1
On the Design of MIMO-NOMA Downlink and Uplink Transmission
Zhiguo Ding, Member, IEEE, Robert Schober, Fellow, IEEE, and H. Vincent Poor, Fellow, IEEE
Abstract—In this paper, a novel MIMO-NOMA framework for
downlink and uplink transmission is proposed by applying the
concept of signal alignment. By using stochastic geometry, closed-
form analytical results are developed to facilitate the performance
evaluation of the proposed framework for randomly deployed
users and interferers. The impact of different power allocation
strategies, such as fixed power allocation and cognitive radio
inspired power allocation, on the performance of MIMO-NOMA
is also investigated. Computer simulation results are provided to
demonstrate the performance of the proposed framework and
the accuracy of the developed analytical results.
I. I NTRODUCTION
Non-orthogonal multiple access (NOMA) has been recog-
nized as a spectrally efficient multiple access (MA) technique
for the next generation of mobile networks [1] and [2]. For
example, the use of NOMA has been recently proposed for
downlink scenarios in long-term evolution (LTE) systems as
well as in fifth generation (5G) mobile networks [3]–[5].
The key idea of NOMA is to exploit the power domain for
multiple access, which means multiple users can be served
concurrently at the same time, frequency, and spreading code.
The performance of NOMA in a network with randomly
deployed single-antenna nodes was investigated in [2]. User
fairness in the context of NOMA was addressed in [6], where
power allocation was optimized under different assumptions
regarding the channel state information (CSI). The applica-
tion of multiple-input multiple-output (MIMO) techniques to
NOMA is important since the use of MIMO provides addi-
tional degrees of freedom for further performance improve-
ment. In [7], the multiple-input single-output (MISO) scenario,
where the base station had multiple antennas and users were
equipped with a single antenna, was considered. Ref. [8]
focused on a special case of MIMO-NOMA with two users,
where a throughput maximization problem was formulated
and two algorithms were proposed to solve the optimization
problem. In many practical scenarios, it is preferable to serve
as many users as possible in order to reduce user latency and
improve user fairness. Following this rationale, in [9], users
were first grouped into small-size clusters, where NOMA was
implemented for the users within each cluster and MIMO
detection was used to cancel inter-cluster interference. Similar
to [10], this method does not need CSI at the base station;
however, unlike [10], it avoids the use of random beamforming
which can cause uncertainties for the quality of service (QoS)
experienced by the users.
This paper considers a general MIMO-NOMA communi-
cation network where a base station is communicating with
multiple users using the same time, frequency, and spreading
Z. Ding and H. V. Poor are with the Department of Electrical Engineering,
Princeton University, Princeton, NJ 08544, USA. Z. Ding is also with the
School of Computing and Communications, Lancaster University, LA1 4WA,
UK. R. Schober is with the Institute for Digital Communications, University
of Erlangen-Nurnberg, Germany.
code resources, in the presence of randomly deployed inter-
ferers. The contributions of this paper are listed as follows.
Firstly, a general MIMO-NOMA framework which is appli-
cable to both downlink and uplink transmission is proposed, by
applying the concept of signal alignment, originally developed
for multi-way relaying channels in [11] and [12]. By exploiting
this framework, the considered multi-user MIMO-NOMA sce-
nario can be decomposed into multiple separate single-antenna
NOMA channels, to which conventional NOMA protocols can
be applied straightforwardly.
Secondly, since the choice of the power allocation coeffi-
cients is key to achieve a favorable throughput-fairness tradeoff
in NOMA systems, two types of power allocation strategies are
studied in this paper. The fixed power allocation strategy can
realize different QoS requirements in the long term, whereas
the cognitive radio inspired power allocation strategy can
ensure that users’ QoS requirements are met instantaneously.
Finally, exact expressions and asymptotic performance re-
sults are developed in order to obtain an insightful understand-
ing of the proposed MIMO-NOMA framework. In particular,
the outage probability is used as the performance criterion
since it not only bounds the error probability of detection
tightly, but can be also used to calculate the outage capac-
ity/rate. The impact of the random locations of the users and
the interferers is captured by applying stochastic geometry, and
the diversity order is computed to illustrate how efficiently the
degrees of freedom of the channels are used by the proposed
framework.
II. S YSTEM M ODEL
Consider a MIMO-NOMA downlink (uplink) communica-
tion scenario in which a base station is communicating with
multiple users. The base station is equipped with M antennas
and each user is equipped with N antennas. In this paper,
we consider the scenario N > M 2 in order to implement the
concept of signal alignment, an assumption more general than
the one used in [9]. This assumption is applicable to various
communication scenarios, such as small cells in heterogenous
networks and 5G cloud radio access networks.
The users are assumed to be uniformly deployed in a disc,
denoted by D, i.e., the cell controlled by the base station.
The radius of the disc is r, and the base station is located at
the center of D. In order to reduce the system load, several
existing studies on NOMA have proposed to pair two users
for the implementation of NOMA, and have demonstrated that
it is ideal to pair two users whose channel conditions are
very different [13], [14]. Based on this insight, we assume
that the disc is divided into two regions. The first region is a
smaller disc, denoted by D1 , with radius r1 (r1 < r) and the
base station located at its origin. The second region is a ring,
denoted by D2 , constructed from D by removing D1 . Assume
that M pairs of users are selected, where user m, randomly
located in D1 , is paired with user m′ , randomly located in D2 .
 2

Hence, the users are randomly scheduled and paired together. where 1m denotes an m × 1 all-one vector, and dIj ,m denotes
The use of more sophisticated schedulers can further improve the distance from user m to the j-th interference source. Note
the performance of the proposed MIMO-NOMA framework that small scale fading has been omitted in the interference
of course, but this is beyond the scope of this paper. model, since the effect of path loss is more dominant for
In addition to the messages sent by the base station, the interferers located far away. In addition, this simplification will
downlink NOMA users also observe signals sent by inter- facilitate the development of tractable analytical results.
ference sources which are distributed in R2 according to a User m applies a detection vector vm to its observation, and
homogeneous Poisson point process (PPP) ΨI of density λI therefore the user’s observation can be re-written as follows:
[15]. The same assumption is made for the uplink case. In Gm
H
vm ym = vm H
√ pm (αm sm + αm′ sm′ ) (3)
practice, these interferers can be cognitive radio transmitters, L(dm )
WiFi access points in LTE in the unlicensed spectrum (LTE- ∑ H Gm
+ vm √ H
pi (αi si + αi′ si′ ) + vm (wIm + nm ),
U), or transmitters from different tiers in heterogenous net- i̸=m
L(d m )
works. In order to obtain tractable analytical results, it is | {z }
interference (including inter−pair interference) + noise
assumed that the interference sources are equipped with a
single antenna and use identical transmission powers, denoted where pm denotes the m-th column of P.
by ρI . In order to remove inter-pair interference, the following
Consider the use of a composite channel model with both constraint has to be met:
quasi-static Rayleigh fading and large scale path loss. In [ H ]
vm Gm
particular, the channel matrix from the base station to user H pi = 02×1 , ∀i ̸= m, (4)
vm ′ G m′
m is Hm = √Gm , where Gm denotes an N × M matrix
L(dm )
whose elements represent Rayleigh fading channel gains, dm where 0m×n denotes the m×n all zero matrix. Without loss of
generality, we focus on p1 which needs to satisfy the following
denotes the distance from the base station to the user, and the constraint:
path loss is modelled as follows: [ ]H
{ α GH
2 v2 GH
2′ v2′ ··· GH
M vM GH
M ′ vM ′ p1 = 0. (5)
dm , if dm > r0
L(dm ) = ,
r0α , otherwise It is easy to verify that for general vm a non-zero vector
pi satisfying (5) does not exist. To overcome this problem, in
where α denotes the path loss exponent and parameter r0
this paper, the concept of interference alignment is applied,
avoids a singularity when the distance is small. It is assumed
which means the detection vectors are designed to satisfy the
that r1 ≥ r0 in order to simplify the analytical results.
following constraint [17], [18]
For notational simplicity, in case of uplink transmission, the
channel matrix from user m to the base station is denoted by H
vm Gm = v mH
′ G m′ , (6)
HHm , where (·)
H
denotes the Hermitian operator. We denote [ ]
the transpose operator by (·)T . Global CSI is assumed to [ ] vm
or equivalently GH m −Gm′
H
= 0M ×1 . Define Um
be available at the users and the base station. The proposed vm′
MIMO-NOMA framework for downlink and uplink transmis- as the 2N × (2N − M [ ) matrix containing
] the (2N − M ) right
sion is described in the following two subsections, respectively. singular vectors of GH m −G H
m ′ corresponding to its zero
singular values. Therefore, the detection vectors at the users
are designed as follows:
A. Downlink MIMO-NOMA Transmission [ ]
The base station sends the following vm
= Um xm , (7)
M × 1 information-bearing vector s = v m′
[ ]T
α1 s1 + α1′ s1′ · · · αM sM + αM ′ sM ′ , where sm where xm is a (2N − M ) × 1 vector to be defined later.
is the signal intended for the m-th user, αm is the power We normalize xm to 2, i.e., |xm |2 = 2, due to the following
2 2
allocation coefficient, and αm + αm ′ = 1. The choice of the two reasons. First, the uplink transmission power has to be
power allocation coefficients will be discussed later. constrained as shown in the following subsection. Second,
Without loss of generality, we focus on user m, whose this facilitates the performance analysis carried out in the
observation is given by next section. It is straightforward to[ show that the] choice of
the detection vectors in (7) satisfies GH m −Gm′ Um xm =
H
Gm
ym = √ Ps + wIm + nm , (1) 0M ×1 .
L(dm )
The effect of the signal alignment based design in (6) is the
where P is the M × M precoding matrix to be defined at the projection of the channels of the two users in the same pair
end of this subsection, wIm denotes the overall co-channel into the same subspace. Define gm , GH m vm as the effective
interference received by user m, and nm denotes the noise channel vector shared by the two users. As a result, the number
vector. Following the classical shot noise model in [16], the of rows in the matrix in (5) can be reduced significantly. In
co-channel interference, wIm , can be expressed as follows: particular, the constraint for pi in (5) can be rewritten as
∑ √ follows:
ρI
wIm , √ 1N , (2) [ ]H
j∈Ψ
L(dIj ,m ) g1 · · · gi−1 gi+1 · · · gM pi = 0(M −1)×1 . (8)
I
 3

[ ]H
Note that g1 · · · gi−1 gi+1 · · · gM is an (M − assume that the total transmission power from one user pair
1) × M matrix, which means that a pi satisfying (8) exists. is normalized as follows:
[ ]H
Define G , g1 · · · gM . A zero forcing based pre-
coding matrix at the base station can be designed as follows:
2
αm |vm |2 + αm
2
′ |vm′ | ≤ 2ρ.
2
(14)

P = G−H D, (9) The base station observes the following signal:

( )
where D is a diagonal matrix to ensure power ∑M
GH GH
m αm vm sm m′ αm′ vm′ sm′
normalization at the base station, i.e., D2 = yBS = √ + √ (15)
m=1 L(dm ) L(dm′ )
diag{ (G−1 G1−H )1,1 , · · · , (G−1 G−H
1
)M,M
}, where (A)m,m
denotes the m-th element on the main diagonal of A. As + wI + nBS ,
a result, the transmission
{ power
} at the base station can be where wI is the interference term defined as follows
constrained, i.e., tr PPH ρ = M ρ, where ρ denotes the √
transmit signal-to-noise ratio (SNR). ∑ ρI
wI , √ ( ) 1M . (16)
With the design in (6) and (9), the signal model for user m
can now be written as follows: j∈ΨI L dIj ,BS
(αm sm + αm′ sm′ )
H
vm ym = √ H
+ vm (wIm + nm ). (10) Here, dIj ,BS denotes the distance between the base station and
(L(dm ))(G−1 G−H )m,m the j-th interferer, and the noise term is defined similarly as in
the previous sub-section. The base station applies a detection
For notational simplicity, we define hm = matrix P to its observations and the symbols from the m-th
√ 1
and hm′ = √ 1
. user pair can be detected based on
L(dm )(G−1 G−H )m,m L(dm′ )(G−1 G−H )m,m
The use of the signal alignment based precoding and detection ( )
H H
matrices decomposes the multi-user MIMO-NOMA channels G αm vm sm G ′ αm ′ vm′ sm′
pH
m yBS =pm
H
√
m
+ m√ (17)
into M pairs of single-antenna NOMA channels. Therefore, L(dm ) L(dm′ )
it is important to point out that hm and hm′ share the same ( )
∑ GH αi vi si GH i′ αi′ vi′ si′
small scale fading gain with different distances. + pmH
√i
+ √
Recall that two users belonging to the same pair are selected i̸=m
L(di ) L(di′ )
from D1 and D2 , respectively, which means that dm < dm′ . + pH
m (wI + nBS ).
Therefore, the two users from the same pair are ordered
without any ambiguity, which simplifies the design of the In order to avoid inter-pair interference, the following con-
power allocation coefficients, i.e., αm ≤ αm′ , following the straint needs to be met
NOMA principle. User m′ decodes its message with the ( )
∑ GH αi vi si GH
following signal-to-interference-plus-noise ratio (SINR) i′ αi′ vi′ si′
pmH
√i
+ √ = 0, ∀m ̸= i. (18)
L(di ) L(di′ )
ρ|hm′ |2 αm
2
′
i̸=m
SIN Rm′ = , (11)
ρ|hm | αm + |vm | + |vm
′ 2 2 ′ 2 H 1 |2 I ′
′ N m Again applying the concept of signal alignment, the con-
where the interference term is given by straint that GH H
m vm = Gm′ vm′ is imposed on the precoding
∑ vectors vm . Therefore, the same design of vm as shown in (7)
ρ
Im ′ = ( I ). (12) can be used. The total transmission power within one pair is
L dIj ,m′ |vm |2 + ραm ′ |vm′ | ≤ 2ρ, which means
2 2 2
j∈Ψ I constrained, i.e., ραm
User m carries out successive interference cancellation (SIC) that the use of the precoding vector in (7) ensures that the total
by first removing the message to user m′ with SINR, transmission power of one user pair is constrained.
ρ|h |2 α2m′ Following steps similar to those used in the previous sub-
SIN Rm,m′ = ρ|hm |2 α2 +|vmm |2 +|v , and then decod-
m 1N | Im′
H 2
m section, one can observe that the use of the proposed pre-
ing its own message with SINR
coding/detection matrices can decompose the original MIMO-
ρ|hm |2 αm
2 NOMA channels into separated SISO-NOMA channels. Due
SIN Rm = , (13)
|vm |2 + |vmH 1 |2 I
N m
to the space limitation, we will only focus on the downlink
case in the rest of this paper. More details about the uplink
which becomes the SNR if ρI = 0. case can be found in the journal version of this paper [19].

B. Uplink MIMO-NOMA Transmission

For the NOMA uplink case, user m will send out an
information bearing message sm , and the signal transmitted by
this user is denoted by αm vm sm . Because of the reciprocity
between uplink and downlink channels, vm , which was used
as a downlink detection vector, can be used as a precoding
vector for the uplink scenario. Similarly, P will be used as
the detection matrix for the uplink case. In this paper, we
III. P ERFORMANCE A NALYSIS FOR D OWNLINK
MIMO-NOMA T RANSMISSION
Two types of power allocation policies are considered in this
section. One is fixed power allocation and the other is inspired
by the cognitive radio concept, as illustrated in the following
two subsections, respectively. Recall that the precoding vectors
vm and vm′ are determined by xm as shown in (7). In this
paper, a random choice of xm is considered.
 4

A. Fixed Power Allocation On the other hand, user m first decodes the message for
In this case, the power allocation coefficients αm and user m′ before decoding its own message via SIC. Again, we
αm′ are constant and not related to the instantaneous re- focus on a modified expression for the outage probability as
alizations of the fading channels. We will first focus on follows:
( ( ) )
the outage performance of user m′ . The outage probability ρ|hm |2 αm2
′
P̃m = P log 1 + < R ′ (23)
of user m′ to decode its information is given by Pom′ = ρ|hm |2 αm2 + 2 + 2δI
m
m
( ( )
P (log (1 + SIN Rm′ ) < Rm′ ), where P(x < a) denotes the ρ|hm |2 αm2
probability of the event x < a. The correlation between vm′ + P log 1 + < Rm ,
2 + 2δIm
and hm′ makes the evaluation of the above outage probability ( ) )
ρ|hm |2 αm2
′
very challenging. Hence, we focus on the following modified log 1 + > R m′ ,
expression for the outage probability ρ|hm |2 αm
2 + 2 + 2δI
m
( ( ) ) which is an upper bound for δ ≥ N as explained in the proof
ρ|hm′ |2 αm
2
′
P̃m = P log 1 +
′ < Rm ,′ for Lemma 2. The following lemma provides an exact expres-
ρ|hm′ |2 αm
2 + 2 + 2δI ′
m sion for this probability as well as its high SNR approximation.
where δ ≥ 0. Since |vm′ |2 + |vm |2 = 2, we have
∑N |vm | 2≤ 2
′
2
′ ≤ αm ϵm′ , the probability P̃m = 1.
2 2
Lemma 2. If αm
and |vm | ≤ 2. In addition, because ( N n=1 xn ) ≤
2 1
∑N Otherwise, the probability P̃m′ can be expressed as follows:
N
1
n=1 xn , |vm′ 1N | ≤ N |vm | . Therefore, we have
2 H 2 ′
2 ∫ r0
2
e−2ϕ̃m r0 φI (r0 )xdx
α
P̃m = 1 − 2 (24)
Pom′ ≤ P̃m′ , (19) r1 0
∫ r1
2
e−2ϕ̃m x φI (x)xdx,
α
for δ ≥ N , which means that P̃m′ provides an upper bound − 2
on Pm′ if δ ≥ N . Note that a choice of δ = 1 is sufficient to r1 r0
ensure that P̃m′ provides a very tight approximation to Pm′ as where ϕ̃m = max{ϕm , ϕm′ } and ϕm = ρα ϵm
2 . If ρI is fixed and
m
shown in [19]. In addition, the use of P̃m′ will be sufficient to the transmit SNR ρ approaches infinity, the outage probability
identify the achievable diversity order of the proposed MIMO- can be approximated as follows:
NOMA scheme.
Given a random choice of xm , the following lemma pro- ϕ̃m (2 + θ̃m′ ) ( α+2 )
P̃m ≈ αr0 + 2r1α+2 , (25)
vides an exact expression for P̃m′ as well as its high SNR r12 (α + 2)
approximation. where θ̃m′ was defined in Lemma 1.
′ ≤ αm ϵm′ , the probability P̃m′ = 1,
2 2
Lemma 1. If αm Proof: Please refer to Appendix B in [19].
where ϵm′ = 2 m′ − 1. Otherwise, the probability P̃m′ can be
R

expressed as follows:
∫ r B. Cognitive Radio Power Allocation
2
e−2ϕm′ x φI (x)xdx,
α
P̃m′ = 1 − 2 (20) In this section, a cognitive radio inspired power allocation
r − r1 r1
2
strategy is studied. In particular, assume that user m′ is viewed
where ϕ m′ = ϵm′
, φI (x) = as a primary user in a cognitive ratio network. With orthogonal
ρα2m′ −ρα2m ϵm′
2 ( ) multiple access, the bandwidth resource occupied by user m′
m′
β (x)
−πλI (β ′ (x)) α γ 1
, α
e m α
, βm′ (x) = 2ϕm′ δρI L (xα ),
r0 cannot be reused by other users, despite its poor channel
and γ(·) denotes the incomplete Gamma function. conditions. In contrast, with NOMA, one additional user, i.e.,
If ρI is fixed and transmit SNR ρ approaches infinity, the user m, can be served simultaneously, under the condition that
outage probability can be approximated as follows: the QoS requirements of user m′ can still be met.
( ) In particular, assume that user m′ needs to achieve a target
2ϕm′ (2 + θ̃m′ ) rα+2 − r1α+2 data rate of Rm′ , which means that the power allocation
P̃m′ ≈ , (21)
r2 − r12 α+2 coefficients of NOMA need to satisfy the following constraint
where θ̃m′ = 2πλI δρI rα0 . When ρI = 0, P̃m′ simplifies to ρ|hm′ |2 αm
2
′
> ϵ m′ , (26)
1 ( α α
) ρ|hm′ |2 αm
2 + |vm′ | + |vm
2 H 1 |2 I ′
′ N m
P̃m′ = 1 − 2 e−2ϕm′ r r2 − e−2ϕm′ r1 r12 (22)
r − r1 2
which leads to the following choice for αm
2 ( ( ) ( )) ( )
(2ϕm′ )− α 2 2 ρ|hm′ |2 − ϵm′ (|vm′ |2 + |vm
H
′ 1 N | Im ′ )
2
− γ + 1, 2ϕm ′r
α
− γ + 1, 2ϕm
α
′ r1 . 2
αm = max 0, .
r2 − r12 α α (1 + ϵm′ )ρ|hm′ |2
Proof: Please refer to Appendix A in [19], which is the (27)
journal version of this paper It is straightforward to show that
By using the high SNR approximation in Lemma 1 and ρ|hm′ |2 −ϵm′ (|vm′ |2 +|vm
H
′ 1N | Im′ )
2

(1+ϵm′ )ρ|hm′ |2 is always less than one.

also the fact that both ϕm′ and θm′ are at the order of ρ1 , the
following corollary can be obtained straightforwardly.
Corollary 1. If αm2 2
′ > αm ϵm′ , the diversity order achieved
by the proposed MIMO-NOMA framework for user m′ is one.
′
An outage at user m means here that all power is allocated
to user m′ , but outage still occurs. As a result, the outage
probability of user m′ is exactly the same as that in conven-
tional orthogonal multiple access systems. Therefore, in this
 5

Outage Sum rate: (1 − Pm‘ )Rm‘ + (1 − Pm )Rm

5.5
section, we only focus on the outage probability of user m
5
which can be expressed as follows:
( { 4.5

Pom =P |hm |2 < max ϕm′ (|vm |2 + |vm H

1N |2 Im ), (28) 4
})
ϕm (|vm |2 + |vm
H
1 N | 2 Im ) , 3.5

3
2 2
if αm ′ > αm ϵm′ , otherwise outage always occurs. It can be
2.5
′ ≤ αm ϵm′ is equivalent to αm = 0, in the
2 2
verified that αm 2
context of cognitive radio power allocation. 1.5
Analyzing this outage probability is very difficult due to MIMO−OMA without precoding
1
the following two reasons. First, hm and vm are correlated, MIMO−OMA with precoding
0.5 MIMO−NOMA without precoding
and second, the users experience different but correlated SA−MIMO−NOMA
0
co-channel interference, i.e., Im ̸= Im′ . Therefore, in this 0 5 10 15 20
Transmission Power in dBm
25 30 35 40

subsection, we only focus on the case without co-channel

(a) Outage Sum Rate
interference, i.e., ρI = 0. In particular, we focus on the
following outage probability 10
0

( { })
P̃m =P |hm |2 < 2 max ϕ̄m′ , ϕ̄m , (29)

Outage Probabilities: Pm‘ and Pm

−1
10
ϵm′
where ϕ̄m = ρϵᾱm2 , ϕ̄m′ = ρᾱ2 −ρ 2
, and ᾱm =
( m ) m′
ᾱ2m ϵm′
ρ|hm′ | −2ϵm′
2
max 0, (1+ϵ . 10
−2

m′ )ρ|hm′ |
2
Pm‘ , Rm‘ =
Similarly to the case with fixed power allocation, the outage 0.5 BPCU
Pm , Rm =
5 BPCU
probability P̃m tightly bounds Pom . The following lemma 10
−3

provides the expression for the outage probability P̃m .

−4
10
Lemma 3. When ρI = 0, the outage probability can be MIMO−OMA without precoding
MIMO−OMA with precoding
expressed as follows: MIMO−NOMA without precoding
( ) ( ) SA−MIMO−NOMA

2ϵm′ 2ϵm (1 + ϵm′ ) 0 5 10 15 20 25 30 35 40

P̃m = 1 − Υ1 Υ2 , (30) Transmission Power in dBm
ρ ρ (b) Outage Probabilities
( −yrα 2 ) Fig. 1. Performance comparison for downlink transmission. Rm = 5 bits
r − e−yr1 r12
α
1
where Υ1 (y) = r 2 −r12
e + per channel use (BPCU) and Rm′ = 0.5 BPCU. r = 20m and r1 = 10m.
y− α
2 ( (2 ) ( )) M = N = 3. r0 = 1m. am′ = 43 . The path loss exponent is α = 3. The
r 2 −r12
γ α + 1, yrα − γ α2 + 1, yr1α , and noise power is −30dBm and the interference power is ρI = 0.
2 −zr0 α ( )
+ r12 e−zr1 r12 − e−zr0 r02 +
r0 e α α
Υ2 (z) = r12
z− α
2 ( (2 ) (2 1
)) design for the two schemes without precoding can be found in
r12
γ α + 1, zr1 − γ α + 1, zr0α . At high SNR,
α
[9]. The MIMO-OMA scheme with precoding serves M users
the outage probability can be approximated as follows: during each orthogonal channel use, e.g. one time slot, whereas
4ϵm′ ( α+2 ) 2M users are served simultaneously by the proposed scheme.
P̃m ≈ r − r1α+2 (31)
ρ(2 + α)(r − r1 )
2 2 The framework proposed in this paper is termed SA-MIMO-
2r2+α ϵm (1 + ϵm′ ) 4ϵm (1 + ϵm′ ) ( α+2 ) NOMA. The size of D1 and D2 is determined by r = 20m,
+ 0 2 + 2 r1 − r0α+2 . and r1 = 10m. The parameter for the bounded path loss model
ρr1 ρ(2 + α)r1
is set as r0 = 1.
Proof: Please refer to Appendix C in [19]. Since the benchmark schemes were proposed for the
By using the above lemma, it is straightforward to show interference-free scenario, Fig. 1 shows the performance com-
that a diversity gain of one is still achievable at user m (i.e., parison of the four schemes for ρI = 0. In Fig. 1(a), the down-
there is no error floor), and it is important to point out that link outage sum rate, defined as Rm′ (1−Pm′ )+Rm (1−Pm ),
this is achieved when user m′ experiences the same outage is shown as a function of transmission power, and the corre-
performance as if it solely uses the channel. Therefore, by sponding outage probabilities are studied in Fig. 1(b). As can
using the proposed cognitive radio NOMA, one additional be seen from the figures, the two NOMA schemes can achieve
user, user m, is introduced into the system to share the larger outage sum rates compared to the two OMA schemes,
spectrum with the primary user, user m′ , without causing any which demonstrates the superior spectral efficiency of NOMA.
performance degradation at user m′ . In Fig. 1(b), the two schemes with precoding can achieve better
outage performance than the two schemes without precoding,
IV. N UMERICAL S TUDIES due to the efficient use of the degrees of freedom at the base
In this section, the performance of the proposed NOMA station. Comparing the proposed SA-MIMO-NOMA scheme
framework is investigated by using computer simulations. The with the one proposed in [9], one can observe that their outage
performance of three benchmark schemes, termed MIMO- sum rate performances are similar, but SA-MIMO-NOMA can
OMA without precoding, MIMO-OMA with precoding, and offer much better reception reliability, particularly for high
MIMO-NOMA without precoding, is shown in Fig. 1. The transmission power. In terms of individual outage probability,

0 0
10
Outage Probabilities
−1
10
Solid lines: simulations
−2
10 Dash-dotted lines: analytical results
P̃m‘ , ρI = ρ Rm’=2 BPCU, Rm=4 BPCU
10
P̃m‘ , ρI = 10dBm
ρ
P̃m , ρI = 10
−3 P̃m , ρI = 10dBm
10
0 5 10 15 20 25
Transmission Power in dBm
Fig. 2. Outage probabilities P̃m′ and P̃m for downlink transmission. λI =
10−4 , δ = 1, r = 20m, r1 = 10m, M = N = 2, r0 = 1m, and am′ = 34 .
Rm′ = Rm = 1 BPCU. The path loss exponent is α = 3 and the noise
power is −30dBm.
SA-MIMO-NOMA can ensure a lower outage probability at
user m, i.e., a smaller Pm , compared to the MIMO-OMA
scheme with precoding, but results in performance degradation
for the outage probability at user m′ , i.e., an increase of Pm′ .
This is consistent with the finding in [14] which shows that the
NOMA user with poorer channel conditions will suffer some
performance loss due to the co-channel interference from its
partner.
In Fig. 2, the accuracy of the analytical results developed
in Lemmas 1 and 2 for downlink transmission is verified. As
can be seen from Fig. 2, the exact expression developed in
Lemma 1 perfectly matches the computer simulations, and the
asymptotic results developed in Lemma 1 are also accurate
at high SNR, as shown in Fig. 2. The accuracy of Lemma
2 can be confirmed similarly. Note that error floors appear
when increasing ρI in Fig. 2, which is expected due to
the strong co-channel interference caused by the randomly
deployed interferers.
In Fig. 3, the performance of the cognitive radio power
allocation scheme proposed in Section III-B is studied. In
particular, given the target data rate at user m′ , the power
allocation coefficients can be calculated opportunistically ac-
cording to (27). As can be seen from the figure, the probability
for this NOMA system to support the secondary user, i.e., user
m, but with a target data rate of Rm approaches one at high
SNR. Note that with OMA, user m cannot be admitted into
the channel occupied by user m′ , and with cognitive radio
NOMA, one additional user, user m, can be served without
degrading the outage performance of the primary user.
V. C ONCLUSIONS
In this paper, we have proposed a signal alignment based
framework which is applicable to both MIMO-NOMA down-
link and uplink transmission. By applying tools from stochas-
tic geometry, the impact of the random locations of the users
and interferers has been captured, and closed-form expressions
for the outage probability achieved by the proposed framework
have been developed to facilitate performance evaluation.
R EFERENCES
[1] Y. Saito, A. Benjebbour, Y. Kishiyama, and T. Nakamura, “System level
performance evaluation of downlink non-orthogonal multiple access
(NOMA),” in Proc. IEEE Annual Symposium on Personal, Indoor and
Mobile Radio Communications (PIMRC), London, UK, Sept. 2013.
6
10
Outage Probabilities
−1
10
Solid lines: simulation
Dotted lines: exact expressions
Dash−dotted lines: approximation
−2
10
R =4 BPCU, R =4 BPCU
m’ m
R =1 BPCU, R =4 BPCU
m’ m
R =1 BPCU, R =1 BPCU
−3 m’ m
10
0 5 10 15 20 25 30 35 40
Transmission Power in dBm
30 35 40
Fig. 3. Outage probability P̃m for cognitive radio downlink transmission.
r = 20m, r1 = 10m, r0 = 1m, δ = 1, ρI = 0, and M = N = 2. The
noise power is −30dBm. The analytical results and the approximations are
based on Lemma 3.
[2] Z. Ding, Z. Yang, P. Fan, and H. V. Poor, “On the performance of
non-orthogonal multiple access in 5G systems with randomly deployed
users,” IEEE Signal Process. Letters, vol. 21, no. 12, pp. 1501–1505,
Dec 2014.
[3] 3rd Generation Partnership Project(3GPP), “Study on downlink mul-
tiuser superposition transmission for LTE,” Mar. 2015.
[4] “5G radio access: requirements, concepts and technologies,” NTT DO-
COMO, Inc., Tokyo, Japan, 5G Whitepaper, Jul. 2014.
[5] “Proposed solutions for new radio access,” Mobile and wireless commu-
nications Enablers for the Twenty-twenty Information Society (METIS),
Deliverable D.2.4, Feb. 2015.
[6] S. Timotheou and I. Krikidis, “Fairness for non-orthogonal multiple
access in 5G systems,” IEEE Signal Process. Letters, vol. 22, no. 10,
pp. 1647–1651, Oct. 2015.
[7] J. Choi, “Minimum power multicast beamforming with superposition
coding for multiresolution broadcast and application to NOMA systems,”
IEEE Trans. Commun., vol. 63, no. 3, pp. 791–800, Mar. 2015.
[8] Q. Sun, S. Han, C.-L. I, and Z. Pan, “On the ergodic capacity of MIMO
NOMA systems,” IEEE Wireless Commun. Letters, (to appear in 2015).
[9] Z. Ding, F. Adachi, and H. V. Poor, “The application of MIMO to non-
orthogonal multiple access,” IEEE Trans. Wireless Commun., (to appear
in 2015) Available on-line at arXiv:1503.05367.
[10] K. Higuchi and Y. Kishiyama, “Non-orthogonal access with random
beamforming and intra-beam SIC for cellular MIMO downlink,” in Proc.
IEEE Vehicular Technology Conference, Las Vegas, NV, US, Sept. 2013,
pp. 1–5.
[11] N. Lee, J.-B. Lim, and J. Chun, “Degrees of the freedom of the MIMO
Y channel: Signal space alignment for network coding,” IEEE Trans
Inform. Theory, vol. 56, pp. 3332 – 3343, Jul. 2010.
[12] Z. Ding and H. Poor, “A general framework of precoding design
for multiple two-way relaying communications,” IEEE Trans. Signal
Process., vol. 61, no. 6, pp. 1531–1535, Mar. 2013.
[13] Y. Saito, Y. Kishiyama, A. Benjebbour, T. Nakamura, A. Li, and
K. Higuchi, “Non-orthogonal multiple access (NOMA) for cellular
future radio access,” in Proc. IEEE Vehicular Technology Conference,
Dresden, Germany, Jun. 2013.
[14] Z. Ding, P. Fan, and H. V. Poor, “Impact of user pairing on 5G non-
orthogonal multiple access,” IEEE Trans. Vehicular Technology, (to
appear in 2015) Available on-line at arXiv:1412.2799.
[15] M. Haenggi, Stochastic Geometry for Wireless Networks. Cambridge
University Press, Cambridge, UK, 2012.
[16] J. Venkataraman, M. Haenggi, and O. Collins, “Shot noise models for
outage and throughput analyses in wireless ad hoc networks,” in Proc.
IEEE Military Communications Conference, Washington, DC, USA,
Oct. 2006.
[17] N. Lee and J.-B. Lim, “A novel signaling for communication on MIMO
Y channel: Signal space alignment for network coding,” in Proc. IEEE
International Symposium on Inform. Theory (ISIT-09), Jul. 2009.
[18] Z. Ding, T. Wang, M. Peng, and W. Wang, “On the design of network
coding for multiple two-way relaying channels,” IEEE Trans. Wireless
Commun., vol. 10, no. 6, pp. 1820–1832, Jun. 2011.
[19] Z. Ding, R. Schober, and H. V. Poor, “A general MIMO framework
for NOMA downlink and uplink transmissions based on signal align-
ment,” IEEE Trans. Wireless Commun., (submitted) Available on-line at
arXiv:1508.07433.