See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/267272067

Cooperative Non-Orthogonal Multiple Access in 5G Systems

Article  in  IEEE Communications Letters · October 2014

DOI: 10.1109/LCOMM.2015.2441064 · Source: arXiv

CITATIONS READS
893 1,415

3 authors:

Zhiguo Ding Mugen Peng

The University of Manchester Beijing University of Posts and Telecommunications
468 PUBLICATIONS   21,103 CITATIONS    482 PUBLICATIONS   9,818 CITATIONS   

SEE PROFILE SEE PROFILE

H. Vincent Poor
Princeton University
2,441 PUBLICATIONS   90,207 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

A Machine Learning Approach for Task and Resource Allocation in Mobile Edge Computing Based Networks View project

Resilience of Energy Infrastructure and Services View project

All content following this page was uploaded by H. Vincent Poor on 14 March 2015.

The user has requested enhancement of the downloaded file.

 1

Cooperative Non-Orthogonal Multiple Access in 5G Systems

Zhiguo Ding, Member, IEEE, Mugen Peng, Senior Member, IEEE, and H. Vincent Poor, Fellow, IEEE

Abstract—Non-orthogonal multiple access (NOMA) has re- II. S YSTEM M ODEL

cently received considerable attention as a promising candidate
for 5G systems. A key feature of NOMA is that users with better
channel conditions have prior information about the messages of
the other users. This prior knowledge is fully exploited in this
paper, where a cooperative NOMA scheme is proposed. Outage
probability and diversity order achieved by this cooperative
NOMA scheme are analyzed, and an approach based on user
pairing is also proposed to reduce system complexity in practice.
arXiv:1410.5846v1 [cs.IT] 21 Oct 2014
Consider a broadcast channel with one base station (the
source), and K users (the destinations). Cooperative NOMA
consists of two phases, as described in the following.
A. Direct Transmission Phase
During this phase, the base station sends K messages to
the destinations based on the NOMA principle, i.e., the base

PK
station sends m=1 pm sm , where sm is the message for the
m-th user, and pm is the power allocation coefficient. The
I. I NTRODUCTION observation at the k-th user is given by
Non-orthogonal multiple access (NOMA) is fundamentally K
different from conventional orthogonal multiple access (MA) y1,k =
X
hk p m sm + n k , (1)
schemes, where multiple users are encouraged to transmit at
m=1
the same time, code and frequency, but with different power
levels [1]. In particular, NOMA allocates less power to the where hk denotes the Rayleigh fading channel gain from
users with better channel conditions, and these users can de- the base station to the k-th user and nk denote the additive
code their own information by applying successive interference Gaussian noise. Without loss of generality, consider that the
cancellation [2]. Consequently the users with better channel users are ordered based on their channel quality, i.e.,
conditions will know the messages intended to the others;
|h1 |2 ≤ · · · ≤ |hK |2 . (2)
however, such prior information has not been exploited by the
existing works about NOMA [3] and [4]. The use of NOMA implies |p1 |2 ≥ · · · ≥ |pK |2 , with
In this paper, a cooperative NOMA transmission scheme
P K 2
m=1 pm = 1. Successive detection will be carried out at
is proposed by fully exploiting prior information available the K-th user at the end of this phase. The receiving signal to
in NOMA systems. In particular, the use of the successive noise ratio (SNR) for the K-th ordered user to detect the k-th
detection strategy at the receivers means that users with better user’s message, 1 ≤ k < K, is given by
channel conditions need to decode the messages for the others,
and therefore these users can be used as relays to improve |hK |2 |pk |2
SN RK,k = PK 1
, (3)
the reception reliability for the users with poor connections to m=k+1 |hH 2
K pm | + ρ
the base station. Local short-range communication techniques,
such as bluetooth and ultra-wideband (UWB), can be used to where ρ is the transmit SNR. After these users’ messages are
deliver messages from the users with better channel conditions decoded, the K-th user can decode its own information at the
to the ones with poor channel conditions. The outage proba- following SNR
bility and diversity order achieved by this cooperative NOMA
SN RK,K = ρ|hK |2 |pk |2 . (4)
scheme are analyzed, and these analytical results demonstrate
that cooperative NOMA can achieve the maximum diversity Therefore the conditions under which the K-th user can decode
gain for all the users. In practice, inviting all users in the its own information are given by
network to participate in cooperative NOMA might not be
realistic due to two reasons. One is that a large amount of log(1 + SN RK,k ) > Rk , ∀1 ≤ k ≤ K,
system overhead will be consumed to coordinate multi-user where Rk denotes the targeted data rate for the k-th user.
networks, and the other is that user cooperation will consume
extra short-range communication resources. User pairing is B. Cooperative Phase
a promising solution to reduce system complexity, and we During this phase, the users cooperate with each other via
demonstrate that grouping users with high channel quality does short range communication channels. Particularly the second
not necessarily yield a large performance gain over orthogonal phase consists of (K − 1) time slots. During the first time
MA. Instead, it is more preferable to pair users whose channel
conditions are more distinctive. PK−1 of the (K − 1)
slot, the K-th user broadcasts the combination
messages with the coefficients qK , i.e., m=1 qK,m sm , where
PK−1 2
The authors are with the Department of Electrical Engineering, Princeton m=1 qK,m = 1. The k-th user observes the following
University, Princeton, NJ 08544, USA. Z. Ding is also with the School of
Computing and Communications, Lancaster University, LA1 4WA, UK. M. K−1
X
Peng is also with the Institute of Telecommunications, Beijing University of y2,k = gK,k qK,m sm + n2,k , (5)
Posts and Telecommunications, Beijing, China. m=1
 2

for k < K, where gK,k denotes the inter-user channel gain. i ≤ n with the special case of aK−n 2
K−n,n = qK−n+1,K−n
The (K−1)-th user uses maximum ratio combining to combine K−n
and bK−n,n = 0. In addition, define ak,0 K−n
= p2k and
the observations from both phases, and the SNR for this user K−n PK 2
bk,0 = m=k+1 pm , for 1 ≤ k ≤ (K − n). By using the
to decode the k-th user’s message, k < (K − 1), is given by
definition of the outage probability, we can have the following
|hK−1 |2 p2k proposition for the diversity order achieved by the proposed
SN RK−1,k = PK 1
(6) cooperative NOMA scheme.
|hK−1 | 2 p2m +
m=k+1 ρ
|gK,K−1 |2 qK,k
2 Proposition 1. Assume that the (n − 1) best users can achieve
+ . reliable detection. The proposed cooperative NOMA scheme
K−1 2
|gK,K−1 |2 m=k+1 qK,m + ρ1
P
can ensure that the (K − n)-th ordered user experiences a
aK−n
After the (K − 1)-th user decodes the other users’ messages, diversity order of K, conditioned on ǫk < k,i
bK−n
, for 1 ≤ k ≤
k,i
it can decode its own information with the following SNR (K − n) and 0 ≤ i ≤ n.
|hK−1 |2 p2K−1 K−n
Proof: For notational simplicity, define zk,i =
SN RK−1,K−1 = + |gK,K−1 |2 qK,K−1
2
. (7)
|hK−1 |2 p2K + ρ1 |gK−i+1,K−n |2 qK−i+1,k
2
P
|gK−i+1,K−n |2 K−i 2 1 , where 1 ≤ k ≤
m=k+1 qK−i+1,m + ρ
Similarly at the n-th time slot, 1 ≤ n ≤ (K − 1), the (K − n+ (K − n) and 1 ≤ i ≤ K−n
n, except zK−n,n =
1)-th user broadcasts the combination of Pthe (K −n) messages ρ|gK−n+1,K−n |2 qK−n+1,K−n
2
. In K−n
addition, define zk,0 =
K−n
with the coefficients qK−n+1,m , i.e., m=1 qK−n+1,m sm . |hK−n |2 p2k
P . The SNRs can be expressed as follows:
The k-th user, k < (K − n + 1), observes |hK−n |2 K 2 1
m=k+1 pm + ρ

K−n n
X
K−n K−n
X
H SN RK−n,k = zk,0 + zk,i , (12)
y2,k = gK−n+1,k qK−n+1,m sm + nn+1,k . (8)
m=1 i=1

Combining the observations from both phases, the (K − n)-th for 1 ≤ k ≤ (K − n). Therefore the outage probability can be
user can decode the k-th user’s message, 1 ≤ k < (K − n), rewritten as follows:
n
!
with the following SNR K−n
X
K−n
K−n
Po = P zk,0 + zk,i < ǫk , ∀k ∈ {1, · · · , K − n}
|hK−n |2 p2k i=1
SN RK−n,k = PK 1
(9)
|hK−n | 2 2
m=k+1 pm +
K−n n
!
ρ X
K−n
X
K−n
n ≤ P zk,0 + zk,i < ǫk .
X |gK−i+1,K−n |2 qK−i+1,k
2
+ K−i
, k=1 i=1
2
|gK−i+1,K−n |2 m=k+1 qK−i+1,m + ρ1
P
i=1 Because channel gains are independent, the outage probability
and it can decode its own information with the following SNR can be further bounded as follows:
K−n n  
|hK−n |2 p2K−n
XY
K−n
PK−n
o ≤ P zk,i < ǫk , (13)
SN RK−n,K−n = PK 1
(10)
|hK−n | 2 2 +
m=K−n+1 pm
k=1 i=0
ρ
K−n K−n
n−1
X |gK−i+1,K−n |2 qK−i+1,K−n
2 All the elements in (12) except zk,0 and zK−n,n share the
+ same structure as follows:
|gK−i+1,K−n |2 K−i 2 1
P
m=K−n+1 qK−i+1,m + aK−n
k,i x
i=1 ρ
K−n
+ ρ|gK−n+1,K−n |2 qK−n+1,K−n
2
. zk,i = K−n . (14)
bk,i x + ρ1
Recall that, without cooperation, the SNR at the k-th user is When x is exponentially distributed, the cumulative density
|hK−n |2 p2K−n K−n
P 1 . Compared it to (10), one can find function (CDF) of zk,i is given by
|hK−n |2 K 2
m=K−n+1 pm + ρ
out that the use of cooperation can boost reception reliability. 
aK−n
 1, k,i
if z ≥ bK−n
III. P ERFORMANCE A NALYSIS PzK−n (Z < z) = k,i ,
z
k,i −
Provided that the (n − 1) best users can achieve reliable 
1 − e ρ(ak,i −bk,i z) , otherwise
K−n K−n

detection, the outage probability for the (K − n)-th user can

be expressed as follows: where the definitions for the coefficients aK−nk,i and bK−n
k,i are
given in the proposition.
PK−n
o , P(SN RK−n,k < ǫk , ∀k ∈ {1, · · · , K − n}), (11) At high SNR, ρ aK−nǫ−b k
→ 0, and the probability for
( k,i K−n
k,i z)
where ǫk = 2Rk − 1. Note that the use of local short- K−n
the event, zk,i < ǫk , can be approximated as follows:
range communications does not reduce the data rate. For ǫk
notational simplicity, define aK−n 2
= qK−i+1,k and bK−n = − ǫk
 
ρ(a ǫk )
K−n K−n
k,i k,i K−n
PK−i 2
P zk,i < ǫk = 1 − e k,i
−b
k,i ≈ K−n , (15)
m=k+1 qK−i+1,m , where 1 ≤ k ≤ (K − n) and 1 ≤ ρak,i
 3

aK−n
which is conditioned on ǫk < k,i
. Reducing System Complexity via User Pairing
bK−n
k,i
The density functions of the two special cases, K−n
zk,0and Inviting all the users to participate in NOMA may not be
K−n preferable in practice because of the extra system overhead to
zK−n,n , can be obtained as follows. Note that the source-user coordinate multiple users. Furthermore, the more users partic-
channels are sorted according to their quality. By applying ipate into cooperation, the more short-range communication
K−n
order statistics [5], the CDF of zk,0 can be found as follows: bandwidth resource is consumed. A promising solution to this
issue is to reduce the number of users for cooperation. Without
PzK−n (Z < z) = loss of generality, we focus on the case to select only two users.
k,0 An important question to be answered here is which two users
aK−n
 1, z≥ k,0

 if bK−n
should be grouped together.
k,0
R z
ρ . Consider that the users are ordered as (2), and the m-th
 a
K−n K−n
z e−x K−n−1 and n-th users are paired together, m < n. The conventional
(K−n−1)! x dx, otherwise
 −b
 k,0 k,0
0 TDMA can achieve the following rates
Again applying the high SNR approximation, the probabil- 1 1
R̄m = log 1 + ρ|hm |2 , R̄n = log 1 + ρ|hn |2 . (19)



K−n
ity, P(zk,0 < ǫk ), can be approximated as follows: 2 2
ǫk

K−n
 Z
ρ(a
K−n K−n
ǫk ) x
K−n−1 −x
e The rates achieved by cooperative NOMA is quite com-
P zk,0 < ǫk =
−b
k,i k,i
dx plicated, so we first consider conventional NOMA which can
0 (K − n − 1)!
achieve the following rates
ǫK−n
k
≈ , (16) ρ|hm |2 p2m
 
 K−n
(K − n)! ak,i K−n
ρK−n Rm = log 1 + , (20)
ρ|hm |2 p2n + 1
aK−n
k,0 and
conditioned on ǫk < bK−n
. Similarly the probability for the
= log 1 + ρpn |hn |2 ,
k,0


K−n Rn (21)
event zK−n,n < ǫk can be approximated as follows:
|h |2 p2
 
K−n ǫk−n where Rn is achievable since log 1 + |hnn|2 p2 m ≥ Rm .
P(zK−n,n < ǫK−n ) ≈ 2 . (17) n +1
qK−n+1,K−n ρ The gap between the two sum rates achieved by TDMA and
Combining (13), (15), (16) and (17), the diversity order conventional NOMA can be expressed as follows:
achieved by the cooperative NOMA scheme can be obtained, Rm + Rn − R̄m − R̄n (22)
which completes the proof. 2 2
log ρ|hn |2
 
p 2 2 log ρ|hm |
≈ log 1 + m + log ρp n |h n | − −
The overall system outage event is defined as the event that p2n 2 2
any user in the system cannot achieve reliable detection, which 2 2
log |hn | log |hm |
means the overall outage probability is defined as follows: = − ,
2 2
K
Y where the approximation is obtained at high SNR. It is
Po , 1 − 1 − Pko .


(18) interesting to observe that the gap is not a function of power
k=1 allocation coefficients pm . By applying order statistics, the
By using Proposition 1 and the fact that the source-destination average of the gap can be calculated as follows:
channels are independent, the following lemma can be obtained
E{Rm + Rn − R̄m − R̄n } (23)
straightforwardly. Z ∞ Z ∞
−x n−1 −x m−1
log x e x log x e x
Lemma 1. The proposed cooperative NOMA scheme can ≈ dx − dx.
0 2 (n − 1)! 0 2 (m − 1)!
ensure that the n-th best user, 1 ≤ n ≤ K, experiences
aK−n
k,i By using Eq. (4.352.1) in [6], the averaged gap is given by
a diversity order of K, conditioned on ǫk < bK−n
, for
k,i
1 ≤ k ≤ (K − n) and 0 ≤ i ≤ n. E{Rm + Rn − R̄m − R̄n } (24)
n−1
Note that K is the maximum diversity order to the addressed log e X1
≈ (ψ(n) − ψ(m)) = ,
scenario. For example, the user with the worst channel con- 2 i
i=m
dition gets help from the other (K − 1) users, in addition to
its own direct channel to the source, which implies that the where ψ(x) denotes the Euler’s integral, and the last equation
maximum diversity for this scenario is K. As can be seen is due to the property of ψ(x), i.e., ψ(x + 1) = ψ(x) + x1 .
from Lemma 1, encouraging user cooperation can ensure that Therefore to conventional NOMA, the worst choice of m
the maximum diversity order of K is achievable to all users, and n is n = m + 1, and it is ideal to group two uses
regardless of the quality of their direct link to the base station, who experience significantly different channel fading. This
whereas a non-cooperative NOMA can achieve only a diversity observation is also valid to cooperative NOMA. Particularly an
order of n for the n-th ordered user [4]. important observation from (3) is that the data rate for the m-th
 4

 
ρ|h |2 p2 1
user is bounded as Rm ≤ log 1 + ρ|hnn|2 p2 m , although Rm Orthogonal MA, ρ=15 dB
n +1 Non−cooperative NOMA, ρ=15 dB
0.99
ρ|hm |2 p2m
  Cooperative NOMA, ρ=15 dB
can be as large as log 1 + ρ|hm |2 p2 +1 + ρ|gn,m |2 , where Orthogonal MA, ρ=10 dB

Probability[ Mutual information > Abscissa]

0.98
n Non−cooperative NOMA, ρ=10 dB
the bound is due to the fact that the n-th user needs 2to2decode 0.97
Cooperative NOMA, ρ=10 dB
ρ|h | p
the m-th user’s information. Since log 1 + ρ|hnn|2 p2 m +1 ≈
  n 0.96
2
p
log 1 + pm 2 , the conclusion obtained for conventional 0.95
n
NOMA can also be applied to cooperative NOMA. 0.94

0
10 0.93

0.92

−1 0.91
10

0.9
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Outage Probability

Mutual information in BPCU

−2
10
Fig. 2. Outage capacity achieved by cooperative NOMA.
others’ messages. Analytical results have been developed to
demonstrate that cooperative NOMA can achieve the maxi-
−3
10 Orthogonal MA mum diversity gain for all users. User pairing and its impact
Non−cooperative NOMA
cooperative NOMA on system throughput have also been discussed in order to
Solid lines: R1=R2=1.5 BPCU
Dashed lines: R1=R2=0.5 BPCU
implement cooperative NOMA with low system complexity.
−4
10
0 5 10 15 20 25 30 35 40
One promising future direction is to apply game theoretic
SNR in dB
algorithms for opportunistic user pairing/grouping, where users
Fig. 1. Outage probability achieved by cooperative NOMA.
can form coalitions in a distributed manner [7].
3
IV. N UMERICAL S TUDIES
In this section, the performance of cooperative NOMA is 2.5

evaluated by using computer simulations. In Fig. 1, the outage

probability achieved by the three schemes, e.g., the orthogonal Averaged sum rate difference
2

MA scheme, non-coopertive NOMA, and cooperative NOMA, Non−cooperative NOMA with

the (K−1)−th ordered user scheduled
1.5
is shown as a function of SNR, with K = 2. As can be seen Coperative NOMA with
the (K−1)−th ordered user scheduled
from the figure, cooperative NOMA outperforms the other two 1
Non−cooperative NOMA with
the first ordered user scheduled
schemes, since it can ensure that the maximum diversity gain Coperative NOMA with
the first ordered user scheduled
is achievable to all the users as indicated by Lemma 1. In 0.5

Fig. 2, the outage capacity achieved by the three schemes

is demonstrated, by setting R1 = R2 . With 10% outage 0

probability, the orthogonal MA scheme can achieve a rate of

−0.5
0.7 bits per channel use (BPCU), non-cooperative NOMA can 0 5 10 15 20 25
SNR in dB
30 35 40 45 50

support 0.95 BPCU, and cooperative NOMA can support 1.7 Fig. 3. The impact of user pairing on the sum rate.
BPCU, much larger than the other two schemes.
In Fig. 3, the impact of user pairing is investigated by R EFERENCES
studying the difference between the sum rates achieved by the [1] Y. Saito, A. Benjebbour, Y. Kishiyama, and T. Nakamura, “System
orthogonal MA scheme and NOMA, i.e., E{Rm + Rn − R̄m − level performance evaluation of downlink non-orthogonal multiple access
(NOMA),” in Proc. IEEE Annual Symposium on Personal, Indoor and
R̄n } as in (23). Particularly, consider that the K-th ordered Mobile Radio Communications (PIMRC), London, UK, Sept. 2013.
user, i.e., the user with the best channel condition, is scheduled, [2] T. Cover and J. Thomas, Elements of Information Theory, 6th ed. Wiley
and Fig. 3 demonstrates how large a sum rate gain can be and Sons, New York, 1991.
obtained by pairing it with different users. As discussed in [3] J. Choi, “Non-orthogonal multiple access in downlink coordinated two-
Section III, it is helpful to increase E{Rm +Rn − R̄m − R̄n } by point systems,” IEEE Commun. Letters, vol. 18, no. 2, pp. 313–316, Feb.
scheduling two users whose channel connections to the source 2014.
are more distinctive. Such a conclusion is confirmed by the [4] Z. Ding, Z. Yang, P. Fan, and H. Poor, “On the performance of non-
results shown in Fig. 3, where pairing the K-th user with the orthogonal multiple access in 5g systems with randomly deployed users,”
IEEE Signal Processing Letters, vol. 21, no. 12, pp. 1501–1505, Dec
first user, i.e., the user with the worst channel condition, can 2014.
yield a significant gain. This observation is consistent to the
[5] H. A. David and H. N. Nagaraja, Order Statistics. John Wiley, New
motivation of NOMA in [1] which is to schedule two users, York, 3rd ed., 2003.
one close to the cell edge and the other close to the BS. [6] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series and
V. C ONCLUSIONS Products, 6th ed. New York: Academic Press, 2000.
[7] Z. Ding, I. Krikidis, B. Sharif, and H. V. Poor, “Wireless information and
In this paper, we have proposed a cooperative NOMA power transfer in cooperative networks with spatially random relays,”
transmission scheme which fully uses the fact that some IEEE Trans. Wireless Commun., vol. 13, no. 8, pp. 4440–4453, Aug.
users in NOMA systems have prior information about the 2014.

View publication stats