You are on page 1of 13

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2017.2744640, IEEE
Transactions on Communications

User Pairing for Downlink Non-Orthogonal Multiple


Access Networks using Matching Algorithm
Wei Liang, Zhiguo Ding, Yonghui Li and Lingyang Song

Abstract— In this paper, we study the user pairing in a coefficients, where the users with poorer channel conditions
downlink non-orthogonal multiple access (NOMA) network, are given more transmission power [4], [9]. The users with
where the base station allocates the power to the pairwise more power allocated are capable of decoding their own
users within the cluster. In the considered NOMA network, a
user with poor channel condition is paired with a user with messages by treating the other users’ information as noise. On
good channel condition, when both their rate requirements are the other hand, the users with less power allocated will employ
satisfied. Specifically, the quality of service for weak users can be the successive interference cancellation (SIC) technique to
guaranteed since the transmit power allocated to strong users is decode their own messages by first removing the other users’
constrained following the concept of cognitive radio. A distributed information from their observations [5], [9]. A cognitive radio
matching algorithm is proposed in the downlink NOMA network,
aiming to optimize the user pairing and power allocation between inspired NOMA networks has been studied from [5], [10],
weak users and strong users, subject to the users’ targeted rate [11], which can be termed as CR-NOMA. The concept of
requirements. Our results show that the proposed algorithm power allocation in CR-NOMA network was proposed in [5],
outperforms the conventional orthogonal multiple access scheme has investigated a more effective approach to allocate power
and approaches the performance of the centralized algorithm, in user pairing-based NOMA systems. In the CR-NOMA
despite its low complexity. In order to improve the system’s
throughput, we design a practical adaptive turbo trellis coded network [5], [11], a user viewed as a cognitive user (CU)
modulation scheme for the considered network, which adaptively with strong channel condition accesses the spectrum occupied
adjusts the code rate and the modulation mode based on the by a user seen as a primary user (PU) with poor channel
instantaneous channel conditions. The joint design work leads condition under a predefined interference constraints. Most
to significant mutual benefits for all the users as well as the of the existing works of the CR-NOMA systems focused on
improved system throughput.
performance analysis, such as the outage probability [5], [12].
Index Terms— Non-orthogonal multiple access scheme, Match- There are very limited studies on the user pair problem in CR-
ing Theory, Adaptive Trellis Turbo Coded Modulation. NOMA. Nevertheless, motivating to maximize the system’s
throughput, the user pairing in CR-NOMA systems leads
I. I NTRODUCTION to combinatorial problem, and is very challenging, thereby
The non-orthogonal multiple access (NOMA) technique becoming NP-hard. In this paper, we employ the matching
emerged as a promising multiple access (MA) technique to im- theory to solve the user pair problem in CR-NOMA network,
prove the access efficiency of the future mobile network [1]– for the sake of finding promising solutions with relatively
[3]. Cooperative communication in conjunction with NOMA small complexity.
scheme was characterized in [4], in which a cooperative Matching theory [13], [14] has been known as an efficient
NOMA scheme was conceived for exploiting the fact that the technique to solve the combinatorial problem of matching
user with better channel condition have the prior information players in two distinct sets, by using the players’ individual in-
about the message for other users. The concept of NOMA have formation and their preferences. The two-side stable matching
been applied in multiple research areas, such as the multiple- problem has been widely investigated from both theoretical
input multiple-output (MIMO) systems [5], high-rate visible and practical perspectives in [13]–[18]. The most classical
light communication (VLC) downlink networks [6], simulta- matching problems are one-to-one matching [17], many-to-
neous wireless information and power transfer (SWIPT) [7] one matching [16] and many-to-many matching [18]. In the
as well as the physical layer security [8]. The key idea of one-to-one matching problem, each player can be matched
the NOMA technique is to explore the power domain for to at most one member from the opposite set. In contrast to
multiple access (MA). More specifically, in a downlink NOMA one-to-one matching, at least one player in the many-to-one
scheme, a base station (BS) can serve multiple users within matching problem can be matched to multiple players in the
the same time/frequency channel via different power allocation opposing set. In the context of many-to-many matching, at
least one player in each of the two sets can be matched to
The work was supported in part by the UK EPSRC under grant number
EP/L025272/1, in part by H2020-MSCA-RISE-2015 under grant number more than one member from the other set. In [18], a many-to-
690750, in part by National Science and Technology Major Project of China many two sided user-subchannel matching algorithm has been
under grant number 2016ZX03001017, and in part by ARC under grant proposed for downlink NOMA networks, aim to optimize the
number DP150104019.
W. Liang and Z. G. Ding are with school of Computing and sub-channel assignment and power allocation problem. In the
Communications, Lancaster University, LA1 4WA, UK. Email: {w.liang, existing literature, there is no study devoted to implement the
z.ding}@lancaster.ac.uk. matching theory on the considered CR-NOMA systems.
Y. H. Li is with school of Electrical and Information Engineering, Univer-
sity of Sydney, NSW, 2006, Australia. Email: yonghui.li@sydeny.edu.au. In this paper, we investigate the user pairing in the CR
L. Y. Song is with school of Electrical Engineering and Computer Science, inspired NOMA networks, where two users within a user pair
Peking University, 100871, China. Email: lingyang.song@pku.edu.cn.

0090-6778 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2017.2744640, IEEE
Transactions on Communications

are allowed to simultaneously transmit information. Specifi- a priori knowledge of the variance of the estimation error.
cally, the paired two users share the same spectrum resources Additionally, we also design a more practical CR-NOMA net-
with the aid of allocating different power levels. The moti- work, which is facilitated by the bandwidth-efficient ATTCM
vation of our considered CR-NOMA is to resist the spectral arrangement in order to improve the system’s throughput
efficiency loss when users’ channel conditions are poor, and performance. The transmission rate/throughput of the system
these weak users1 can be served for the sake of improving is adapted according to the instantaneous channel conditions.
the user fairness. Specifically, the users with strong channel 4) We provide a comprehensive performance analysis for our
condition would have more chances to access the weak users’ CR-NOMA system with/without the practical ATTCM scheme,
spectrum bands which will not case too much performance when employing three user pairing and power allocation
degradation of the weak users. Different to [18], we associate algorithms. Our simulation results show that, in comparison
our user pairing problem with equivalent to two-side one-to- with the OMA assisted CR system, our CR-NOMA system
one matching problem, which aims to increase the individual can always achieve a significant throughput enhancement,
data rates of the paired users as well as the throughput of the regardless of user pairing and power allocation algorithms
overall network, while both the minimum rate requirements employed. Additionally, our proposed conventional distributed
of the PUs and CUs are satisfied. In [17], [18], the time algorithm significantly outperforms the random algorithm,
slots, frequency bands and sub-channel assignments are the and its throughput performance is close to the centralized
exchange proposals. In contrast to the implementations of the algorithm’s. However, when considering imperfect CSI, the
matching theory in NOMA and CR systems, in our paper, the throughput performance of our proposed DMA gets much
CUs negotiate with the PUs regarding the amount of power closer to that of the CA, which reveals that our distributed
coefficients available for them, then these CUs are allowed algorithm is a promising solution to practical CR-NOMA sys-
to access their paired PUs’ spectrum bands. Additionally, we tems. Importantly, we also show that, the CR-NOMA system’s
also develop a more realistic system, wherein considering the throughput can be further improved at a slight cost of using
imperfect channel state information (CSI). Finally, we design the ATTCM scheme.
the practical adaptive coding and modulation schemes based The rest of the paper is organized as follows. The system
on turbo trellis coded modulation (TTCM) for our CR-NOMA model of the CR-NOMA network is outlined in Section II. The
system. general problem of the user pairing and power allocation is
The key contributions of the paper are summarize as fol- discussed in Section III. The proposed distributed approach for
lows: solving the user pairing problem is introduced in Section IV.
1) We conceive a downlink CR-NOMA network, in which A practical adaptive coded modulation scheme is described in
the CUs could transmit simultaneously with the PUs under Section VI. The theoretical performance of the proposed CR-
the quality of service (QoS) constraint that the interferences NOMA scheme is evaluated in Section V and the practical
inflicted by the CUs on the PUs do not degrade the PUs’ performance of the ATTCM aided CR-NOMA scheme is
communication quality. The PUs and CUs are assumed to be investigated in Section VI-B. Finally, our conclusions are
paired with each other in the CR-NOMA network. Particularly, presented in Section VII.
the PU and CU, with two very different channel conditions,
will be paired in each small cluster in order to perform the II. S YSTEM M ODEL
NOMA technique in the power domain, while reducing the
system load. Additionally, the achievable rate of the CUs can
Frequency

PUl CUk
be increased by employing the SIC technique, which leads to User pair
an improved overall system throughput as well. PUl ’s receiver

2) We propose a distributed algorithm for the user pairing and Power


Decode the message

of CUk
power allocation in the CR-NOMA system based on the one-to-
PUl
CUk ’s receiver
one matching theory, where the PUs negotiate with the CUs
for obtaining the appropriate power allocations, aiming to SIC of PUl ’s Decode the message
message of PUl
maximize the system’s throughput. For the sake of comparison,
we develop two extreme benchmarks. The first one with the BS Downlink
CUk

highest complexity demanded, namely centralized algorithm, is


Fig. 1. The system model of user pairing among PUl and CUk in CR-NOMA
to find the optimal pairing of the PUs and CUs by exhaustive networks.
search, while assigning transmit power optimally. Another
called random algorithm, is to randomly pair the PUs and Refer to Fig. 1, each cluster in our CR-NOMA system
CUs and allocate power in a fixed manner. consists of a single BS, as well as the user l and user k.
3) To address practical issues, we investigate the impact of As shown in Fig. 2, the users are assumed to be uniformly
imperfect CSI on the performance of the CR-NOMA network. deployed in a disc, namely Do with the radius ro . The BS
More specifically, we assume the channel estimation error is located at the center of this disc. We further assume that
model, where the BS has an estimation of the channel and this disc is divided into two regions. A smaller disc Di with
radius ri is located inside the disc Do and they have the
1 In the CR inspired NOMA networks, the weak user can be viewed as the
same origin. Specifically, the radius of Di is smaller than
primary user (PU) and the strong user can be seen as the cognitive user (CU) that of the Do , which is ro > ri . Additionally, K users are
[5], [11]. assumed to be randomly distributed in Di . On the contrary,

0090-6778 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2017.2744640, IEEE
Transactions on Communications

L users are uniformly distributed in a region, in which Di is pair. Therefore, the CUk would detect the PUl ’s message when
subtracted from Do , i.e. |Do − Di |. Hence, user l and user hpl < hck , and then remove the message from its observation
k are randomly scheduled and matched with each other. In in a successive manner. Meanwhile, the message of the CUk
the considered downlink CR inspired NOMA system, the L will be treated as noise at PUl .
users who located at the edge of the cell are treated as PUs. Then the achievable rate of PUl is given by [5], [9]:
Specifically, those stronger users which located close to the  
γlP U |hpl |2 αl,k
RlP U = log2 1 + , (3)
1 + γlP U |hpl |2 (1 − αl,k )
BS Note that the transmit
p
signal to noise ratio (SNR)3 of the lth
α P αck PS
PU is γlP U = N l
0
S
and that of the kth CU is γ CU
k = N0 .
R Di
As a result, the achievable rate of CUk can be written as:

R Do RkCU = log2 1 + (1 − αl,k )γkCU |hck |2 , (4)
Fig. 2. The system design of distance setting. Specifically, the PU is willing to perform NOMA with the
BS can be viewed as the CUs. For the sake of improving the CU only if it can achieve a higher rate in comparison with the
spectrum efficiency, those CUs have been squeezed into the conventional orthogonal multiple access (OMA) transmission
spectrum occupied by the PUs. Therefore, the so-called PUs [5].
and CUs have the significant channel condition difference in We set the rate of the PUl using conventional OMA-CR as
our considered system. Without loss of generality, the channels the minimum rate requirement, namely Rl,req PU
, and it can be
are sorted as 0 < |hpl |2 < · · · < |hpL |2 < |hck |2 < · · · < |hcK |2 , expressed as:
where hpl defines the channel from BS to the lth PU and hck
1
is the channel form the BS to the kth CU. We assume that PU
Rl,req = log2 (1 + γlP U |hpl |2 ) , (5)
the CUs and PUs can transmit simultaneously, and each PU 2
can be paired with one particular CU. The paired PU and where the factor 21 of Eq.(5) is due to the fact that conventional
CU can transmit information on the same spectrum, but users OMA results in a multiplexing loss of 12 . Specifically, PUs and
from different pairs should transmit on orthogonal channels. CUs are served in a TDMA manner for conventional OMA-CR
Additionally, the pathloss is defined as ̺ = 1/daab [19], where which has a multiplexing loss of 21 for the PUs. Additionally,
dab is the geometrical distance between node a and node b, the minimum rate constraints of PUl is defined as RlP U ≥
PU
and the path-loss exponent considered in our system is a = 4. Rl,req , which can be re-written as:
Similar to [20], [21], all channels in our system are assumed 
γlP U |hpl |2 αl,k

1
to impose quasi-static Rayleigh fading, where the channel log2 1 + P U p 2 ≥ log2 (1 + γlP U |hpl |2 ) . (6)
1 + γl |hl | (1 − αl,k ) 2
coefficients are constant for each transmission block, but vary
independently between different blocks and all the CSI of the After that, the power allocation fraction of CUk can be derived
users are assumed to be known at BS. as:
q
The signal received at PUl is given by 1 + γlP U |hpl |2 − 1
q p (1 − αl,k ) ≤ . (7)
ylP U = hpl ( αlp PS xpl + αkc PS xck ) + npl , (1) γlP U |hpl |2
and the signal received at CUk is expressed as: Moreover, we have a minimum rate requirement of the CUk ,
CU
q p namely Rk,req , in order to make sure CUk can benefit from
ykCU = hck ( αlp PS xpl + αkc PS xck ) + nck . (2) accessing the PU’s spectrum band. Furthermore, the minimum
wherepPS is the transmit power emitted from the BS. The rate requirement of each CU is dependent on its channel
p
term αlp PS xpl + αkc PS xck of Eq. (1) and Eq. (2) represents conditions.
the combined signal of PUl and CUk emitted from the BS.
The channels hpl and hck are assumed to be Rayleigh dis- III. F ORMULATION OF THE O PTIMIZATION P ROBLEM
tributed, following the complex-valued Gaussian distribution Let us define a (L× K)-element matching matrix M. When
of CN (0, σ 2 ). In Eq. (1) and Eq. (2), npl and nck are the PUl is matched to CUk , their corresponding matching matrix
Gaussian noise terms, which have a zero mean and a noise entry is given by ml,k = 1, otherwise ml,k = 0. We also
variance of N0 /2 per dimension. Additionally, αlp and αkc define a (L × K)-element power allocation matrix P with
represent the power coefficients for PUl and CUk , and we elements αl,k . In particular, when given the users’ predefined
have αkc + αlp = 1. For simplify, we use the notation αl,k QoS requirements, the objective function of maximizing the
to represent the power allocation coefficient, and we define system throughput are formulated as:
αl,k = αlp . Then, the power allocated to the CU is (1 − αl,k ).
L X
K
The SIC technique2 is only employed at the CU within each X
{Mcen , Pcen } = max ml,k W (αl,k ) , (8)
2 The M,P
SIC is applied at the user device in the considered CR inspired l=1 k=1
NOMA system, in order to cancel the multi-user interference. Therefore, in
our proposed framework, only two users within each pair share the same 3 The definition of the transmit SNR is unconventional, because it relates the
spectrum. Additionally, the SIC is only employed by the user with strong transmit-power to the noise-power at the receiver, which are quantities mea-
channel condition. Hence, the complexity of implementing SIC is just one sured at different locations. Nonetheless, this convenient definition simplifies
and is affordable at user device. our discussion, as proposed in [22].

0090-6778 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2017.2744640, IEEE
Transactions on Communications

s.t. (a) RlP U (αl,k ) ≥ Rl,req


PU
, ∀l∀k their corresponding matching matrix entry can be defined as
(b) RkCU (1 − αl,k ) ≥ CU
Rk,req , ∀l∀k mi,j . Then the optimization value of αl,k can be formulated
as:
(c) 0 ≤ αl,k ≤ 1 , ∀l∀k
cen
L
X αl,k = max W (αl,k )mi,j , (12)
αl,k
(d) ml,k ≤ 1 , ∀k
l=1 where mi,j satisfies the matching constraints in (d), (e) and
XK (f) of Eq. (8).
(e) ml,k ≤ 1 , ∀l According to condition (a) of Eq. (8), the lower bound of
k=1 αl,k is given by
(f ) ml,k ∈ {0, 1} , ∀l∀k q
K
L X γlP U |hpl |2 + 1 − γlP U |hpl |2 + 1
X min
(g) Nf = ml,k .∀l∀k αl,k ≥ . (13)
γlP U |hpl |2
l=1 k=1
Moreover, from condition (b) of Eq. (8) the upper bound of
The conditions (a) and (b) ensure that the minimum rate αl,k can be given by
requirements of the PUs and CUs can be achieved. The
CU
condition (c) specifies the range of the power allocation factor max 1 + γkCU |hck |2 − 2Rk,req
of the PU. Therefore, conditions (d) and (e) ensure that each αl,k ≤ . (14)
γkCU |hck |2
PU (or CU) will only be matched to one CU (or PU). The
condition (f) states that the corresponding matching matrix By considering the three cases of the optimization problems
entry is either 1 or 0. Conditions (g) defines the total number as described in Eq. (9), Eq. (10) and Eq. (11), the setting of
cen
of available frequency bands Nf is equal to the number of αl,k in Eq. (12) are derived as:
PUs. Without loss of generality, we assume that each PU • Case 1:
occupies a single spectrum band. Moreover, the choice of the CU
cen 1 + γkCU |hck |2 − 2Rk,req
weight W (αl,k ) of Eq. (8) depends on the objective function αl,k = . (15)
to be optimized. We consider three cases to set up the relative γkCU |hck |2
optimization functions, which are described below: • Case 2:
q
• Case 1: objective function is to maximize the sum-rate γlP U |hpl |2 + 1 − γlP U |hpl |2 + 1
of matched PUs: cen
αl,k = . (16)
γlP U |hpl |2
WP U (αl,k ) = RlP U (αl,k ) . (9)
• Case 3:

• Case 2: objective function is to maximize the sum-rate CU
Rk,req
 1+γkCU |h
 c 2
k | −2 ∗
CCU < CP∗ U ;
of matched CUs: cen
αl,k =
CU
γk |hk |√c 2
p 2
PU
γlP U |hp 2

WCU (αl,k ) = RkCU (1 − αl,k ) . (10)  γl |hl | +1−



PU p 2
l | +1 ∗
CCU > CP∗ U ;
γl |hl |
(17)
• Case 3: objective function is to maximize the total sum-
rate of matched PUs and CUs: The derivation of Eq. (17) is detailed in Appendix A. After
cen
obtaining the optimization value of αl,k of Eq. (12), the
Wtotal (αl,k ) = RlP U (αl,k ) + RkCU (1 − αl,k ) . (11) optimization function of Eq. (8) can be re-written as :
Note that, the objective function of Eq. (8) is a function X K
L X

of continuous variables αl,k and binary variables ml,k . The {Mcen } = max W (αcen
l,k )ml,k , (18)
M
objective function in Eq. (8) can be decoupled with respect to l=1 k=1

two entries, which are the optimal power allocation matrix P


L
and the optimal matching matrix M [23]. Hence we solve this X
s.t. (d) ml,k ≤ 1 , ∀k
optimization problem in two steps in Section III-A.
l=1
XK

A. Decoupling the Optimization Function (e) ml,k ≤ 1 , ∀l ,


k=1
For a given matched pair (PU, CU), mapping is valid (f ) ml,k ∈ {0, 1} ∀l∀k .
only if the conditions (a) and (b) of Eq. (8) have been
cen cen
satisfied. During the first step, the optimum value, αl,k , and its The weight W (αl,k ) can be solved according to the terms
cen
corresponding weight W (αl,k ) for each pair (PU, CU) can be PU cen PU
Rl (αl,k ), Rl,req , RkCU (αl,k
cen CU
) and Rk,req . These four terms
found under the conditions (a), (b) and (c) of Eq. (8), where the can be calculated based on the known channel conditions. This
cen
computations of the achievable rate of the PU and the CU have follows the fact that the weight W (αl,k ) only depends on
cen cen
been introduced in Section II. Note that all W (αl,k ) depend the variable αl,k . Specifically, αl,k is the maximum value of
cen
on a single variable αl,k and then maximizing the individual variable αl,k . In a given matching pair, αl,k can be obtained
W (αl,k ) leads to the maximal value of the overall objective. from Eq (15), Eq. (16) and Eq. (17) according to different
Suppose we have a particular matched pair (PUi , CUj ), where optimization objectives. Note that, the optimal objective of

0090-6778 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2017.2744640, IEEE
Transactions on Communications

cen
Eq. (18) is W (αl,k )ml,k , which is a specific example of the list. When a CU transmits on the spectrum band occupied by
formulation shown in Eq. (8), where W (αl,k ) of Eq. (8) is the preferred PU, its achievable transmission rate should be
cen CU
substituted by W (αl,k ) of Eq. (18). Therefore, the solution for higher than its minimum rate requirement Rk,req . Thus, the
Eq. (18) is a feasible solution for Eq. (8), which is equivalent preference list for CUk is given by:
to find the best matching pair.  L′
CU LISTk = P Uι(l) l=1 , (21)

B. Centralized Algorithm where the function ι(l) satisfies the following condition:
CU CU
In order to find the optimal solution, we use the centralized Rι(l),k (αι(l),k ) > Rk,req , l ∈ (1, . . . , L) . (22)
algorithm (CA). We consider all possible matching pairs of
the PUs and CUs, and then select that particular matched pair, Again, the elements in CULISTk are ranked in decreasing
which has the maximum sum-rate according to our intended order in terms of the achievable rate. L′ in Eq. (21) is the
objective functions as shown in Eq. (9), Eq. (10) and Eq. (11). number of PUs in CU LISTk , i.e. L′ ≤ L. Additionally,
By referring to Eq. (18), the centralized solution requires an the PUs and CUs would exchange their roles in the Case 2
exhaustive search over all possible matching pairs and the scenario. Specifically, the CUs would make offers to the PUs
power allocation combinations. Specifically, the CA relies on in the Case 2 scenario. In the Case 3, the PUs and CUs would
an exhaustive search method that imposes the highest number compare first, and then decide which one to be the proposers.
of operations for the sake of finding the optimum solution. This is based on the optimization results of Eq. (11).
Therefore, the CA has the highest complexity. Additionally,
The amount of complexity required for this CA increases with B. Conventional Distributed Matching Algorithm (DMA)
the number of users, which may become quite high, rendering
it impractical. Therefore, we will propose a distributed ap- Different to the CA as described in Section III, we motivate
proach in association with the matching algorithm as described to use the matching theory to find a suboptimal solution of
in Section IV. Eq. (8). In the context of the CA, the BS needs to know
all the CSIs of all the PUs and CUs. Additionally, in our
proposed DMA, there is no central controller and does not
IV. T HE D ISTRIBUTED A PPROACH require the BS to know the full CSI of users, which reduces
A. Preference Lists the system complexity. Instead, the BS only needs to have
the rank information of channels, and is able to broadcast
Our problem can be modeled as a two-side one-to-one this information to PUs and CUs for operating distributed
matching problem [13], [14], in which a set of PUs will be user paring and power allocation. Then, PUs and CUs can
matched with a set of CUs. We consider a canonical matching construct their preference list without any CSI. The key idea
problem [15], where the preferences of one player set depends of this distributed algorithm in Case 1 scenario is that each
on the other player set. Specifically, before the information PU make an offer to its most preferred CU from its preference
exchange (proposals), each player is required to merely collect list. Then each CU has right to accept or reject these offers.
the information from the opposite set users they are interested, When all PUs make their offer once, one round of proposals
and performs a ranking according to its preferences. In the is preformed. In order to construct the user pairs among the
considered CR-NOMA system, the BS broadcasts the ranking PUs and CUs in the considered CR inspired NOMA networks,
information of the channels to all the PUs and CUs. According the PU having a poorer channel condition matches with the
to the objective functions presented in Eq.(9), Eq.(10) and CU having a better channel condition. Then we distinguish the
Eq.(11), we treat one sets of users as the proposers and the matched PUs and CUs in the power domain, aiming to improve
opposite sets of users as selectors. The users who to be the the total sum-rate of the PUs and CUs. Additionally, each CU
proposers or/and selectors depend on the consideration of the and its paired PU transmit simultaneously in appropriate dif-
objective functions. In case 1 of Eq.(9), PUs are act as the ferent power levels by exchanging their information. The BS
proposers and CUs are the selectors, then before any offer is allocates the initialized power allocations to the PU and CU,
made to the CUs, each of the PUs construct a preference list which the PU would get more power than the CU. After that,
of CUs, which can satisfy the PU’s rate requirement. Thus, the PUs and CUs negotiate with each other for a specific value
the preference list for PUl is given by: of power, i.e. αl,k , as introduced in Section III. Employing the
 K ′ DMA for efficiently representing the interaction among the
P U LISTl = CUκ(k) k=1 , (19)
PUs, where each PU chooses its allocation independently of
where the function κ(k) satisfies the following condition: the others.
PU
Rl,κ(k) PU
(αl,κ(k) ) > Rl,req , k ∈ (1, . . . , K) . (20) Additionally, we first initialize the power allocation factor of
PUl as αl,k = αinit and set the value of the power step counter
The indexes of the CUs are recorded in the PULIST, when to τ . We construct P U LISTl and CU LISTk based on the
their corresponding rate have satisfied the PU’s rate require- initialized value αinit . As an example, an offer of αl,k is given
ment. K ′ in Eq. (19) is the number of CUs in P U LISTl , by PUl to the CUk in the top of its preference list PULISTl .
where K ′ ≤ K. Additionally, we have assumed that the first Then this CUk grants a power allocation factor 1 − αl,k for
CUκ(k) at the top of the P U LISTl provides the highest rate itself. When CUk receives an offer, it has two options: (1)
PU
Rl,κ(k) (αl,κ(k) ). Similarly, each CU also has its preferred PU reject the offer if PUl is not in CUk ’s CULISTk ; (2) accept

0090-6778 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2017.2744640, IEEE
Transactions on Communications

TABLE I TABLE II
C ONVENTIONAL D ISTRIBUTED M ATCHING A LGORITHM (DMA) IN C ASE T HE A LGORITHM FOR C ASE 2 SCENARIO
1
Step 1: Initialization
Step 1: Initialization
1) Set αl,k = αinit , and set τ .
1) Set αl,k = αinit , and set τ . 2) Construct CULISTk based on (1 − αinit ), k ∈ K.
2) Construct PULISTl based on αinit , l ∈ L. 3) Construct PUN OT M AT CH to list the PUs which are not matched,
3) Construct PUN OT M AT CH to list the PUs which are not matched, yielding PUN OT M AT CH = {P U1 , . . . , P UL }Ll=1 .
yielding PUN OT M AT CH = {P U1 , . . . , P UL }L 4) Construct CUN OT M AT CH to list the CUs which are not matched,
l=1 .
4) Construct CUN OT M AT CH to list the CUs which are not matched, yielding CUN OT M AT CH = {CU1 , . . . , CUK }K l=1 .
yielding CUN OT M AT CH = {CU1 , . . . , CUK }K 5) Construct a list of all CUs which are matched, yielding
l=1 .
5) Construct a list of all PUs which are matched, yielding CUM AT CH = {CUM AT CH (k)}K , where CUM AT CH (k) =
P UM AT CH = {P UM AT CH (l)}L , where P UM AT CH (l) = 1 1 if the CU is matched to the PU, otherwise set CUM AT CH (k) =
if the PU is matched to the CU, otherwise set PUM AT CH (l) = 0. 0.
6) Set l = 1 and k = 1. 6) Set k = 1 and l = 1.

Step 2: The PUs negotiate with the CUs: Step 2: The CUs negotiate with the PUs:
• Step 2A: PU selects its current best CU: PUl offers αl,k = αinit • Step 2A: CU selects its current best PU: CUk offers 1 − αl,k =
to the top CU from its PULISTl . αinit to the top PU from its CULISTl .
1) If this PUl is in CULISTk of CUk : 1) If this CUk is in PULISTl of PUl :
– PUl and CUk are temporal matched. – CUk and PUl are temporal matched.
– Remove PUl from PUN OT M AT CH . Update the lth el- – Remove CUk from CUN OT M AT CH . Update the kth
ement of PUM AT CH to PUM AT CH (l) = 1. Remove element of CUM AT CH to CUM AT CH (k) = 1. Remove
CUk from CUN OT M AT CH . PUl from PUN OT M AT CH .
2) Else: 2) Else:
– Update the value of power as α∗l,k = αl,k − τ . – Update the value of power as α∗l,k = αl,k − τ .
– Update PULISTl based on α∗l,k , then go back to Step 2A – Update CULISTk based on (1 − α∗l,k ), then go back to Step
and set PUM AT CH (l) = 0. Remove PUl from the set 2A and set CUM AT CH (k) = 0. Remove CUk from the set
PUN OT M AT CH , if PULISTl = ⊘ and go to Step 3. CUN OT M AT CH , if CULISTk = ⊘ and go to Step 3.
• Step 2B: If this preferred CUk has a current matched pair, PUcur . • Step 2B: If this preferred PUl has a current matched pair, CUcur .

1) if (1 − αl,k ) > (1 − αlcur ,k ), this CUk would discard its 1) if αl,k > αlcur ,k , this PUl would discard its current matching
current matching PUcur . Then PUl and CUk are matched and CUcur . Then CUk and PUl are matched and the CUcur goes
the PUcur goes to Step 2A-(2). to Step 2A-(2).
2) Else: The PUl goes to Step 2A-(2). 2) Else: The CUk goes to Step 2A-(2).

Step 3: If no further matching is required Step 3: If no further matching is required


Then go to Step 4, otherwise go to Step 2 for the next PU, l = l + 1. Then go to Step 4, otherwise go to Step 2 for the next CU, k = k + 1.
Step 4: End of the the algorithm. Step 4: End of the the algorithm.

it, when this PUl is in the CULISTk , creating a matching described in Table II. Additionally, when we consider the
pair between PUl and CUk . Additionally, if two or more PUs Case 3 scenario, the PUs and CUs would take turn to be the
send offers to CUk , a conflict will happen [16]. In order to proposers depending on their current avenue as shown in Table
avoid this conflict, CUk will only pair with the PU who could III. Note that, we have some particular situations that two
provide a higher revenue. Specifically, if this intended CUk players may compete with each other for appealing the same
has already been matched to PUcur , and this PUcur failed to choice. During the competition, each player reduces its own
provide a lower power allocation factor to the CUk , i.e. power allocation factor, and this competition completes when
one of the user’s achievable rate is inferior to its minimum
αlcur ,k > αl,k , (23) rate requirement. Then we allow the losing user to select its
then CUk discards its current matching in favor of the new surplus best choice from its current preference list.
matching. This is because a larger value of αl,k leads to a
lower achievable rate of the CU, which is refer to Eq. (4). V. P ERFORMANCE A NALYSIS OF THEORETICAL
Moreover, the rejected PUcur will update the value of the CR-NOMA NETWORK
power allocated factor by setting it to
A. The Stability of our Conventional DMA
αl∗cur ,k = αlcur ,k − τ . (24)
Before introducing the definition of stable matching, we
and then it will reconstruct its preference list based on αl∗cur ,k would like to highlight our related notations and definitions.
and repeat the matching process. This algorithm aims to find We denote the set of PUs’ indexes by L = {1, . . . , l, . . . , L}
the specific power allocation factors, which can be accepted and the set of CUs’ indexes by K = {1, . . . , k, . . . , K}.
by both the PUl as well as the CUk , and the algorithm is Definition 1: We define a matching µ as a one-to-one matching
terminated when each PU has found its appropriate matched of the set L ∪ K. We refer to µ(l) and µ(k) as the partner of l
partner, provided that both their rate requirements can be and k, respectively. Then we have µ(l) ∈ K and µ(k) ∈ L [14,
satisfied. Similar to the DMA of Case 1 as shown in Table I, p. 19].
when we consider the Case 2 scenario, the CUs become Definition 2: A matching µ is deemed to be blocked by an
the proposer and the PUs are the selectors. The details are individual P Ul (or CUk ), where we have l ∈ L (or k ∈ K).

0090-6778 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2017.2744640, IEEE
Transactions on Communications

TABLE III
neither PUl nor CUq blocks the matching µ, hence the DMA
T HE A LGORITHM FOR C ASE 3 SCENARIO
is stable.
Step 1: Initialization
16
1) Set αl,k = αcen
l,k of Eq. (17), and set τ .
L=6
∗ L=4
2) Compute CCU of Eq. (17).

Average throughput of matched PU


14 L=2
3) Compute CP ∗
U of Eq. (17).
12
Step 2: Select the players to act as the proposer.
∗ ∗ . 10
• Compare the value of CCU and CP U

• if CCU < CP ∗ , Select the PUs to act as the proposers and use the
U 8
DMA in Case 1 of Table I.
• Else, select the CUs to act as the proposers and use the DMA in
6
Case 2 of Table II.
• Set l = l + 1, k = k + 1. 4

until converge
Step 3: End of Algorithm 2
0 20 40 60 80 100 120 140 160 180 200
Number of iterations

Fig. 3. The performance of average throughput of matched PUs versus the


The P Ul (or CUk ) would rather prefer not to be matched, number of iterations by employing the conventional DMA in the CR-NOMA
systems.
instead of being matched to its current partner candidate under
µ [14, p. 20], when a low resultant rate for P Ul (or CUk ) Fig. 3 shows the average throughput of matched PUs versus
under the current matching µ fails to satisfy their minimum the number of iterations N for the DMA in CR-NOMA
rate requirements at a given power allocation factor αl,k . This system. It implies that the DMA converges to a certain value
mathematically implies that: after a finite number of iterations. From Fig. 3, it can be noted
PU
that the DMA converges to the identical throughput of Rmatch
PU PU at L = 2, L = 4 and L = 6. Furthermore, we can observe that
Rl,µ(l) (αl,µ(l) ) < Rl,req or
CU CU the number of iterations required for converge sightly increases
Rµ(k),k (αµ(k),k ) < Rk,req . (25)
as L gets bigger. That is because, the DMA would use more
Definition 3: The pair (P Ul , CUk ) is referred to as a blocking iterations to process the matching procedure between the PUs
pair for the matching µ, if both P Ul and CUk would prefer to and CUs, while we have more number of user pairs. Therefore,
be matched with each other, but they are not matched under the we can imply that, our DMA requires low implementation
current matching µ [14, p. 21]. Both P Ul and CUk achieve complexity, which will be further analyzed in Section V-B.
a higher rate if they are matched in comparison with their
current matching partners under µ at a given power-allocation B. Complexity Analysis
factor αl,k . This mathematically implies that:
The complexity required by the CA and the conventional
PU PU
Rl,k (αl,k ) > Rl,µ(l) (αl,µ(l) ) and DMA will be discussed in this section in terms of the number
CU
Rl,k CU
(1 − αl,k ) > Rµ(k),k (1 − αµ(k),k ). (26) of operations [24]. The optimal CA requires the exhaustive
search. All possible matching pairs over PUs and CUs are
Definition 4: A matching µ is stable if it is not blocked by searched, and then the most profitable matched pair is selected
any individual or any pair [14, p. 21]. according to the objective function as discussed in Eq. (8) of
Based on Definition 1, 2, 3 and 4, we now proceed to prove Section III. The complexity of the CA can be expressed as:
that our DMA as shown in Section IV-B constitutes a stable O L! 2L+K

, (27)
matching.
Let us assume that the final matching result provided by our where the L! of Eq. (27) is the total number of matching
DMA is µ. There are no blocking individuals under µ because combinations between the PUs and CUs for the case of
all users in each preference list (Eq.(20) and Eq.(22)) satisfy L = K. Additionally, the complexity of solving the linear
the corresponding minimum rate requirements. Furthermore, programming problem for all possible matching combinations
we assume that P Ul and CUq (q ∈ K) are not matched with is given by: 2L+K [25].
each other under µ, but PUl prefers to be matched with CUq For the conventional DMA, the PUs and CUs negotiate with
instead of its current partner CUµ(l) under µ. This illustrates each other for transmitting power as discussed in Section IV-
that CUq is acceptable to P Ul , hence P Ul must have made B. In the DMA, the PUs make the offers to the CUs, which
an offer to CUq before asking its current partner CUµ(l) (or then decide either to accept or reject the offers. The matched
before being rejected by all its acceptable partners). Since P Ul CU rejects its current matching in favor of a new matching,
was not matched to CUq , when the algorithm was terminated, whenever the new matching pair is capable of providing a
this implies that P Ul must have been rejected by CUq , where higher rate for this CU. In that case, the rejected PU updates its

CUq is in favor of P Ut (t 6= l, t ∈ L), which offers a better power allocation factor according to αlcur ,k = αlcur ,k −τ . The
rate compared to that offered by P Ul , as detailed in Step-2B worst case happens for a PU is that, it updates its preference
of Table I. Therefore, CUq is matched to its current partner list ( αinit −α
τ
limit
) times until it finds a final matched partner.
P Ut , where t = µ(q), which is better than PUl . As a result, The αlimit is the tolerable minimum power allocation factor

0090-6778 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2017.2744640, IEEE
Transactions on Communications

that can be accepted by this PU. Note that, the PU’s achievable can be used in large CR-NOMA systems. In our simulations,
rate is less than its minimum rate requirement when αl,k < we assume that the transmit SNRs of all CUs are equal, and set
αlimit . Let us consider the most worst case which requires the γ1CU = · · · = γkCU = 5 dB. We also assume that the SNRs of
highest complexity. A PU has K CUs in its preference list. all PUs are γ1P U = · · · = γlP U = 3 dB. The rate requirement
This PU offers ( αinit −α
τ
limit
) times for each of the (K − 1) of the CUs are assumed to Rk,req CU
= 2. Fig. 4 shows the
CUs in its preference list. Then, the PU finds a final matched average sum-rate of the matched PUs versus the number of
CU when the PU makes the last offer to the CU in the bottom PUs in the CR-NOMA system, when L = K = {2, 3, . . . , 10}.
of its list. For this extreme case, the number of operations The average sum-rate of the matched PUs is defined as:
required should be (αinit −αlimit2τ
)×K(K−1)
, which corresponds X K
L X
2
to the complexity of O = {K }. Therefore, the complexity of PU
Rmatch = ml,kmatched × RlP U (αl,kmatched ) , (29)
the DMA is given by l k

O LK 2

. (28) where αl,kmatched represents the final value of power allocated
to the matched PU. In our simulations, we evaluate the
It can be seen from Eq. (27) and Eq. (28) that, the CA has a bandwidth normalized throughput of the systems. In Fig. 4 we
significantly higher complexity compared with the DMA, since compare the performances achieved by the three algorithms,
the complexity of CA increases exponentially by the number which include the DMA of Section IV-B, the CA described
of PUs or CUs. in Section III-B and another benchmark, namely random
algorithm (RA). In the context of the RA, each PU makes
an offer αl,k to a CU, which is randomly selected from its
C. Rate Analysis preference list. The CU chooses the specific matching pair
that provides an increased rate to itself and discards the one
25
CA
having a lower rate. More specifically, the αl,k is chosen in
DMA order to maximize the optimization weight. In Fig. 4, we have
Average throughput of matched PU

RA
20 OMA considered the scenario for case 1 of Eq. (9) as introduced
in Section III. In our simulations, we used the optimal power
15 allocation factor for the RA according to different objective
functions as shown in Eq. (9), Eq. (10) and Eq. (11).
10
In Fig. 4, we observe that the CA achieves the highest
average sum-rate among these three algorithms, while the
5
RA achieves the lowest sum-rate. Furthermore, as seen from
Fig. 4 that our CR-NOMA system always achieve a higher
0
2 4 6 8 10 sum-rate than that of the OMA system, regardless of the
Number of PUs, L=K
user pair and power allocation approaches used. Specifically,
when L = 6, for the DMA under the CR-NOMA system, the
Fig. 4. The performance of the average throughput of matched PUs versus
the number of PUs in the CR-NOMA system over Rayleigh fading channel.
average throughput of matched PUs is 12.5 − 7.5 = 5 bit-
The objective function of case 1 is considered, which is described in Eq. (9) per-second-per-Hz (bit/s/Hz), which is higher than that of the
of Section III. The “CA”, “DMA” and “RA” techniques are detailed in OMA scheme. Moreover, we find that our CR-NOMA system
Section IV-B. The transmit SNRs of the CUs are γ CU = 5 dB and the
transmit SNRs of the PUs are γ P U = 3 dB.
achieves a higher performance gain over the OMA system
as the value of L increases. Even when the RA is employed
70
under the CR-NOMA, it still outperforms the OMA case, such
CA
DMA
as 9.2 − 7.5 = 1.7 bit/s/Hz can be obtained at L = 6.
60 RA
OMA Additionally, Fig. 5 shows the average throughput of
Case 3
50 matched PUs and CUs as well as the sum-rate of matched
Average throughput

Case 2

40
CUs versus the number of PUs in the CR-NOMA system. In
Fig. 5, we consider the performances for the two optimization
30
objective functions, which are the case 2 of Eq. (9) and case 3
20 of Eq. (11) as described in Section III. The average sum-rate
of matched CUs is formulated as:
10
X L X K
CU
0
2 4 6 8 10
Rmatch = ml,kmatched × RkCU (1 − αl,kmatched ) . (30)
Number of PUs, L=K l k
Then the average sum-rate of matched PUs and CUs is given
Fig. 5. Performance of the average throughput versus the number of PUs in by:
the CR-NOMA system over Rayleigh fading channel. The objective functions P L PK
total
Rmatch = k ml,kmatched ×
of case 2 and case 3 are considered, which are described in Eq. (10) and (31)
l
Eq. (11) of Section III.  PU 
Rk (αl,kmatched ) + RkCU (1 − αl,kmatched ) .
In our current design work, we have considered multiple
pairs of users, which the maximum number of pairs is M = As shown in Fig. 5, for both two optimization objective
10. Note that, both our proposed algorithms are scalable and functions, maximizing the average throughput of matched PUs

0090-6778 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2017.2744640, IEEE
Transactions on Communications

and CUs, and maximize the throughput of matched CUs, the the DMA, and that of the RA is the worst. Hence, the rate-
CA achieves the best performance. We observe from Fig. 5 improvement observed in Fig. 4 is linked to a higher Pmatch ,
that, the performance of our DMA is close to that of the resulting from a better matching. Therefore, Pmatch of the
CA, which implies the DMA provides a promising suboptimal DMA is identical to that of the CA when L = K = 5.
solution to the problem in Eq. (8). However, the RA achieves Additionally, Pmatch of the RA achieves a average value,
the worst performance among the three algorithms in the which is Pmatch = 0.8 for all different numbers of the PUs
CR-NOMA system. Once again, Fig. 5 illustrates our CR- and CUs. Moreover, the percentage of matched users found by
NOMA system can outperform the OMA system under all the DMA is seen to be significantly higher than that of the RA.
cases considered. For example, in Fig. 7, the DMA achieves Pmatch = 99.99%
at K = 10, while the RA only achieves Pmatch = 80%. Thus
1.0 there is an approximately Pmatch = 99.99% −80% = 19.99%
l,k

difference between these two algorithms.


The average power allocation factor

0.9

E. Imperfect channel estimation

0.8
25
CA-Perfect
= 0.05 CA-Imperfect w=15dB

Average throughput of matched PU


= 0.10 CA-Imperfect w=2dB
= 0.15 20
DMA
= 0.20
0.7
2 4 6 8 10
Number of PUs, L=K
15

Fig. 6. The performance of the average power allocation factor αl,k versus
10
the step size of the power allocation factor τ in the CR-NOMA system.
In Fig. 6, we investigate how the step size τ affects the
5
power allocation, i.e. αl,k . As expected, the power allocated
factor αl,k was decreasing with increasing τ . However, a small 2 4 6 8 10

τ would lead to a longer matching period. Hence, we have Number of PUs, L=K

chosen τ = 0.05 as a compromise in our simulations.


Fig. 8. Performance of the average throughput versus the number of PUs
of CR-NOMA system by implementing the imperfect channel estimations,
associate with the error estimation factor w = 2dB and w = 15dB.
D. Relative Frequency of Successfully matched PUs
However perfect CSI estimation is not realistic in practice,
especially in a mobile communication system having time-
Relative-frequency of successfully matched pairs

1.0 varying wireless channels. In this section, we will investigate


the performance of the considered NOMA-CR network, when
the CSI estimation is impaired by estimation errors. We
0.9
employ the widely used assumptions [26], [27] in which the
feedback to the transmitter is instantaneous and error free.
More specifically, the CSI is also achievable at the transmitter
0.8
whatever CSI the receiver has. In the case of imperfect
channel knowledge, the estimate of the channel gain matrix
CA H̃ is constructed by the channel gain matrix H modeled
DMA
RA by independent zero-mean complex-valued Gaussian random
0.7
2 4 6 8 10 variables and the channel error matrix Ñ that has a variance of
Number of PUs, L=K
var(Ñ ). The estimation of the channel gain matrix is written
as [28]:
Fig. 7. Pmatch versus L performance of the CR-NOMA system over
Rayleigh fading channel. The “CA”, “DMA” and “RA” techniques are
detailed in Section IV-B. The transmit SNRs of the CUs are γ CU = 5 dB H̃ = H + Ñ . (33)
and the transmit SNRs of the PUs are γ P U = 3 dB.
Fig. 7 illustrates the relative frequency of successfully The channel H is defined in Eq. (1) and Eq. (2). The
matched PUs, i.e. Pmatch , versus the number of PUs, and channel estimation error Ñ is assumed to be independent
Pmatch is evaluated as: identical Gaussian distributed obeying the standard Gaussian
" L K # distribution of zero mean. The degree of CSI estimation errors
1 XX
is governed by the Channel Estimation Factor (CEF), namely
Pmatch = E ml,k , (32)
L w (dB) defined by:
l k

where E[.] denotes the expected value of [.]. As shown in 1


w = 10 log10 [dB] . (34)
Fig. 7, Pmatch of the CA is the highest, followed by that of var(Ñ )

0090-6778 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2017.2744640, IEEE
Transactions on Communications

10

Then the received signal of our considered NOMA-CR system to that of binary turbo codes, but each decoder alternately
impaired by imperfect channel estimation is rewritten as: processes its corresponding encoder’s channel-impaired output
symbol, and then the other encoder’s channel-impaired output
Y = H̃X + n
symbol [29, p.764]. More details on the ATTCM principles
= (H + Ñ )X + n . (35) can be found in [29].
However, we emphasize that, regardless of the estimation We have employed a ATTCM scheme for the CR-NOMA
techniques employed, the estimation error is actually non- system, where the information Bit-per-Symbol (iBPS) is se-
Gaussian, hence using a Gaussian model constitutes an ap- lected from the set iBPS= {0, 1, 2, 3, 5} Bit-per-Symbol (BPS)
proximation. The simulation results of our proposed scheme corresponding to the case of no transmission, QPSK, 8PSK,
subjected to CSI estimation channel errors are discussed 16QAM and 64QAM modulations, respectively. Moreover, the
in the following. Fig. 8 shows a comparison between the ATTCM mode switching thresholds Υ =[γ0 , γ1 , γ2 , γ3 ] were
perfect channel estimation for the average throughput and the determined based on the BER performance curves of each of
imperfect channel estimation results associated with the CSI the four TTCM schemes in a Rayleigh fading channel, which
estimation errors of w = 2dB and w = 15dB. It is evident that is shown in Fig. 9. Specifically, the ATTCM mode switching
the average throughput under the perfect CSI is higher than operation and the throughput of the modes are specified by
that of imperfect CSI scenarios, when the other parameters the following algorithm:

stay the same. Furthermore, the average throughput of PU  γ3 ≤ γR , TTCM-64QAM, iBPS=5 BPS;

achieved by the CA gets worse under the imperfect CSI, when 



 2γ ≤ γ R < γ 3 , TTCM-16QAM, iBPS=3 BPS;
the value of the error estimation factor decreases, i.e. when
error estimation becomes more severe. It is worth noting that, MODE = γ1 ≤ γR < γ2 , TTCM-8PSK, iBPS=2 BPS;


when considering imperfect CSI, the throughput performance 

 γ0 ≤ γR < γ1 , TTCM-4PSK, iBPS=1 BPS;

of our proposed DMA gets much closer to that of the CA, γ < γ ,
R 0 No transmission, iBPS=0 BPS.
which reveals that our distributed algorithm, i.e. the DMA is a PU
promising solution to practical implementation of user pairing We use the notation γR to refer to the instantaneous received
CU
and power allocation in practical CR-NOMA systems. SNR of the link between the BS and the PUl and γR
represents the instantaneous received SNR of the link between
the BS and the CUk . The received SNR at PUl is given by
VI. P RACTICAL ATTCM AIDED CR-NOMA N ETWORK
PU αl,k PS ̺p |hB,pl |2
A. ATTCM scheme γR = . (36)
(1 − αl,k )PS + N0
100
The received SNR at CUk can be expressed as:
64QAM
16QAM
8PSK CU (1 − αl,k )PS ̺c |hB,cl |2
10-1 4PSK γR = , (37)
N0
10-2
where the term αl,k is the power allocation coefficient of PUl ,
PS is the transmit power emit from the BS and the pathloss
BER

̺ is detailed in Section II. The quasi-static Rayleigh fading


CCMC-ibps=1

DCMC-ibps=1

CCMC-ibps=2

DCMC-ibps=2

CCMC-ibps=3

DCMC-ibps=3

CCMC-ibps=5

DCMC-ibps=5

10-3
channels between the BS and PUl is denoted as hB,pl , while
10-4 this between the BS and CUk is represented by hB,ck . The
channel gains are independent of each other.
10-5
0 5 10 15 20 25
Each of the communication links in our CR-NOMA sys-
SNRr[dB] tem will be assisted by the ATTCM scheme. We chose the
switching thresholds to ensure that the target BER is lower
Fig. 9. The BER versus SNRr performance of TTCM using a frame length
of 120,000 symbols, when communicating over Rayleigh fading channels. than 10−5 , which is given by ΥAT T CM =[4.8, 12, 16, 24]
Four TTCM iterations were invoked. dB as seen in Fig. 9. In contrast to the practical power-
In this section we investigate the practical design of the CR- and bandwidth-efficient ATTCM scheme, we also make use
NOMA system advocated using ATTCM. Employing TTCM of both continuous-input continuous-output memoryless chan-
has the advantage that the system’s effective throughput can nel (CCMC) and the discrete-input continuous-output mem-
be increased upon increasing the code rate, when the channel- oryless channel (DCMC) [31] schemes. More specifically,
quality improves. Additionally, the Bit error ratio (BER) the CCMC based adaptive scheme assumes that idealistic
performance of the system may be improved when TTCM coding and modulation schemes are employed for communi-
is used [29]. The TTCM encoder comprises two identical cating exactly at Shannon’s capacity. By contrast, the DCMC
parallel-concatenated TCM encoders [30] linked by a symbol based adaptive scheme assumes that an idealistic capacity-
interleaver. The first TCM encoder directly processes the achieving code is employed for allowing the PSK/QAM modu-
original input bit sequence, while the second TCM encoder lation schemes considered to operate right at the modulation-
manipulates the interleaved version of the input bit sequence. dependent DCMC capacity. We note that Shannon’s CCMC
Then the bit-to-symbol mapper maps the input bits to complex- capacity is only restricted by the SNR and the bandwidth.
valued ATTCM symbols using the Set Partition based labelling The CCMC-based switching thresholds are represented as
method [29]. The structure of the TTCM decoder is similar ΥCCM C =[1.75, 6, 11, 14] dB, while the switching thresholds

0090-6778 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2017.2744640, IEEE
Transactions on Communications

11

of the corresponding modulation dependent DCMC based of successful matching of the PUs and CUs becomes higher,
scheme are given by ΥDCM C =[2, 8, 12.5, 20] dB, which which results in a higher sum-rate. As expected, the DMA
are also explicitly shown in Fig. 9. The reason why we have algorithm using CCMC achieves a higher average throughput
chosen the target BER to be lower than 10−5 is because the than that for employing the DCMC, as shown in Fig. 10. When
error floor emerging at BER< 10−5 can be removed by using we consider the ATTCM scheme, there is a minor difference
a long outer code, such as a Reed Solomon (RS) code, albeit of (7.3 BPS − 5.3 BPS) = 2 BPS between the DMA and
no RS code was used here. For quasi-static fading channels, the CA.Additionally, the performance by employing the DMA
the achievable rates over different links become random and in CR-NOMA is much better than the RA. The percentage
vary as the channel changes. The relatively performances of Pmatch versus the number of PUs, L, is investigated in Fig. 11.
our ATTCM aided CR-NOMA network will be discussed in The DMA based on CR-NOMA achieve higher percentage
Section VI-B. compare with the RA when they have the same number of
PUs. Moreover, the CR-NOMA employing the DMA imposes
B. Performance of our ATTCM aided CR-NOMA Network a lower complexity compared with the system using the CA,
and its performance can approach to that of the CA.
35
CCMC
DCMC VII. C ONCLUSIONS
Average throughput of matched PUs

30 TTCM
CA

25
DMA In this paper, we have applied matching theory to solve the
RA
user pairing and power allocation problem in the CR-NOMA
20
systems. In our proposed distributed matching algorithm, the
15 PUs trade the available power with the CUs by negotiating the
power allocation coefficients, which guarantees that the rate
10
requirements of both the PUs and CUs are satisfied. We have
5 shown that the proposed DMA results in a stable matching
0
and implement a low complexity. Our numerical analysis has
2 4 6 8 10
Number of PUs, L=K revealed that the DMA achieved a better performance than the
RA benchmark scheme, and it is close to the optimum CA. Im-
PU
Fig. 10. The performance of Rmatched versus the number of PUs for the portantly, we have also shown that the CR-NOMA system can
ATTCM, CCMC and DCMC aided CR-NOMA systems over block fading achieve a significant performance advantage over the OMA
Rayleigh channel. A BER below 10−5 is maintained. The “CA”, “DMA” system. Moreover, we have investigated a practical ATTCM
and “RA” techniques are detailed in Section IV-B. The transmit SNRs of the
PUs are γ P U = 10 and the transmit SNRs of the CUs are γ CU = 15 dB. for the CR-NOMA system, and the system’s throughput can
be further improved according to our simulation results.
Relative-frequency of successfully matched pairs

1.0

A PPENDIX A
0.9

As discussed in Section III, based on Eq. (8) and Eq. (11)


0.8 , the objective function that maximizes the total sum-rate of
matched PU and CU can be rewritten as:
h i
0.7
max W (αl,k ) = max RlP U (αl,k ) + RkCU (1 − αl,k ) . (38)
CCMC
αl,k αl,k
DCMC
0.6 TTCM
CA The term RlP U (αl,k ) is defined in Eq. (3) and RkCU (1 −
DMA
RA αl,k ) is expressed in Eq. (4). The problem of Eq. (38) is a
0.5
2 4 6 8 10 concave function problem and its relative proof is discussed
Number of PUs, L=K
in Appendix B. Additionally, its three linear constraints which
are the condition (a), (b) and (c) of Eq. (8) are shown as
Fig. 11. Pmatch versus L performance of the ATTCM, CCMC and DCMC
aided CR-NOMA systems over block fading Rayleigh channel. A BER below follows:
10−5 is maintained. The “CA”, “DMA” and “RA” techniques are detailed in
Section IV-B.The transmit SNRs of the PUs are γ P U = 10 and the transmit s.t. (a) RlP U (αl,k ) ≥ Rl,req
PU
, ∀l ∀k (39)
SNRs of the CUs are γ CU = 15 dB.
(b) RkCU (αl,k ) ≥ CU
Rk,req , ∀l ∀k (40)
In this section, we investigate the performance of the
CCMC, DCMC and ATTCM aided CR-NOMA systems by (c) 0 ≤ αl,k ≤ 1 , ∀l ∀k . (41)
employing the DMA, aiming to find the appropriate power Hence, the Lagrangian function for Eq. (38) using the corre-
allocations of the PU and CU in the same cluster. In Fig. 10, sponding constrains of Eq. (39), Eq. (40) and Eq. (41) can be
we show that the rate performance of the CCMC, DCMC and expressed as:
ATTCM aided CR-NOMA systems in terms of the average cen
L(αl,k , λ, µ) = W (αl,k )
sum-rate of matched PUs. As seen from Fig. 10, when L  
increases, the sum-rate of matched PU pairs also increases. + λl,k RlP U (αl,k ) − Rl,req
PU

That is because the number of matching pairs gets bigger  


+ µl,k RkCU (1 − αl,k ) − Rk,req
CU
, (42)
as the total number of PUs increases. Thus, the probability

0090-6778 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2017.2744640, IEEE
Transactions on Communications

12


CP
where the weight W (αl,k ) in Eq. (38) aims to maximize the z }|
U
{
sum-rate of the PU and CU. λ = (λl,k : l ∈ L, k ∈ K) is 1 + γP U |hpl |2 − αl,k γP U |hpl |2
a matrix of Lagrange multipliers corresponding to the PUs’ λl,k > ∗
CCU
−1 . (50)
rate requirement constraint of Eq. (39) with λl,k ≥ 0. Then z }| {
the vector µ = (µl,k : l ∈ L, k ∈ K) of Lagrange multipliers γCU |hck |2 − αl,k γCU |hck |2 + 1
corresponds to the CUs’ rate constraint of Eq. (40) with µl,k ≥
0. The maximization problem of Eq. (42) can be simplified as: Moreover, when the conditions in Eq. (49) and Eq. (50) are
γ P U |hpl |2 + 1
satisfied, we could obtain the bounds related to Eq. (45)
L(αcen
l,k , λ, µ) = log2 ( P U p 2l ) and Eq. (46). Finally, the optimization value of αl,k cen
that
γl |hl | − αl,k γlP U |hpl |2
+1
 maximizes the total sum-rate of PU and CU is given by
CU c 2 CU c 2
+ log2 (1 + γk |hk | − αl,k γk |hk | )  CU
Rk,req

γlP U |hpl |2 + 1
!  1+γkCU |h

CU
c 2
k | −2
c

CCU < CP∗ U ;
PU cen γk |hk |√ 2
+ λl,k log2 ( ) − Rl,req αl,k = (51)
γlP U |hpl |2 − αl,k γlP U |hpl |2 + 1 PU p 2 P U p 2
 γl |hl | +1−

P U
γl |hl | +1
p 2 C ∗
CU > C ∗
P U ;
  γ |h |
+ µl,k log2 (1 + γkCU |hck |2 − αl,k γkCU |hck |2 ) − Rk,req
CU
. l l

(43)
A PPENDIX B
cen
The optimization allocation of αl,k for fixed values of λl,k
According to Eq. (3) and Eq. (4), the optimization problem
and µl,k can be calculated for each PU and CU by applying
of Eq. (38) can be rewritten as:
the Karush-Luhn-Tucker (KKT) conditions [32], such that we  
have: αl,k γlP U |hpl |2
max log2 1 +
cen
∂H(αl,k , λ, µ) 1 + λl,k αl,k (1 − αl,k )γlP U |hpl |2 + 1)
= PU p 2 | {z }
∂αl,k γl |hl | − αl,k γlP U |hpl |2 + 1 f (XP )

1 − µl,k + log2 1 + γkCU |hck |2 − αl,k γkCU |hck |2 . (52)
− . (44)
1 + γkCU |hck |2 − αl,k γkCU |hck |2 | {z }
f (XC )
Then the Lagrange multipliers corresponding to the constraint The second-derivation of Eq. (52) is derived as:
of Eq (39) is given by ′′
cen
[f (XP ) + f (XC )] =
∂H(αl,k , λ, µ) γ P U |hpl |2 + 1 f (XA )
= log2 ( P U p 2 l )
∂λl,k γl |hl | − αl,k γlP U |hpl |2 + 1 z }| {
PU
(γP U |hpl |2 )2  P U p 2 P U p 2 −2

− Rl,req , (45) γl |hl | + 1 − αl,k γl |hl |
ln2
PU f (XB )
where Rl,req is defined in Eq. (5). Then the optimization z }| {
value of αl,k√based on Eq. (45) can be obtained as: αl,k = (γkCU |hcl |2 )2  CU  −2
γlP U |hp 2
l | +1− γlP U |hp 2
l | +1
− γk |hcl |2 + 1 − αl,k γkCU |hcl |2 .(53)
P U p 2 . Additionally, the Lagrange multi- ln2
γl |hl |
plies µl,k corresponds to the constraint of Eq (40) can be As shown in Eq. (53), the term f (XA ) > 0 and the term
expressed as: f (XB ) < 0. In our system, we assume that the term f (XB )
is always greater than f (XA ), thus f (XA ) + f (XB ) < 0.
∂H(αl,k cen
, λ, µ) Therefore, the function [f (XP ) + f (XC )]′′ < 0, which shows
= log2 (1 + γkCU |hck |2 − αl,k γkCU |hck |2 ) − Rk,req
CU
that the problem of Eq. (52) is a concave problem [33].
∂µl,k
CU
1 + γkCU |hck |2 − 2Rk,req R EFERENCES
→ αl,k = . (46)
γkCU |hck |2 [1] Y. Saito, Y. Kishiyama, A. Benjebbour, T. Nakamura, A. Li, and
K. Higuchi, “Non-Orthogonal Multiple Access (NOMA) for Cellular
Hence, we could obtain two bounds on the optimization Future Radio Access,” in IEEE 77th Vehicular Technology Conference
cen
value for αl,k . Based on Eq (44), λ and µ can be expressed (VTC Spring), pp. 1–5, June 2013.
[2] S. L. Shieh and Y. C. Huang, “A Simple Scheme for Realizing the
as: Promised Gains of Downlink Nonorthogonal Multiple Access,” IEEE
Transactions on Communications, vol. 64, pp. 1624–1635, April 2016.
γlP U |hpl |2 − αl,k γlP U |hpl |2 + 1 [3] J. Choi, “Minimum Power Multicast Beamforming With Superposition
λl,k = (1 + µl,k ) − 1 , (47)
1 + γkCU |hck |2 − αl,k γkCU |hck |2 Coding for Multiresolution Broadcast and Application to NOMA Sys-
tems,” IEEE Transactions on Communications, vol. 63, pp. 791–800,
March 2015.
γkCU |hck |2 − αl,k γkCU |hck |2 + 1 [4] Z. Ding, M. Peng, and H. Poor, “Cooperative Non-Orthogonal Mul-
µl,k = (1 + λl,k ) − 1 . (48) tiple Access in 5G Systems,” IEEE Communications Letters, vol. 19,
1 + γlP U |hpl |2 − αl,k γlP U |hpl |2 pp. 1462–1465, Aug 2015.
[5] Z. Ding, F. Adachi, and H. Poor, “The Application of MIMO to Non-
Since λl,k > 0 and µl,k > 0, Eq. (47) and Eq. (48) may be Orthogonal Multiple Access,” IEEE Transactions on Wireless Commu-
further derived as: nications, vol. PP, no. 99, pp. 1–1, 2015.
∗ [6] H. Marshoud, V. M. Kapinas, G. K. Karagiannidis, and S. Muhaidat,
CCU
z }| { “Non-orthogonal multiple access for visible light communications,”
IEEE Photonics Technology Letters, vol. 28, pp. 51–54, Jan 2016.
γCU |hck |2 − αl,k γCU |hck |2 + 1 [7] Y. Liu, Z. Ding, M. Elkashlan, and H. V. Poor, “Cooperative non-
µl,k > ∗ −1 , (49)
CP U orthogonal multiple access with simultaneous wireless information and
z }| { power transfer,” IEEE Journal on Selected Areas in Communications,
1 + γP U |hpl |2 − αl,k γP U |hpl |2 vol. PP, no. 99, pp. 1–1, 2016.

0090-6778 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2017.2744640, IEEE
Transactions on Communications

13

[8] Y. Zhang, H. M. Wang, Q. Yang, and Z. Ding, “Secrecy sum rate max- [32] H. Hindi, “A tutorial on convex optimization II: duality and interior
imization in non-orthogonal multiple access,” IEEE Communications point methods,” in American Control Conference, p. 11, June 2006.
Letters, vol. PP, no. 99, pp. 1–1, 2016. [33] S. Boyd and L. Vandenberghe, Convec Optimization. Cambridge
[9] Z. Ding, Z. Yang, P. Fan, and H. Poor, “On the Performance of Non- University Press, 2004.
Orthogonal Multiple Access in 5G Systems with Randomly Deployed Wei Liang received her M.Sc. and Ph.D. degree in
Users,” IEEE Signal Processing Letters, vol. 21, pp. 1501–1505, Dec wireless communication at University of Southamp-
2014. ton, Southampton, U.K in 2010 and 2015, respec-
[10] A. Goldsmith, S. Jafar, I. Maric, and S. Srinivasa, “Breaking spectrum tively. She is a Research Fellow in Lancaster Uni-
gridlock with cognitive radios: An information theoretic perspective,” versity now. Her research interests include adaptive
Proceedings of the IEEE, vol. 97, pp. 894 –914, May 2009. coded modulation, network coding, matching theory,
[11] Z. Ding, R. Schober, and H. V. Poor, “A General MIMO Frame- game theory, cooperative communication, cognitive
work for NOMA Downlink and Uplink Transmission Based on Signal radio network and Non-orthogonal multiple access
Alignment,” IEEE Transactions on Wireless Communications, vol. 15, scheme.
pp. 4438–4454, June 2016.
[12] L. Lv, J. Chen, and Q. Ni, “Cooperative Non-Orthogonal Multiple Zhiguo Ding (S’03-M’05) received his B.Eng in
Access in Cognitive Radio,” IEEE Communications Letters, vol. 20, Electrical Engineering from the Beijing University
pp. 2059–2062, Oct 2016. of Posts and Telecommunications in 2000, and the
[13] D. Gusfield and R. W. Irving, The Stable Marriage Problem: Structure Ph.D degree in Electrical Engineering from Imperial
and Algorithms. The MIT Press, 1989. College London in 2005. From Jul. 2005 to Aug.
[14] A. Roth and M. Sotomanyor, Two sided matching: a study in game- 2014, he was working in Queen’s University Belfast,
theoretic modeling and analysis, 1st ed. Cambridge University Press, Imperial College and Newcastle University. Since
1989. Sept. 2014, he has been with Lancaster University
[15] Y. Gu, W. Saad, M. Bennis, M. Debbah, and Z. Han, “Matching theory as a Chair Professor. From Oct. 2012 to Sept. 2016,
for future wireless networks: fundamentals and applications,” IEEE he has also been an academic visitor in Princeton
Communications Magazine, vol. 53, pp. 52–59, May 2015. University.
[16] Y. Xiao, K. C. Chen, C. Yuen, Z. Han, and L. DaSilva, “A bayesian over- Dr Ding’ research interests are 5G networks, game theory, cooperative and
lapping coalition formation game for device-to-device spectrum sharing energy harvesting networks and statistical signal processing. He is serving as
in cellular networks,” IEEE Transactions on Wireless Communications, an Editor for IEEE Transactions on Communications, IEEE Transactions on
vol. 14, pp. 4034–4051, July 2015. Vehicular Technology, and Journal of Wireless Communications and Mobile
[17] S. Bayat, R. Louie, Z. Han, B. Vucetic, and Y. Li, “Physical-layer Computing, and was an Editor for IEEE Wireless Communication Letters,
security in distributed wireless networks using matching theory,” IEEE IEEE Communication Letters from 2013 to 2016. He received the best paper
Transactions on Information Forensics and Security, vol. 8, pp. 717–732, award in IET Comm. Conf. on Wireless, Mobile and Computing, 2009, IEEE
May 2013. Communication Letter Exemplary Reviewer 2012, and the EU Marie Curie
[18] B. Di, S. Bayat, L. Song, and Y. Li, “Radio Resource Allocation for Fellowship 2012-2014.
Downlink Non-Orthogonal Multiple Access (NOMA) Networks Using
Matching Theory,” in 2015 IEEE Global Communications Conference
(GLOBECOM), pp. 1–6, Dec 2015. Yonghui Li (M’04-SM’09) received his PhD degree
[19] K. Liu, A. K. Sadek, W. Su, and A. Kwasinski, Cooperative Commu- in November 2002 from Beijing University of Aero-
nications and Networking. Cambridge University Press , New York, nautics and Astronautics. From 1999 - 2003, he was
2009. affiliated with Linkair Communication Inc, where he
[20] S. Timotheou and I. Krikidis, “Fairness for non-orthogonal multiple ac- held a position of project manager with responsibil-
cess in 5g systems,” IEEE Signal Processing Letters, vol. 22, pp. 1647– ity for the design of physical layer solutions for the
1651, Oct 2015. LAS-CDMA system. Since 2003, he has been with
[21] Y. Liu, Z. Qin, M. Elkashlan, Y. Gao, and L. Hanzo, “Enhancing the the Center of Excellence in Telecommunications,
physical layer security of non-orthogonal multiple access in large-scale the University of Sydney, Australia. He is now a
networks,” IEEE Transactions on Wireless Communications, vol. 16, Professor in School of Electrical and Information
pp. 1656–1672, March 2017. Engineering, University of Sydney. He is the recipi-
[22] B. Zhao and M. C. Valenti, “Distributed turbo coded diversity for relay ent of the Australian Queen Elizabeth II Fellowship in 2008 and the Australian
channel,” Electronics Letters, vol. 39, pp. 786 – 787, May 2003. Future Fellowship in 2012.
[23] M. Shamaian, S. H. Lee, S. Vishwanath, and H. Vikalo, “Distributed His current research interests are in the area of wireless communications,
algorithms for spectrum access in cognitive radio relay networks,” IEEE with a particular focus on MIMO, millimeter wave communications, machine
Journal on Selected Areas in Communications, vol. 30, pp. 1947 –1957, to machine communications, coding techniques and cooperative communi-
November 2012. cations. He holds a number of patents granted and pending in these fields.
[24] J. Jalden and B. Ottersten, “On the complexity of sphere decoding He is now an editor for IEEE transactions on communications and IEEE
in digital communications,” IEEE Transactions on Signal Processing, transactions on vehicular technology. He received the best paper awards from
vol. 53, pp. 1474–1484, April 2005. IEEE International Conference on Communications (ICC) 2014 and IEEE
[25] A. Kan and J. Telgen, “The complexity of linear programming,” J. Wireless Days Conferences (WD) 2014.
Statistica Neerlandica, no. 2, pp. 91–107, 1981.
[26] Z. Yang, Z. Ding, P. Fan, and G. K. Karagiannidis, “On the Performance
of Non-orthogonal Multiple Access Systems With Partial Channel Infor- Lingyang Song (S’03-M’06-SM’12) received his
mation,” IEEE Transactions on Communications, vol. 64, pp. 654–667, PhD from the University of York, UK, in 2007,
Feb 2016. where he received the K. M. Stott Prize for excellent
[27] T. Yoo and A. Goldsmith, “Capacity and power allocation for fading research. He worked as a research fellow at the
MIMO channels with channel estimation error,” IEEE Transactions on University of Oslo, Norway until rejoining Philips
Information Theory, vol. 52, pp. 2203–2214, May 2006. Research UK in March 2008. In May 2009, he joined
[28] T. Weber, A. Sklavos, and M. Meurer, “Imperfect channel-state infor- the School of Electronics Engineering and Computer
mation in MIMO transmission,” IEEE Transactions on Communications, Science, Peking University, China, as a full pro-
vol. 54, pp. 543 – 552, March 2006. fessor. His main research interests include MIMO,
[29] L. Hanzo, S. X. Ng, T. Keller, and W. Webb, Quadrature Amplitude cognitive and cooperative communications, security,
Modulation:From Basics to Adaptive Trellis-Coded, Turbo-Equalised and big data. Dr. Song wrote 2 text books,“Wireless
and Space-Time Coded OFDM, CDMA and MC-CDMA Systems. Wiley- Device-to-Device Communications and Networks” and “Full-Duplex Com-
IEEE Press, 2004. munications and Networks” published by Cambridge University Press, UK.
[30] G. Ungerböck, “Channel coding with multilevel/phase signals,” IEEE He is the recipient of IEEE Leonard G. Abraham Prize in 2016 and IEEE
Transactions on Information Theory, vol. 28, pp. 55–67, January 1982. Asia Pacific (AP) Young Researcher Award in 2012. He is currently on the
Editorial Board of IEEE Transactions on Wireless Communications. He is an
[31] S. Ng and L. Hanzo, “On the MIMO channel capacity of multidimen-
IEEE distinguished lecturer since 2015.
sional signal sets,” IEEE Transactions on Vehicular Technology, vol. 55,
pp. 528 – 536, March 2006.

0090-6778 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.