You are on page 1of 5

Comparison Study of Non-Orthogonal Multiple

Access Schemes for 5G


Bichai Wang1 , Kun Wang2 , Zhaohua Lu3 , Tian Xie1 , and Jinguo Quan4
1
Tsinghua National Laboratory for Information Science and Technology (TNList),
Department of Electronic Engineering, Tsinghua University, Beijing 100084, China
2
National Computer Network Emergency Response Technical Team/Coordination Center (CNCERT), Beijing, China
3
ZTE Cooperation, Shenzhen, China
4
Division of Information Science & Technology,
Shenzhen Graduate School, Tsinghua University, Shenzhen 518055, China
E-mail: 18810655172@163.com

Abstract—With the development of mobile Internet and Inter- division multiple access (OFDMA) [12], NOMA introduces
net of things (IoT), the 5th generation (5G) wireless communica- some controllable interferences to realize overloading at the
tions will foresee explosive increase in mobile traffic. To address cost of slightly increased receiver complexity. As a result,
challenges in 5G such as higher spectral efficiency, massive
connectivity, and lower latency, some non-orthogonal multiple higher spectral efficiency and massive connectivity can be
access (NOMA) schemes have been recently actively investigated, achieved by NOMA for 5G.
including power-domain NOMA, multiple access with low-density Among available NOMA technologies, SCMA, MUSA, and
spreading (LDS), sparse code multiple access (SCMA), multi- PDMA are three typical schemes recently proposed. In SCMA,
user shared access (MUSA), pattern division multiple access
(PDMA), etc. Different from conventional orthogonal multiple bit streams are directly mapped to sparse codewords, and
access (OMA) schemes, NOMA can realize overloading by thus it is available to use near-optimal multi-user detection
introducing some controllable interferences at the cost of slightly (MUD) based on message passing algorithm (MPA) [13] with
increased receiver complexity, which can achieve significant gains acceptable complexity. MUSA uses low-correlation spreading
in spectral efficiency and accommodate much more users. In sequences at the transmitter, and successive interference can-
this paper, we will discuss basic principles and key features of
three typical NOMA schemes, i.e., SCMA, MUSA, and PDMA. cellation (SIC) [14] is performed at the receiver to realize
What’s more, their performance in terms of uplink bit error rate MUD. While in PDMA, non-orthogonal patterns are designed
(BER) will be compared. Simulation results show that in typical to maximize the diversity and minimize the overlaps of
Rayleigh fading channels, SCMA has the best performance, while multiple users. All of them can realize overloading by non-
the BER performance of MUSA and PDMA are very close to each orthogonal resource allocation, which can improve spectral
other. In addition, we also analyze the performance of PDMA
using the same factor graph as SCMA, which indicates that the efficiency and accommodate much more users.
performance gain of SCMA over PDMA comes from both the However, different NOMA schemes are individually dis-
difference of factor graph and the codebook optimization. cussed in the literature, and no work has been published to
compare their performance. This paper aims to fill in the
I. I NTRODUCTION
gap by comparing the performance of three typical NOMA
In the history of broadband multimedia communications schemes under hot investigation: SCMA, MUSA, and PDMA.
and broadcasting [1], lots of investigations have been done Particularly, Section II will present the basic principles and key
to satisfy various requirements in practical systems. However, features of SCMA, MUSA, and PDMA, respectively. Then,
with the rapid development of mobile Internet and Internet of in Section III, we will compare the performance of these
things (IoT), rapidly increased requirements are expected to NOMA schemes in typical Rayleigh fading channels. Sim-
be satisfied in the 5th generation (5G) wireless communica- ulation results show that SCMA performs best, while MUSA
tions [2] [3], e.g., higher spectral efficiency, massive connec- and PDMA have very similar performance. Meanwhile, we
tivity, lower latency, etc. To address these challenges, some also show that SCMA outperforms PDMA not only due to the
non-orthogonal multiple access (NOMA) schemes have been difference of factor graph, but also due to the optimal design
recently proposed, such as power-domain NOMA [4], multiple of codebooks. Finally, conclusions are drawn in Section IV.
access with low-density spreading (LDS) [5] [6], sparse code
multiple access (SCMA) [7]–[9], multi-user shared access
(MUSA), pattern division multiple access (PDMA) [10], and II. T YPICAL NOMA S CHEMES
so on. Different from conventional orthogonal multiple access
(OMA) schemes, including frequency division multiple access In this section, basic principles and key features of three
(FDMA) [11], time division multiple access (TDMA), code NOMA schemes under hot investigation, i.e., SCMA, MUSA,
division multiple access (CDMA), and orthogonal frequency and PDMA, will be discussed separately.
[ [ [ [ [ [
A. SCMA
9DULDEOH
SCMA is an improved scheme based on LDS-OFDM that QRGHV
exploits sparse spreading sequences to realize overloading. Un-
like LDS-OFDM, bit-to-constellation mapping and spreading
with low-density signatures are combined together in SCMA,
namely, the bit streams are directly mapped to different sparse )DFWRU
codewords. Particularly, different users’ codewords are taken QRGHV
from different codebooks, which are generated by multi- \ \ \ \
dimensional constellation. By making full use of the sparsity
of codewords, at the receiver, MUD based on MPA can be
Fig. 1. Factor graph representation of MPA.
used to separate symbols with acceptable complexity. In this
subsection, system model of SCMA will be firstly discussed,
and then MUD based on MPA will be presented. where N (k) expresses the set of users whose sparse codeword
1) System model: In the uplink SCMA, different codebooks has a nonzero element at subcarrier k, namely, N (k) =
are assigned to different users. We assume that the number {j|xkj = 0}.
of codebooks or users is J, each codebook contains M 2) MUD based on MPA: MPA is the typical receiver
codewords of length K, the number of nonzero elements in algorithm of SCMA. The procedure of MPA can be explained
each codeword is N and N  K, i.e., the codewords are by the factor graph as shown in Fig. 1, which is a bipartite
sparse. We consider the case of K < J, namely, overloading graph including variable nodes and factor nodes. Particularly,
can be realized by SCMA, which accommodates the demand in SCMA, variable nodes consist of transmitted codewords
of massive connectivity in 5G. At the transmitter, log2 M bits from J users, factor nodes can be seen as the received signals
for user j are mapped to a sparse codeword from the jth over K subcarriers, and there exists an edge between a variable
codebook, and then J codewords are multiplexed over K node xj and a factor node yk if and only if xkj = 0.
shared orthogonal resources, e.g., OFDM orthogonal subcar- Message can be passed among variable nodes and factor nodes
riers [15]–[17]. The received signal on subcarrier k can be through the edges, and the marginal distribution computed for
represented by a variable node can be seen as a function of the messages
received by that node. Without loss of generality, we assume
J
 that the transmitted codewords and noise vector are identically
yk = hkj xkj + nk , (1) independently distributed (i.i.d) as well as the transmitted
j=1
codewords obey uniform distribution, then the iterative form
where xkj is the kth component of the codeword xj for user of MUD based on MPA can be represented as
j, hkj is the channel gain of user j at the kth subcarrier, (t)
 1 1
and nk is a complex-valued noise sample taken from a zero mk→j (xj ) ∝ √ exp{− 2 ||yk
2πσ 2σ
mean Gaussian distribution with variance σ 2 . We combine the {xi |i∈N (k)\j}
  (t−1)
received signals from all subcarriers, and then the received − hkj xkj ||2 } mi→k (xi ) ,
signal vector y = (y1 , y2 , · · · , yK )T can be described as i∈N (k)\j
j
J
 xkj = 0
y= diag(hj )xj + n, (2)  (4)
(t) (t−1)
j=1 mj→k (xj ) ∝ mi→j (xj ) , (5)
i∈N (j)\k
where xj = (x1j , x2j , · · · , xKj )T is the SCMA codeword of
user j, hj = (h1j , h2j , · · · , hKj )T is the channel vector for (t)
where mk→j (xj ) denotes the message transmitted from factor
the jth user, and n ∼ CN (0, σ 2 IK ) is the Gaussian noise. (t)
node yk to variable node xj at the tth iteration, mj→k (xj )
By multiplexing J users over K orthogonal resources, the
presents the message transmitted from variable node xj to
overloading factor is defined as λ = J/K.
factor node yk , and N (k)\j presents all elements in N (k)
We can see from (1) that when all xj for j = 1, · · · , J
except for j. If the maximum iteration number is T , the
are sparse codewords, each user will spread its data only
marginal probability distribution for each codeword can be
over a small number of subcarriers. As a result, the number
finally calculated by
of the superimposed signals at each subcarrier will be less  (T )
than the number of active users, which means the interference p (xj ) ∝ mk→j (xj ) . (6)
efficiently decreases among multiple users with appropriate k∈N (j)
design of sparse codewords. Then, (1) can be also rewritten
as B. MUSA

yk = hkj xkj + nk , (3) MUSA is another NOMA scheme via code domain mul-
j∈N (k) tiplexing, which can be regarded as an improved scheme of
[Ö()
CDMA. Different from conventional CDMA, MUSA can real- 006(IRUXVHU 
ize overloading by using low-correlation spreading sequences
at the transmitter. At the receiver, SIC is performed to separate
\Ö () [Ö(  )
superimposed symbols according to the received signal-to- \Ö () = \ − K () [Ö () 006(IRUXVHU 
interference-plus-noise-power ratio (SINR) difference. In this
subsection, the system model and SIC receiver of MUSA will
\Ö (  ) [Ö()
be discussed in detail. \Ö ( ) = \Ö () − K( ) [Ö( ) 006(IRUXVHU 
1) System model: In the uplink MUSA system, all transmit-
ted symbols of one user are multiplied with the same spreading
sequence. Of course, different spreading sequences can be \Ö ( ) [Ö(  )
\Ö ( ) = \Ö (  ) − K ( ) [Ö( ) 006(IRUXVHU 
also used to different symbols for the same user, which can




further obtain the benefit of interference averaging. Then, all
symbols after spreading are transmitted over the same time-
frequency resources, e.g., subcarriers of OFDM. Without loss
Fig. 2. SIC receiver based on MMSE.
of generality, we assume that one user just transmits a symbol
every time, and there are K1 users and N1 subcarriers. Similar
to SCMA, overloading can be also realized by MUSA, i.e., where H can be obtained via training pilots [18]–[20]. Then,
K1 > N1 , which will introduce interferences among multiple the SIC detection method as shown in Fig. 2 can be summa-
users. The received signal on subcarrier n can be represented rized as follows.
by Firstly, we calculate the SINR for each user. Users are then
K1 sorted by the order of decreasing SINR, i.e., user with the
yn = gn,k sn,k xk + vn , (7) maximum SINR is detected at first by using MMSE, which is
k=1 x̂(1) = w(1) y, where w(1) is the first row of WM M SE after
where xk is the transmitted symbol for user k, sn,k is the ordering by SINR.
nth component of the spreading sequence sk of user k, gn,k Secondly, the estimated interference from the first user is
is the channel gain of user k at the nth subcarrier, and subtracted from the received signal y, i.e., ŷ(1) = y−h(1) x̂(1) .
vn is a complex-valued noise sample taken from a zero The channel vector h(1) of the first user is removed from
mean Gaussian distribution with variance σ 2 . We combine the the channel matrix H. Then SINRs for remained users are
received signals from all subcarriers, and then the received calculated again and the detection order is based on their
signal vector y = (y1 , y2 , · · · , yN1 )T can be modeled as received SINRs. MMSE is used to detect the second user’s
transmitted symbol. Other users’ symbols are detected in the
y = Hx + v, (8) same way, namely, interference cancellation, ordering by SINR
where, x = (x1 , x2 , · · · , xK1 )T , H is the channel matrix and linear detection using MMSE are performed at each stage.
of size N1 × K1 , and the element hn,k in the nth raw
C. PDMA
and the kth column of H is equal to gn,k sn,k . Finally,
v = (v1 , v2 , · · · , vK1 )T , and v ∼ CN 0, σ 2 IN1 . PDMA is a novel NOMA scheme that can be realized
At the receiver, SIC is performed to separate superimposed in multiple domains. At the transmitter, PDMA uses non-
symbols according to the received SINR difference. Note orthogonal patterns which are designed to maximize the di-
that the design principle of spreading sequences in MUSA versity and minimize the overlaps of multiple users. Then,
is to maintain low correlation, which can be generated by multiplexing can be realized in code domain, power domain,
pseudorandom sequences, and thus interference cancellation space domain or their combinations. Multiplexing in code
capability can be improved at the receiver. domain is similar to that in SCMA, but the number of
2) SIC receiver: In MUSA, SIC is used to realize MUD subcarriers connected to the same symbols in the factor graph
at the receiver. MUD based on SIC performs interference can be different. At the receiver, MPA is performed to realize
cancellation in multiple stages. Linear detection is used to interference cancellation. In the case of multiplexing in power
estimate the transmitted symbol at each stage. Particularly, we domain, power allocation needs to be carefully considered
use minimum mean square error (MMSE) in this paper. Given under the total power constraint. SIC can be also used at the re-
the system model (8), the MMSE detector finds the transfor- ceiver according to SINR difference among multiplexed users.
mation matrix by minimizing the mean square error between Multiplexing in space domain, i.e., spatial PDMA, can be
the transmitted vector and the estimated vector x̂ =WM M SE y combined with the multi-antenna technique. The advantage of
as follows: spatial PDMA compared with multi-user MIMO (MU-MIMO)
is that PDMA doesn’t require joint precoding to realize spatial
WM M SE = min E[||x − Wy||2 ], (9) orthogonality, which significantly reduces system complexity.
W
In addition, multiple domains can be combined in PDMA to
whose solution is
 −1 H make full use of various available wireless resources. Without
WM M SE = HH H + σ 2 I H , (10) loss of generality, the performance of PDMA multiplexed in
code domain will be considered in this paper. 10
0

Consider an uplink code domain PDMA system with K2


users and N2 orthogonal subcarriers of OFDM. We assume
that K2 > N2 , i.e., the overloaded system. The system 10
í1

model of code domain PDMA is similar to that of MUSA,


but in code domain PDMA, the nonzero elements of any
spreading sequence bk for user k are equal to 1, and the

BER
í2
10
spreading matrix B = (b1 , b2 , · · · , bK2 ), which is similar
to S = (s1 , s2 , · · · , sK1 ) in MUSA, is designed with the
following principles [10]: í3 MUSA
10
• The number of groups with different number of 1’s in the PDMA
PDMA(Same factor graph with SCMA)
spreading sequence should be maximized.
SCMA
• The number of the overlapped spreading sequences which í4
10
have the same number of 1’s should be minimized. 0 5 10 15 20
SNR
Then the maximum number of supported users with N2
orthogonal subcarriers can be calculated by
      Fig. 3. BER performance comparison of SCMA, MUSA, and PDMA in
N2 N2 N2 Rayleigh fading channels.
+ + ··· + = 2N2 − 1. (11)
1 2 N2
from {−1, 0, 1}, and the design of non-orthogonal patterns
For example, spreading matrices for N2 = 2, K2 = 3, N2 = in PDMA refers to the principle in [10].
3, K2 = 7, and N2 = 4, K2 = 15 can be designed as follows: We can observe from Fig. 3 that SCMA has the best BER
  performance, while MUSA and PDMA have very similar BER
1 1 0
B2,3 = , (12) performance. This is caused by the fact that error propagation
1 0 1
⎛ ⎞ of SIC receivers has a severe effect on the system performance.
1 1 0 1 1 0 0 On the other hand, when PDMA adopts the same factor graph
B3,7 = ⎝ 1 1 1 0 0 1 0 ⎠ , (13) as SCMA, the performance of SCMA is still better than that
1 0 1 1 0 0 1 of PDMA, which indicates that the near-optimal design of
sparse codewords in SCMA is beneficial for the performance
B4,15
⎛ ⎞ improvement.
1 1 0 1 1 1 1 1 0 0 0 1 0 00
⎜ 1 1 1 0 1 1 0 0 1 1 0 0 1 00 ⎟ .
=⎜
⎝ 1
⎟ IV. C ONCLUSIONS
1 1 1 0 0 1 0 1 0 1 0 0 10 ⎠
1 0 1 1 1 0 0 1 0 1 1 0 0 01 In this paper, we compare the performance of three typical
(14) NOMA schemes, i.e., SCMA, MUSA, and PDMA. It is found
At the receiver, MPA is performed to separate signals that SCMA has the best performance due to the near-optimal
for different users. The purpose of the design principle for design of sparse codewords together with the near-optimal
spreading matrices in code domain PDMA is to facilitate MPA receiver, while MUSA and PDMA have very similar
interference cancellation. Take the B4,15 for example, the performance. This comparison study reveals that to obtain
spreading sequence of the first user has 4 non-zero elements, better system performance for NOMA, the design of sparse
i.e., the diversity order is 4, and thus the first user’s symbol codebooks in SCMA, low-correlation spreading sequences in
is the most reliable among those of all other users. Therefore, MUSA, and non-orthogonal patterns in PDMA should be
the probability of the first user’s symbol can be quickly con- optimized in the future work. Additionally, robust receiver
verged, which will be benefit for the convergence of symbols’ with low complexity is also expected for NOMA.
probability of other users with lower diversity orders.
ACKNOWLEDGEMENT
III. S IMULATION R ESULTS
This work was supported by Shenzhen Visible
The uplink bit error rate (BER) performance comparison Light Communication System Key Laboratory
of SCMA, MUSA, and PDMA in Rayleigh fading channels (ZDSYS20140512114229398) and Shenzhen Peacock
is shown in Fig. 3. In addition, the performance of PDMA Plan (Grant No. 1108170036003286).
using the same factor graph with SCMA is also analyzed.
Particularly, QPSK modulation is used in this paper. The R EFERENCES
number of transmitted symbols is 6, and the number of [1] L. Dai, Z. Wang, and Z. Yang, “Next-generation digital television
orthogonal resources is 4. Therefore, the overloading factor terrestrial broadcasting systems: key technologies and research trends,”
is 150%. The codebooks in SCMA are designed accord- IEEE Commun. Mag., vol. 50, no. 6, pp. 150–158, June 2012.
[2] F. Boccardi, R. W. Heath Jr, A. Lozano, T. L. Marzetta, and P. Popovski,
ing to [7], spreading sequences in MUSA are generated by “Five disruptive technology directions for 5G,” IEEE Commun. Mag.,
pseudorandom sequences whose real and image are taken vol. 52, no. 2, pp. 74–80, Feb. 2014.
[3] A. Osseiran, F. Boccardi, V. Braun, K. Kusume, P. Marsch, M. Maternia, [12] L. Dai, Z. Wang, and J. Song, “TDS-OFDMA: a novel multiple access
O. Queseth, M. Schellmann, H. Schotten, H. Taoka, et al., “Scenarios system based on TDS-OFDM,” IEEE Trans. Consum. Electron., vol. 57,
for 5G mobile and wireless communications: the vision of the METIS no. 4, pp. 1528–1534, Nov. 2011.
project,” IEEE Commun. Mag., vol. 52, no. 5, pp. 26–35, May 2014. [13] F. R. Kschischang, B. J. Frey, and H.-A. Loeliger, “Factor graphs and the
[4] Y. Saito, Y. Kishiyama, A. Benjebbour, T. Nakamura, A. Li, and sum-product algorithm,” Trans. Inf. Theory, vol. 47, no. 2, pp. 498–519,
K. Higuchi, “Non-orthogonal multiple access (NOMA) for future radio 2001.
access,” in Proc. IEEE VTC Spring 2013, June 2013, pp. 1–5. [14] Y. S. Cho, J. Kim, W. Y. Yang, and C. G. Kang, MIMO-OFDM wireless
[5] J. Van De Beek and B. M. Popovic, “Multiple access with low-density communications with MATLAB. John Wiley & Sons, 2010.
signatures,” in Proc. IEEE GLOBECOM 2009, Dec. 2009, pp. 1–6. [15] L. Dai, Z. Wang, and Z. Yang, “Compressive sensing based time domain
[6] M. Al-Imari, P. Xiao, M. A. Imran, and R. Tafazolli, “Uplink non- synchronous OFDM transmission for vehicular communications,” IEEE
orthogonal multiple access for 5G wireless networks,” in Proc. 11th In- J. Sel. Areas Commun., vol. 31, no. 9, pp. 460–469, Sep. 2013.
ternational Symposium on Wireless Communications Systems (ISWCS), [16] L. Dai, J. Fu, J. Wang, and J. Song, “A multi-user uplink TDS-OFDM
Aug. 2014, pp. 781–785. system based on dual PN sequence padding,” IEEE Trans. Consum.
[7] H. Nikopour and H. Baligh, “Sparse code multiple access,” in Proc. Electron., vol. 55, no. 3, pp. 1098–1106, Aug. 2009.
IEEE PIMRC 2013, Sep. 2013, pp. 332–336. [17] L. Dai, J. Wang, Z. Wang, P. Tsiaflakis, and M. Moonen, “Spectrum-and
[8] K. Au, L. Zhang, H. Nikopour, E. Yi, A. Bayesteh, U. Vilaipornsawai, energy-efficient OFDM based on simultaneous multi-channel reconstruc-
J. Ma, and P. Zhu, “Uplink contention based SCMA for 5G radio access,” tion,” IEEE Trans. Signal Process., vol. 61, no. 23, pp. 6047–6059, Dec.
in Proc. IEEE GLOBECOM 2014, Dec. 2014, pp. 1–5. 2013.
[9] M. Taherzadeh, H. Nikopour, A. Bayesteh, and H. Baligh, “SCMA [18] L. Dai, Z. Wang, and Z. Yang, “Spectrally efficient time-frequency
codebook design,” in Proc. IEEE VTC Fall 2014, Sep. 2014, pp. 1– training OFDM for mobile large-scale MIMO systems,” IEEE J. Sel.
5. Areas Commun., vol. 31, no. 2, pp. 251–263, Feb. 2013.
[10] D. Xiaoming, C. Shanzhi, S. Shaohui, K. Shaoli, W. Yinmin, S. Zukang, [19] L. Dai, Z. Wang, J. Wang, and Z. Yang, “Joint channel estimation
and X. Jin, “Successive interference cancelation amenable multiple and time-frequency synchronization for uplink TDS-OFDMA systems,”
access (SAMA) for future wireless communications,” in Proc. IEEE IEEE Trans. Consum. Electron., vol. 56, no. 2, pp. 494–500, May 2010.
ICCS 2014, Nov. 2014, pp. 1–5. [20] L. Dai, Z. Wang, and Z. Yang, “Time-frequency training OFDM with
[11] L. Dai, Z. Wang, and S. Chen, “A novel uplink multiple access scheme high spectral efficiency and reliable performance in high speed environ-
based on TDS-FDMA,” IEEE Trans. Wireless Commun., vol. 10, no. 3, ments,” IEEE J. Sel. Areas Commun., vol. 30, no. 4, pp. 695–707, May
pp. 757–761, Mar. 2011. 2012.

You might also like