specifically, considering that two signals s 1(t) and s 2(t), which are orthogonal either in