IEEE Communications Magazine • September 201574 0163-6804/15/$25.00 © 2015 IEEE

Linglong Dai, BichaiWang, and ZhaochengWang are with TsinghuaUniversity.

Yifei Yuan is with ZTECorporation.

Shuangfeng Han andChih-Lin I are with ChinaMobile Research Institute.

This work was supportedin part by the Interna-tional Science & Tech-nology CooperationProgram of China(Grant No.2015DFG12760), theNational Natural ScienceFoundation of China(Grant Nos. 61571270and 61201185), and theBeijing Natural ScienceFoundation (Grant No.4142027)..

1 Note that “NOMA” isalso used by NTT DoCo-Mo to refer to NOMAvia power domain multi-plexing.

INTRODUCTIONIn the history of wireless communications fromthe first generation (1G) to 4G, the multipleaccess scheme has been the key technology todistinguish different wireless systems. It is wellknown that frequency-division multiple access(FDMA) for 1G, time-division multiple access(TDMA) mostly for 2G, code-division multiple

access (CDMA) for 3G, and orthogonal frequen-cy-division multiple access (OFDMA) for 4G areprimarily orthogonal multiple access (OMA)schemes. In these conventional multiple accessschemes, different users are allocated to orthog-onal resources in either the time, frequency, orcode domain in order to avoid or alleviate inter-user interference. In this way, multiplexing gaincan be achieved with reasonable complexity.

However, the fast growth of mobile Internethas propelled 1000-fold data traffic increase by2020 for 5G. Hence, the spectral efficiencybecomes one of the key challenges to handlesuch explosive data traffic. Moreover, due to therapid development of the Internet of Things(IoT), 5G needs to support massive connectivityof users and/or devices to meet the demand forlow latency, low-cost devices, and diverse servicetypes. To satisfy these requirements, enhancedtechnologies are necessary. So far, some poten-tial candidates have been proposed to addresschallenges of 5G, such as massive MIMO, mil-limeter wave communications, ultra dense net-work, and non-orthogonal multiple access(NOMA) [1]. In this article, we focus on NOMA,which is highly expected to increase systemthroughput and accommodate massive connec-tivity. Note that Third Generation PartnershipProject (3GPP) Long Term Evolution (LTE)Rel-13 is doing ongoing studies toward NOMAin the form of multi-user superposition transmis-sion (MUST). NOMA allows multiple users toshare time and frequency resources in the samespatial layer via power domain or code domainmultiplexing. Recently, several NOMA schemeshave attracted lots of attention, and we can gen-erally divide them into two categories,1 that is,power domain multiplexing [2–4] and codedomain multiplexing, including multiple accesswith low-density spreading (LDS) [5, 6], sparsecode multiple access (SCMA) [7], multi-usershared access (MUSA) [8], and so on. Someother multiple access schemes such as pattern-division multiple access (PDMA) and bit divisionmultiplexing (BDM) [9] are also proposed. Keyfeatures and advantages of NOMA are discussed

ABSTRACT

The increasing demand of mobile Internetand the Internet of Things poses challengingrequirements for 5G wireless communications,such as high spectral efficiency and massive con-nectivity. In this article, a promising technology,non-orthogonal multiple access (NOMA), is dis-cussed, which can address some of these chal-lenges for 5G. Different from conventionalorthogonal multiple access technologies, NOMAcan accommodate much more users via non-orthogonal resource allocation. We divide exist-ing dominant NOMA schemes into twocategories: power-domain multiplexing andcode-domain multiplexing, and the correspond-ing schemes include power-domain NOMA,multiple access with low-density spreading,sparse code multiple access, multi-user sharedaccess, pattern division multiple access, and soon. We discuss their principles, key features, andpros/cons, and then provide a comprehensivecomparison of these solutions from the perspec-tive of spectral efficiency, system performance,receiver complexity, and so on. In addition, chal-lenges, opportunities, and future research trendsfor NOMA design are highlighted to providesome insight on the potential future work forresearchers in this field. Finally, to leverage dif-ferent multiple access schemes including bothconventional OMA and new NOMA, we pro-pose the concept of software defined multipleaccess (SoDeMA), which enables adaptive con-figuration of available multiple access schemesto support diverse services and applications infuture 5G networks.

SOFTWARE DEFINED 5G NETWORKS FORANYTHING AS A SERVICE

Linglong Dai, Bichai Wang, Yifei Yuan, Shuangfeng Han, Chih-Lin I, and Zhaocheng Wang

Non-Orthogonal Multiple Access for 5G:Solutions, Challenges, Opportunities,and Future Research Trends

DAI_LAYOUT_Author Layout 9/4/15 3:36 PM Page 74

IEEE Communications Magazine • September 2015 75

later. The design principles, key features, advan-tages and disadvantages of existing dominantNOMA schemes are discussed and compared.More importantly, although NOMA can provideattractive advantages, some challenging prob-lems should be solved, such as advanced trans-mitter design and the trade-off betweenperformance and receiver complexity. Thus,opportunities and research trends are highlight-ed to provide some insights on the potentialfuture work for researchers in this field. In addi-tion, unlike the conventional way of designing aspecific multiple access scheme separately andindividually, we propose the concept of softwaredefined multiple access (SoDeMA), in whichseveral candidates among multiple accessschemes can be adaptively configured to satisfydifferent requirements of diverse services andapplications in future 5G networks. Finally, con-clusions are drawn.

FEATURES OF NOMAIn conventional OMA schemes, multiple usersare allocated with radio resources which areorthogonal in time, frequency, or code domain.Ideally, no interference exists among multipleusers due to the orthogonal resource allocationin OMA, so simple single-user detection can beused to separate different users’ signals. Theo-retically, it is known that OMA cannot alwaysachieve the sum-rate capacity of multiuser wire-less systems [10]. Apart from that, in convention-al OMA schemes, the maximum number ofsupported users is limited by the total amountand the scheduling granularity of orthogonalresources.

Recently, NOMA has been investigated todeal with the problems of OMA as mentionedabove. Basically, NOMA allows controllableinterferences by non-orthogonal resource alloca-tion with the tolerable increase in receiver com-plexity. Compared to OMA, the main advantagesof NOMA include the following.

Improved spectral efficiency: According tothe multi-user capacity analysis in the pioneeringwork [10], Fig. 1 shows the channel capacitycomparison of OMA and NOMA, where twousers in the additive white Gaussian noise(AWGN) channel are considered as an examplewithout loss of generality. Figure 1a shows thatthe uplink NOMA is able to achieve the capacitybound, while OMA schemes are in general sub-optimal except at point C. However, at this opti-mal point, the user throughput fairness is quitepoor when the difference of the received powersof the two users is significant, as the rate of theweak user is much lower than that of the stronguser. In the downlink, Fig. 1b shows that theboundary of rate pairs of NOMA is outside ofthe OMA rate region in general. In multi-pathfading channels with intersymbol interference(ISI), although OMA could achieve the sumcapacity in the downlink, NOMA is optimalwhile OMA is strictly suboptimal if channel stateinformation (CSI) is only known at the mobilereceiver [10].

Massive connectivity: The non-orthogonalresource allocation in NOMA indicates that thenumber of supported users or devices is not

strictly limited by the amount of availableresources and their scheduling granularity.Therefore, NOMA can accommodate signifi-cantly more users than OMA by using non-orthogonal resource allocation; for example,MUSA can still achieve reasonably good perfor-mance when the overloading is 300 percent [8].

Low transmission latency and signaling cost:In conventional OMA with grant-based trans-mission, a user has to send a scheduling requestto the base station (BS) at first. Then, based onthe received request, the BS performs schedulingfor the uplink transmission and sends a grantover the downlink channel. This procedureresults in large latency and high signaling cost,which becomes worse or even unacceptable inthe scenario of massive connectivity anticipatedfor 5G. In contrast, such dynamic scheduling isnot required in some uplink schemes of NOMA,rendering a grant-free uplink transmission thatcan drastically reduce the transmission latencyand signaling overhead.

Due to the potential advantages above,NOMA has been actively investigated as apromising technology for 5G. In the next section,existing dominant NOMA schemes are discussedand compared in detail.

DOMINANT NOMA SOLUTIONSIn this section, we discuss dominant NOMAschemes by grouping them into two categories:power domain multiplexing and code domainmultiplexing. Power domain multiplexing meansthat different users are allocated different powerlevels according to their channel conditions toobtain the maximum gain in system perfor-mance. Such power allocation is also beneficialto separate different users, where successiveinterference cancellation (SIC) is often used tocancel multi-user interference. In this article,power domain multiplexing is applied only todownlink NOMA. Code domain multiplexing issimilar to CDMA or multicarrier CDMA (MC-CDMA), that is, different users are assigned dif-ferent codes, and are then multiplexed over thesame time-frequency resources. The difference

Figure 1. Channel capacity comparison of OMA and NOMA in an AWGNchannel: a) uplink AWGN channel; b) downlink AWGN channel.

NOMA

NOMA

OMA

A

C

B

Rate

of

user

2

Rate

of

user

2

Rate of user 1(a)

Rate of user 1(b)

OMA

DAI_LAYOUT.qxp_Author Layout 9/3/15 5:41 PM Page 75

IEEE Communications Magazine • September 201576

between power domain multiplexing and codedomain multiplexing is that code domain multi-plexing can achieve certain spreading gain andshaping gain at the cost of increased signal band-width.

NOMA VIA POWER DOMAIN MULTIPLEXINGBasic NOMA with a SIC Receiver: Figure 2aillustrates the basic NOMA scheme via powerdomain multiplexing with a SIC receiver in thedownlink. Note that this NOMA scheme canalso be applied in the uplink [2]. At the BStransmitter, signals for different users are linear-ly added up under certain power partitions tobalance the sum rate of all multiplexed usersand the throughput fairness among individualusers.

At the receiver, SIC is commonly used torealize multi-user detection (MUD). Due to thenear-far effect, the channel conditions may varysignificantly among users. SIC is performed atusers with relatively high signal-to-interference-plus-noise ratio (SINR), and should be carriedout in descending order of SINR.

As we can see, the basic form of NOMA withSIC exploits SINR difference among users,either due to the natural near-far effect or bynon-uniform power allocation at the transmitter.A similar scheme can be used for uplink toincrease the uplink system capacity.

NOMA in Massive MIMO Systems: NOMAcan be used in conjunction with multi-user mul-tiple-input multiple-output (MU-MIMO) to fur-ther improve the system spectral efficiency [3].As illustrated in Fig. 2b, multiple transmit anten-nas at a BS are used to form different beams in

the spatial domain, where each beam adopts thebasic NOMA discussed above.

At the receiver, the inter-beam interferencecan be suppressed by spatial filtering [3], andthen intra-beam SIC can be used to remove theinter-user interference. The extension of NOMAin massive MIMO systems can further improvespectral efficiency.

Network NOMA: When transmit power alloca-tion is biased toward far away users in downlinkNOMA, cell edge users experience increasedinterference from neighboring cells. As an exam-ple, a cellular system with two cells and fourusers is depicted in Fig. 2c, where a two-userNOMA scheme is assumed: user 1 and user 2are served by BS 1, while user 3 and user 4 areserved by BS 2. Strong interference is expectedbetween users 1 and 3, which may degrade theperformance of network NOMA, that is, multi-cell NOMA.

To mitigate the inter-cell interference, jointprecoding of NOMA users’ signals across neigh-boring cells can be considered. This requires thatall users’ data and CSI should be available atmultiple BSs, but finding the optimal precoder isnot trivial. Moreover, the multi-user precodingused for single-cell NOMA maybe not be feasi-ble for the network NOMA scenario, since theprecoder for geographically separate BS anten-nas does not actually form the physical beamthat can readily be used for intra-beam NOMA.Based on the fact that large-scale fading wouldbe quite different between the links to differentcells, a complexity-reduced precoding scheme fornetwork NOMA has been proposed in [4], wherethe multi-cell joint precoder is applied only tocell edge users (e.g., user 1 and user 3 as shownin Fig. 2c).

NOMA VIA CODE DOMAIN MULTIPLEXINGLow-Density Spreading CDMA: The ideabehind LDS-CDMA is to use sparse spreadingsequences instead of dense spreading sequencesin conventional CDMA [5] to reduce the inter-ference at each chip. Therefore, LDS-CDMAcan improve system performance by exploitingLDS sequences in CDMA [5], which is the keyfeature distinguishing conventional CDMA andLDS-CDMA. In this way, interference will beefficiently decreased among multiple users withappropriate spreading sequence design, andoverloading can be achieved.

At the receiver, a message passing algorithm(MPA) can be used to realize MUD. MPA isvery efficient for the factor graph [11], which is abipartite graph including variable nodes and fac-tor nodes as illustrated in Fig. 3. Messages arepassed among variable nodes and factor nodesover edges, which can be interpreted as the soft-values that represent the reliability of the symbolassociated with each edge. The marginal distri-bution of a variable node can be regarded as afunction of the messages received by that node[11].

Low-Density Spreading OFDM: LDS orthogo-nal frequency-division multiplexing (LDS-OFDM) can be considered as a combinedversion of LDS-CDMA and OFDM, in which

Figure 2. Illustration of NOMA via power domain multiplexing: a) basicNOMA with a SIC receiver; b) NOMA in MIMO systems; c) networkNOMA.

User 2

Frequency

Pow

er

User 2(a)

(b)

(c)

User 1

User 1

User 2

User 2

User 5

Beam 1

Beam 2User 3

User 3

User 4

User 4Cell 2

NOMA user pair NOMA user pair

Cell 1

BS 2BS 1

User 1

User 1

SIC of user 2signal

User 1 signaldetection

User 2 signaldetection

DAI_LAYOUT.qxp_Author Layout 9/3/15 5:41 PM Page 76

IEEE Communications Magazine • September 2015 77

the chips are subcarriers of OFDM in order tocombat the multipath fading. In LDS-OFDM,the transmitted symbols are first mapped to cer-tain LDS sequences, and then transmitted ondifferent OFDM subcarriers. The number ofsymbols can be greater than the number of sub-carriers, that is, overloading is allowed toimprove spectral efficiency [6]. MPA in LDS-CDMA can also be used in an LDS-OFDMreceiver. Essentially, LDS-OFDM can be viewedas an improved form of multi-carrier CDMA(MC-CDMA) by replacing the dense spreadingsequences with LDS.

Sparse Code Multiple Access: The recentlyproposed SCMA [7] is an enhanced version ofLDS-CDMA. Unlike LDS-CDMA, SCMAdirectly maps different bitstreams to differentsparse codewords, as illustrated in Fig. 4, whereeach user has a predefined codebook (there are6 users in Fig. 4). All codewords in the samecodebook contain zeros in the same two dimen-sions, and the positions of zeros in differentcodebooks are distinct to facilitate the collisionavoidance of any two users. For each user, twobits are mapped to a complex codeword. Code-words for all users are multiplexed over fourshared orthogonal resources (e.g., OFDM sub-carriers).

The key difference between LDS-CDMA andSCMA is that a multi-dimensional constellationfor SCMA is designed to generate codebooks,which brings the “shaping gain” that is not possi-ble for LDS [7]. Here, “shaping gain” is the gainin the average symbol energy when the shape ofa constellation is changed. In general, the shap-ing gain is higher when the shape of a constella-tion is closer to a sphere, and the maximumachievable shaping gain by the optimization of amulti-dimensional constellation is 1.53 dB [7].For the concatenated approach in high modula-tion order, the multi-dimensional constellationcan be optimized to obtain shaping gain, andthen codebooks are generated based on themulti-dimensional constellation [7]. The SCMAcodebook design is a complicated problem, sincedifferent layers are multiplexed with differentcodebooks. As the appropriate design criterionand specific solution to the multi-dimensionalproblem are still unknown, a multi-stageapproach has been proposed to realize a subop-timal solution [7]. Specifically, an N-dimensionalcomplex constellation with M points (which iscalled the mother constellation) is first opti-mized to improve the shaping gain, and thensome codebook-specific operations are per-formed to the mother constellation to generatethe N-dimensional constellation for each code-book. Three typical operations are phase rota-tion, complex conjugate, and dimensionalpermutation of the constellation [7]. In the gen-erated N-dimensional constellations after code-book-specific operations, each N-dimensionalconstellation point is multiplied with a projec-tion matrix to generate a K-dimensional code-word (K >> N), which has N non-zero elementsfrom the components of the N-dimensional con-stellation point. In this way, codebooks with Mcodewords can be obtained. Readers can findmore details in [7].

Multi-User Shared Access : In the uplinkMUSA system shown in Fig. 5 [8], symbols ofeach user are spread by a spreading sequence.Multiple spreading sequences constitute a poolfrom which each user can randomly pick one ofthe sequences. Note that for the same user, dif-ferent spreading sequences may also be used fordifferent symbols, which may further improvethe performance via interference averaging.Then all spreading symbols are transmitted overthe same time-frequency resources. The spread-ing sequences should have low cross-correlationand can be M-ary. At the receiver, codeword-level SIC is used to separate data from differentusers. The complexity of codeword-level SIC isless of an issue in the uplink as in any case thereceiver needs to decode the data for all users.The only noticeable impact on the receiverimplementation would be that the pipeline ofprocessing may be changed in order to performSIC operation. The difference between MUSAand MC-CDMA is that MUSA assumes that it isbasically synchronous when users’ signals arriveat the BS, which is easier to realize SIC, whileMC-CDMA dose not require this synchroniza-tion in the uplink. In addition, MUSA uses non-binary spreading sequences, while binary

Figure 3. Illustration of LDS-CDMA: six users and four chips for transmis-sion, which means 150 percent overloading.

User 1 Encoder

Variablenodes

Factornodes

Mapper

Mapper

Mapper

Mapper

Mapper

Encoder

Encoder

Encoder

Encoder

Encoder

User 2

User 3

User 4

User 5

User 6

x1sn,k 0

y1x2

y2x3

y3x4

y4x5

x6

Mapper

Figure 4. SCMA encoding and multiplexing.

Codebook 1

(0,0) (0,0)(1,0) (0,1) (1,1) (1,1)

Bit streamsare mapped

to sparsecodewords

MUDbased on

MPA

Six sparse codewordsare transmitted over

four orthogonal resources

Codebook 2Codebook 3 Codebook 4 Codebook 5Codebook 6

DAI_LAYOUT.qxp_Author Layout 9/3/15 5:41 PM Page 77

IEEE Communications Magazine • September 201578

spreading sequences are usually considered inclassical MC-CDMA systems.

In downlink MUSA, users are separated intoK groups. In each group, different users’ sym-bols are mapped to different constellations in away that can ensure Gray mapping in the com-bined constellation of superposed signals. Thecombined constellation is determined not onlyby the modulation order of each user, but alsoby the transmit power partition among multi-plexed users. Orthogonal sequences can beused to spread the superposed symbols to gettime or frequency diversity gain. Gray mappingof the combined constellation reduces thereliance on advanced receivers, so less process-ing-intensive receivers such as symbol-level SICcan be used.

OTHER NOMA SCHEMESIn addition to the power domain multiplexingand code domain multiplexing discussed above,a few other NOMA schemes are currently beinginvestigated. Pattern-division multiple access(PDMA) is a NOMA scheme that can be real-ized in several domains. At the transmitter,PDMA uses non-orthogonal patterns, which aredesigned by maximizing the diversity and mini-mizing the overlaps among multiple users. Thenthe actual multiplexing can be carried out in thecode domain, spatial domain, or a combinationof the two. Multiplexing in the code domain cor-responds to the case of successive interferencecancellation amenable multiple access (SAMA)[12], which is similar to LDS-CDMA, with LDSsequences being replaced by non-orthogonal pat-terns. Hence, MPA can also be used for thesequence detection in PDMA. The multiplexingin the spatial domain, called spatial PDMA,requires multiple antennas at the BS. The diver-sity of PDMA can come from multiple transmitantennas, which is preferred for macrocelldeployment. Different from multi-user MIMO,precoding is not needed in spatial PDMA sincethe aim is to increase the spatial diversity ratherthan spectral efficiency. PDMA can be used inboth downlink and uplink transmissions.

Bit-division multiplexing (BDM) [9] is anoth-er form of NOMA particularly useful for down-link transmission. Its basic concept is based onhierarchical modulation, and the resources ofmultiplexed users are partitioned at the bit level.Although strictly speaking the resource alloca-tion of BDM is orthogonal in the bit domain,multi-user signals share the same constellation(e.g., superposed in the modulation symboldomain).

Some other NOMA schemes were also pro-posed, such as interleave-division multiple access(IDMA), which performs interleaving of chipsafter symbols are multiplied by spreadingsequences. As shown in [13], compared toCDMA, IDMA is able to achieve an Eb/N0 gainof about 1 dB when bit error rate (BER) perfor-mance of 10–3 is considered in highly loaded sys-tems with 200 percent overloading.

In many of the NOMA schemes mentionedabove, especially when used for grant-free uplinktransmission, there is an issue that the users’activity or instantaneous system loading is notreadily known to the receiver. This would have anegative impact on the performance. Compressivesensing (CS) is a promising technique to estimatethe resource occupancy. Some work on CS-basedrandom access has been carried out recently, suchas compressive random access [14].

COMPARISON OF NOMA SOLUTIONSFrom the theoretical perspective, code-domainNOMA can obtain spreading gain due to the useof spreading sequences or codewords, which canbe achieved only in the case that there is no CSIat the transmitter. Spreading gain is similar tothat in CDMA, that is, the transmitted band-width can be spread by spreading sequences orcodewords, and thus, according to Shannon’sequation, signals can still be transmitted with thesame capacity even when signal-to-noise ratio(SNR) is low. The spreading gain can be calcu-lated by 10log N, where N is the spreading fac-tor. However, introducing redundancy throughspreading will affect the system spectral efficien-cy [15]. In addition, SCMA can achieve extra“shaping gain” due to the optimization of multi-dimensional constellation [7].

We also compare these NOMA schemes interms of the computational complexity of themulti-user signal detection algorithm. In power-domain NOMA, SIC is the key method formulti-user interference cancellation with com-plexity O(K3), where K is the number of users.Therefore, the complexity of SIC is much lessthan that of the optimal maximum likelihood(ML) detection, whose complexity O(|X|K)increases exponentially with the number of usersK, where |X| denotes the cardinality of the con-stellation set X . On the other hand, in code-domain NOMA like LDS-CDMA, LDS-OFDM,and SCMA, spreading sequences or codebooksshould be known at the receiver to realize MUD,and the complexity of the MPA-based receiver isO(|X|w), where w is the maximum number ofnonzero signals superimposed on each chip orsubcarrier. Thus, an MPA-based receiver usuallyhas higher complexity than a SIC-based receiveras w is usually larger than 3 in typical 5G sys-tems with massive connectivity.

Figure 5. Uplink MUSA system.

Each user’s symbols arespread by a speciallydesigned sequence

Spread symbolsare transmitted onsame orthogonal

resources

User 1

User 1

Spreading

Spreading

Spreading

User 2

User n User n

SIC

User 2

.

.

.

.

.

.

.

.

.

.

.

.

DAI_LAYOUT.qxp_Author Layout 9/3/15 5:41 PM Page 78

IEEE Communications Magazine • September 2015 79

CHALLENGES, OPPORTUNITIES, ANDFUTURE RESEARCH TRENDS

THEORETICAL ANALYSIS OFACHIEVABLE RATE AND OVERLOADING BOUNDS

In NOMA schemes, theoretical analysis isrequired to provide some insights for systemdesign. Achievable rate of multiple access is akey metric of system performance. The achiev-able rate of code-domain NOMA with LDSneeds to be studied, and can refer to the analyti-cal approach of MC-CDMA. Particularly, due tothe special structure of spreading sequences,some approximations can be used to simplify thecalculation. It is expected to derive the closed-form expression to reveal the relationshipbetween the achievable rate and LDS parame-ters such as sequence sparsity and overloadingfactor. Such theoretical results can shed light onhow to design the system parameters accordingto the specific application requirements.

On the other hand, the interference cancella-tion capability and the affordable complexity atthe receiver play an important role in the overallperformance, for example, the maximum over-loading factor that the system can support.

DESIGN OFSPREADING SEQUENCES OR CODEBOOKS

In LDS systems, due to non-orthogonal resourceallocation, interference exists among multipleusers. A factor graph in MPA should be opti-mized to get good trade-off between overloadingfactor and receiver complexity.

In addition, it has been proved that MPA canobtain the exact marginal distribution with acycle-free factor graph and the precise solutionwith a local tree-like factor graph. Graph theorycan be used to design a cycle-free or local tree-like factor graph in NOMA without compromis-ing spectral efficiency. In addition, the matrixdesign principle and methods in low-density-par-ity check (LDPC) can be considered whendesigning the factor graph for NOMA.

RECEIVER DESIGNFor an MPA-based receiver, the complexity maystill be high for massive connectivity in 5G.Therefore, simplified improvement of MPA canbe used to reduce receiver complexity, such asGaussian approximation of interference (GAI),which models the interference-plus-noise asGaussian distributed, and such approximationtends to be more accurate as the amount of con-nectivity becomes larger in 5G. In addition,MPA can be used to jointly detect and decodethe received symbols, in which the constructedgraph consists of variable nodes, observationnodes, and check nodes corresponding to thecheck equations of the LDPC code. In this way,intrinsic information between the decoder andthe demodulator can be used more efficiently toimprove the detector’s performance.

For a SIC-based receiver, error propagation maydegrade the performance of some users. Therefore,at each stage of SIC, some nonlinear detection algo-rithms with higher detection accuracy can be consid-ered to suppress the error propagation.

OTHER CHALLENGES

There are also some other engineering aspectsof NOMA, including reference signal design,channel estimation, and CSI feedback mecha-nism that can deliver robust performance whencross-user interference is severe, resource alloca-tion signaling that can support different trans-mission modes for NOMA, extension to MIMO(especially massive MIMO) that can reap theperformance benefits of both NOMA and multi-user MIMO, peak-to-average-power ratio(PAPR) reduction in multi-carrier NOMA, sys-tem scalability that can support different trafficloading and radio environment, and so on. Thesechallenges need to be addressed before NOMAbecomes part of 5G standards in the future.

THE CONCEPT OF SODEMAAs discussed above, NOMA can be used forcapacity improvement and massive connectionsin 5G. However, this does not mean that con-ventional OMA schemes will be completelyreplaced by NOMA in future 5G networks. Forexample, when the number of users is small andthe near-far effect is not significant, such as inthe case of small cells, OMA would be a betterchoice. In this sense, both OMA and NOMAwill coexist in 5G to fulfill diverse requirementsof different services and applications.

To this end, we borrow the idea of softwaredefined radio (SDR) for multiple access designto propose the SoDeMA concept for 5G asshown in Fig. 6, where different NOMA schemescan coexist in a system assuming all of them willbe specified in 5G standards. SoDeMA providesa very flexible configuration of multiple accessschemes to support different services and appli-cations in 5G. For example, for cell-center usersor real-time services like ultra-high-definitionvideo, conventional OMA schemes can be adopt-ed to support high data rate transmission, whichcapitalizes on the orthogonality and synchroniza-tion. On the other hand, when high spectral effi-ciency, massive connectivity, and frequent accessof small packets are required in some practicalscenarios (e.g., dense population areas andmobile social applications), NOMA schemes canbe selected. Moreover, different NOMA orOMA schemes have their own appropriate appli-cation situations, and can be adaptively config-ured to realize the trade-off between

Figure 6. Illustration of the concept of software definded multiple access.

CDMA

MUSA

TDMA

OFD

MA

PDMA

Power-domain NOMA

tComplexity

fPerform

ance

LDS-OFDM

SCMA . . . . . .

DAI_LAYOUT.qxp_Author Layout 9/3/15 5:41 PM Page 79

IEEE Communications Magazine • September 201580

performance and implementation complexity.For instance, if a large difference among users’channel conditions exists due to the near-fareffect or in moving networks, power-domainNOMA with a SIC receiver can be used with rel-atively low complexity. On the other hand, ifhigh reliability should be guaranteed, especiallywhen channel condition is bad or the locationdistribution of users is concentrated, SCMA is afeasible solution due to its shaping gain andnear-optimal MPA detection. Of cause, whenthe number of users is large enough, it may bedifficult to design a codebook for each user, andin this case, LDS-OFDM or MUSA can also beused to reduce the design complexity at thetransmitter or receiver, separately.

As elaborated in previous sections, certainsignal processing modules are common to sever-al NOMA schemes, for example, MPA at thereceiver or spreading operation at the transmit-ter, which can be shared in hardware so that thehardware cost would be reduced at both userterminals and base stations. These general-pur-pose modules can be combined in differentforms at the software level to implement differ-ent schemes. The switching between NOMAschemes is fast and flexible with software definedhardware architecture, and can quickly adapt todifferent deployment scenarios, that is, fromcapacity achieving to user loading improvement.

To enable SoDeMA, the frame structureshould be flexible enough so that the time andfrequency resources are partitioned into differ-ent blocks freely for different services and users.In each resource block, one specific multipleaccess scheme is configured with specific wave-form, duplex mode, pilot signals, power level,and so on. Note that the inter-subcarrier inter-ference between different resource blocks needsto be carefully mitigated. The proposed SoDe-MA concept provides a flexible configuration ofmultiple access schemes to support different ser-vices and applications. It is highly expected thatSoDeMA can be carefully designed to adapt tovarious application scenarios to support the sys-tem design goal of “anything as a service” infuture 5G networks.

CONCLUSIONSIn this article, we have discussed and comparedseveral major NOMA schemes for 5G from theaspects of basic principles, key features, receivercomplexity, engineering feasibility, and so on.Compared to conventional OMA, NOMA allowscontrollable interferences to realize overloadingat the cost of a tolerable increase of receivercomplexity. Therefore, the demands of spectralefficiency and massive connectivity for 5G canbe partially fulfilled by NOMA. We have alsohighlighted key challenges, opportunities andfuture research tends for the design of NOMA,including theoretical work, optimal design ofspreading sequences or codebooks, receiverdesign, a grant-free NOMA mechanism, and soon. The proposed concept of SoDeMA is able toflexibly support diverse services and applicationswith different requirements. It is expected thatNOMA will play an important role in future 5Gwireless communications.

REFERENCES[1] F. Boccardi et al., “Five Disruptive Technology Directions

for 5G,” IEEE Commun. Mag., vol. 52, no. 2, Feb. 2014,pp. 74–80.

[2] Y. Saito et al., “Non-Orthogonal Multip le Access(NOMA) for Future Radio Access,” Proc. IEEE VTC-Spring ’13, June 2013, pp. 1–5.

[3] K. Higuchi and Y. Kishiyama, “Non-Orthogonal Accesswith Random Beamforming and Intra-Beam SIC for Cel-lular MIMO Downlink,” Proc. IEEE VTC-Fall ’13, Sept.2013, pp. 1–5.

[4] S. Han et al., “Energy Efficiency and Spectrum EfficiencyCo-Design: From NOMA to Network NOMA,” IEEEMMTC E-Letter, vol. 9, no. 5, Sept. 2014, pp. 21–24.

[5] R. Hoshyar, F. P. Wathan, and R. Tafazolli, “Novel Low-Density Signature for Synchronous CDMA Systems overAWGN Channel,” IEEE Trans. Signal Proc., vol. 56, no.4, Apr. 2008, pp. 1616–26.

[6] M. Al-Imari et al., “Uplink Nonorthogonal MultipleAccess for 5G Wireless Networks,” Proc. 11th Int’l.Symp. Wireless Commun. Sys., Aug. 2014, pp. 781–85.

[7] H. Nikopour and H. Baligh, “Sparse Code Multiple Access,”Proc. IEEE PIMRC 2013, Sept. 2013, pp. 332–36.

[8] Z. Yuan, G. Yu, and W. Li, “Multi-User Shared Accessfor 5G,” Telecommun. Network Technology, vol. 5, no.5, May 2015, pp. 28–30.

[9] J. Huang et al., “Scalable Video Broadcasting Using BitDivision Multiplexing,” IEEE Trans. Broadcast., vol. 60,no. 4, Dec. 2014, pp. 701–06.

[10] D. Tse and P. Viswanath, Fundamentals of WirelessCommunication, Cambridge Univ. Press, 2005.

[11] F. R. Kschischang, B. J. Frey, and H.-A. Loeliger, “FactorGraphs and the Sum-Product Algorithm,” IEEE Trans. Info.Theory, vol. 47, no. 2, Feb. 2001 , pp. 498–519.

[12] X. Dai et al., “Successive Interference CancelationAmenable Multiple Access (SAMA) for Future WirelessCommunications,” Proc. IEEE ICCS 2014, Nov. 2014,pp. 1–5.

[13] K. Kusume, G. Bauch, and W. Utschick, “IDMA vs.CDMA: Analysis and Comparison of Two Multip leAccess Schemes,” IEEE Trans. Wireless Commun., vol.11, no. 1, pp. 78–87, Jan. 2012.

[14] G. Wunder, P. Jung, and C. Wang, “Compressive Ran-dom Access for Post-LTE Systems,” Proc. IEEE ICC ’14,June 2014, pp. 539–44.

[15] V. V. Veeravalli and A. Mantravadi, “The Coding-Spreading Tradeoff in CDMA Systems,” IEEE JSAC, vol.20, no. 2, Feb. 2002, pp. 396–408.

BIOGRAPHIESLINGLONG DAI [M’11, SM’14] (daill@tsinghua.edu.cn)received his B.S. degree from Zhejiang University in 2003,his M.S. degree (with highest honor) from the ChinaAcademy of Telecommunications Technology (CATT) in2006, and his Ph.D. degree (with the highest honor) fromTsinghua University, Beijing, China, in 2011. From 2011 to2013, he was a postdoctoral fellow with the Departmentof Electronic Engineering, Tsinghua University, where hehas been an assistant professor since July 2013. Hisresearch interests are in wireless communications, with afocus on multi-carrier techniques, multi-antenna tech-niques, and multi-user techniques. He has published over60 journal and conference papers. He has received the Out-standing Ph.D. Graduate of Tsinghua University award in2011, the Excellent Doctoral Dissertation of Beijing awardin 2012, the IEEE ICC Best Paper Award in 2013, theNational Excellent Doctoral Dissertation Nomination Awardin 2013, the IEEE ICC Best Paper Award in 2014, the URSIYoung Scientists Award in 2014, and the IEEE Scott HeltMemorial Award in 2015 (IEEE Transactions on Broadcast-ing Best Paper Award). He currently serves as Co-Chair ofthe IEEE Special Interest Group (SIG) on Signal ProcessingTechniques in 5G Communication Systems.

BICHAI WANG [S’15] (wang-bc11@mails.tsinghua.edu.cn)received her B.S. degree in electronic engineering fromTsinghua University in 2015. She is currently workingtoward her Ph.D. degree in the Department of ElectronicEngineering, Tsinghua University. Her research interests arein wireless communications, with emphasis on new multi-ple access techniques. She received the Freshman Scholar-ship of Tsinghua University in 2011, Academic MeritScholarships of Tsinghua University in 2012, 2013, and2014, respectively, and the Excellent Thesis Award ofTsinghua University in 2015.

Compared with con-

ventional OMA,

NOMA allows con-

trollable interferences

to realize overloading

at the cost of toler-

ate increase of

receiver complexity.

Therefore, the

demands of spectral

efficiency and mas-

sive connectivity for

5G can be partially

fulfilled by NOMA.

DAI_LAYOUT.qxp_Author Layout 9/3/15 5:41 PM Page 80

IEEE Communications Magazine • September 2015 81

YIFEI YUAN (yifei.yuan@ztetx.com) received Bachelor’s andMaster’s degrees from Tsinghua University, and a Ph.D.from Carnegie Mellon University, Pennsylvania He was withAlcatel-Lucent from 2000 to 2008 working on 3G/4G keytechnologies. Since 2008, he has been with ZTE, responsi-ble for standards research on LTE-Advanced physical layerand 5G technologies. His research interests include MIMO,iterative codes, resource scheduling, non-orthogonalaccess, and small cells. He was admitted to the ThousandTalent Plan Program of China in 2010. He has publishedextensively, including a book on LTE-A relay and a book onLTE-Advanced key technologies. He has over 30 grantedpatents.

SHUANGFENG HAN (hanshuangfeng@chinamobile.com)received his M.S. and Ph.D. degrees in electrical engineer-ing from Tsinghua University in 2002 and 2006, respec-tively. He joined Samsung Electronics as a senior engineerin 2006 working on MIMO, multi-BS MIMO, and so on.Since 2012, he has been a senior project manager in theGreen Communication Research Center at the ChinaMobile Research Institute. His research interests are green5G, massive MIMO, full duplex, NOMA, and EE-SE co-design.

CHIH-LIN I (icl@chinamobile.com) received her Ph.D. degreein electrical engineering from Stanford University. She hasbeen working at multiple world-class companies and re-search institutes leading R&D, including AT&T Bell Labs,AT&T HQ, ITRI of Taiwan, and ASTRI of Hong Kong. Shereceived the IEEE Transactions on Communications StephenRice Best Paper Award and is a winner of the CCCP Nation-al 1000 Talent program. Currently, she is China Mobile’schief scientist of wireless technologies and has establishedthe Green Communications Research Center, spearheadingmajor initiatives including system architectures, technolo-gies, and devices; green energy; and C-RAN and soft base

stations. She was an elected Board Member of IEEE Com-Soc, Chair of the ComSoc Meetings and ConferencesBoard, and Founding Chair of the IEEE WCNC SteeringCommittee. She is currently an Executive Board Member ofGreenTouch and a Network Operator Council Member ofETSI NFV. Her research interests are green communications,C-RAN, network convergence, and bandwidth active anten-na arrays.

ZHAOCHENG WANG [M’09, SM’11] (zcwang@tsinghua.edu.cn)received his B.S., M.S., and Ph.D. degrees from TsinghuaUniversity in 1991, 1993, and 1996, respectively. From1996 to 1997, he was a postdoctoral fellow of NanyangTechnological University, Singapore. From 1997 to 1999,he was with OKI Techno Centre (Singapore) Pte. Ltd.,where he was first a research engineer and later became asenior engineer. From 1999 to 2009, he was with SonyDeutschland GmbH, where he was first a senior engineerand later became a principal engineer. He is currently aprofessor with the Department of Electronic Engineering,Tsinghua University, and serves as director of the Broad-band Communication Key Laboratory, Tsinghua NationalLaboratory for Information Science and Technology. He hasauthored or coauthored over 80 international journalpapers (SCI indexed). He is the holder of 34 grantedU.S./EU patents. He co-authored two books, one of which,Millimeter Wave Communication Systems, was selected forthe IEEE Series on Digital & Mobile Communication andpublished by Wiley-IEEE Press. His research areas includewireless communications, visible light communications, mil-limeter-wave communications, and digital broadcasting. Heis a Fellow of the Institution of Engineering and Technolo-gy. Currently he serves as an Associate Editor of IEEE Trans-action on Wireless Communications and IEEECommunications Letters, and has also served as TechnicalProgram Committee Co-Chair of various international con-ferences.

DAI_LAYOUT.qxp_Author Layout 9/3/15 5:41 PM Page 81