You are on page 1of 5

Coordinated Non-Orthogonal Mulitple Access

(CO-NOMA)
Anthony Beylerian Tomoaki Ohtsuki
Graduate School of Science and Technology, Department of Information and Computer Science,
Keio University, Faculty of Science and Technology, Keio University,
Yokohama, Japan Yokohama, Japan
anthonybeylerian@ohtsuki.ics.keio.ac.jp ohtsuki@ics.keio.ac.jp

Abstract—We propose a Coordinated Non-Orthogonal Mul-


tiple Access (CO-NOMA) scheme for resource allocation in the
context of mobile networks. NOMA refers to schemes where mul-
tiple users can access the wireless channel in the same frequency
band simultaneously. In this study, we consider NOMA with
power domain multiplexing, where the near-far property in space
is exploited through appropriate power allocations. Moreover, the
space domain is further exploited by coordinating transmissions
from several distributed radio units. This approach, known as
Coordinated Multi-Point (CoMP), can complement NOMA in
order to further make use of the available degrees of freedom.
Therefore, we discuss and propose a suboptimal scheduling
strategy that achieves NOMA with coordinated transmitters in
the downlink, with linear complexity, and compare performances
with other schemes. The proposed scheme is shown to enhance Fig. 1: CO-NOMA on one sub-band.
performances for low to medium number of users per cell, as
observed from system level simulations. frequency block, as long as we can separate and decode each
stream at the receivers, typically via a SIC decoder [4].
I. I NTRODUCTION On the other hand, several studies on CoMP [5] have
Resource allocation for the wireless interface is a core shown that performance improvements are possible through
task that is continuously improved, in order to be able to distributed space diversity, particularly in the downlink, where
provide mobile network users with a pleasing user expe- multiple remote radio units (RRU) transmit to respective users
rience. The multiple access and allocation policies directly simultaneously. CoMP can also help solve performance degra-
affect the performances experienced on the wireless network dation issues, and user fairness can be achieved with adequate
and represent the main pivoting point between the different policies [6]. It is therefore interesting to combine both CoMP
generations. Conventional solutions adopt policies based on and NOMA concepts to benefit from the available degrees
the orthogonality of resources in either time, code or fre- of freedom. Effectively, we discuss how Coordinated NOMA
quency domains (i.e. TDMA, CDMA, OFDMA). In recent (CO-NOMA) can achieve this from a resource allocation
work, NOMA [1] was proposed as a way to improve on perspective and propose a scheme that scales linearly with
these solutions through allocating resources non-orthogonally, the number of users.
and consequently improving the overall network capacity.
II. R ELATED W ORKS
As stated in [2], NOMA can be mainly achieved by either
power domain (PD) or code-domain (CD) multiplexing. Due There are many studies in the literature related to both
to complexity constraints, PD multiplexing combined with strategies (CoMP and NOMA). CoMP has been heavily re-
successive interference cancellation (SIC) at the receiver is searched in its several aspects [5]. Howeover, over the recent
preferred and has been recently receiving larger interests as it years, NOMA has received a large amount of attention from
is also currently under the standardization process [3]. many researchers leading to many interesting studies such as
In PD NOMA, we make use of the near-far property. work by the authors of [1] and [7]. Researchers have also
This means that users that experience higher channel gains, studied the fairness in resource allocation for NOMA such
appearing “near” the transmitter, in the downlink, can be as in [8], while others have also proposed switching schemes
allocated a lower transmit power than those experiencing lower between OMA and NOMA [9]. However, few authors discuss
channel gains, while still satisfying performance requirements. leveraging NOMA in a coordinated context. We therefore
This property allows us to multiplex in the power domain argue that it would be worthwhile to investigate access and
different user streams, at the same time, and in the same allocation policies for such a strategy.

978-1-5090-1328-9/16/$31.00 ©2016 IEEE


Authorized licensed use limited to: UNIVERSITE DE MONASTIR. Downloaded on December 10,2021 at 08:19:27 UTC from IEEE Xplore. Restrictions apply.
III. R ESOURCE A LLOCATION FOR CO-NOMA Naturally, this policy can be easily extended to support
Resource allocation is performed at periodic scheduling frequency resource block groups (RBG) instead of RBs. This
intervals, in order to allocate available resources such as power will decrease the user feedback granularity at the cost of
and frequency blocks to the network’s active users. We denote coarser frequency selectivity. Typically time-domain (TD)
by 𝑈 (𝜓) being the set of active users, connected to the 𝜓 𝑡ℎ metrics with QoS/QoE considerations could also be used, at
RRU. We consider one antenna per network point (RRU/user), the cost of additional complexity. However, for conciseness,
having the same transmit power, operating on the same system our investigation focuses on FD based policies.
bandwidth and assume perfect channel estimations. In the fol- On the other hand, in the power domain (PD), we have to
lowing subsections, we discuss the proposed approach incre- decide on how the power will be distributed across sub-bands.
mentally starting from standard orthogonal schemes, moving Although a water-filling technique can be used, it comes at
on to non-orthogonal and coordinated schemes. the cost of extra complexity and more difficult parallelization.
Herein, we assume a static power profile across RBs, where in
A. Non-Coordinated Orthogonal Scheduling all cases we assume that the maximum power is split evenly
In conventional OMA scheduling, only one active user 𝑢 per active RB 𝑓 from the operating set of sub-bands 𝐹 (𝜓),
is scheduled per resource block (RB) 𝑓 from the available following:
frequency block set 𝐹 . It is assumed that we have a resource 𝑃𝑇 (𝜓)
map 𝑀𝑡 (𝜓, 𝑓 ), mapping an RB 𝑓 to an active user 𝑢 (non- 𝑃𝑇 (𝜓, 𝑓 ) = . (3)
∣𝐹 (𝜓)∣
empty buffer), whenever 𝑢 is allocated 𝑓 at scheduling time 𝑡
on RRU 𝜓. Typically, the allocation policy would be aiming B. Non-Coordinated Non-Orthogonal Scheduling
to optimize a target objective. For example, the objective For NOMA, at any scheduling time, multiple users can
can be derived from proportional fairness (PF) or Quality of be served by a certain RRU per frequency block. A RB
Service/Experience metrics, and updated by user reports. 𝑓 is saturated (full) at 𝜓, when the maximum supported
In non-coordinated scheduling, the task can be easily dis- number of users 𝜇𝑚𝑎𝑥 have been allocated 𝑓 . Conditions to
tributed per RRU. Indeed, for a distributed architecture, the allocate multiple users on the same RB are related to SIC
policy can be run locally per RRU 𝜓 as well as in parallel per decoding [4], which works iteratively and introduces power
RB 𝑓 . Although the optimal allocation solution would be to constraints between the multiplexed users. In fact, to decode
find the best user/RB combinations, it would be more difficult streams successfully at all receivers, we need to ensure that
to parallelize as we would have to search the user values on the power level allocated to each multiplexed user is feasible.
all RBs. For practical purposes, sub-optimal approaches where Consequently, if we represent the power level as 𝑃 (𝜓, 𝑓, 𝑢)
the allocation policy is run separately per RB, keep a low with maximum RRU power per RB 𝑃𝑇 (𝜓, 𝑓 ), we would first
scheduling runtime and are parallelizable. As for a centralized have the general constraint:
architecture, such as cloud-RAN, where RRUs and their base ∑
band unit (BBUs) are virtualized, we can have independent 𝑃𝑡 (𝜓, 𝑓, 𝑢) ≤ 𝑃𝑇 (𝜓, 𝑓 ). (4)
𝑢
parallel scheduling tasks per virtual RRU per RB. With this
in mind, let 𝜙(𝜓, 𝑓, 𝑢) represent the frequency domain (FD) Otherwise, the typical approach for power allocation is to
objective metric. The resource allocation policy in its simplest assign different ratios of the total power to each user :
form can be written as follows:
𝑃𝑡 (𝜓, 𝑓, 𝑢) = 𝛽𝑡 (𝜓, 𝑓, 𝑢) × 𝑃𝑇 (𝜓, 𝑓 ), (5)
𝑀𝑡 (𝜓, 𝑓 ) = arg max 𝜙𝑡 (𝜓, 𝑓, 𝑢). (1) ∑𝜇(𝜓,𝑓 )
𝑢∈𝑈 (𝜓) such that 𝑖=1 𝛽𝑡 (𝜓, 𝑓, 𝑢𝑖 ) ≤ 1, following the constraint
For example, 𝜙 can be defined as : (4), with 𝑢𝑖 being the 𝑖𝑡ℎ multiplexed user on the RB 𝑓 and
𝜇(𝜓, 𝑓 ) the actual total number of users multiplexed.
{ Then, considering the deliberate intra-band interference
1, if u ∈
/ 𝑀𝑡−1 (𝜓, 𝑓 ), caused by PD multiplexing, and assuming that the channel
𝜙𝑡 (𝜓, 𝑓, 𝑢) = ˆ 𝑡 (𝜓,𝑓,𝑢)
𝑅 (2) gains are ordered as ∣ℎ𝑡 (𝜓, 𝑓, 𝑢𝑖 )∣ < ∣ℎ𝑡 (𝜓, 𝑓, 𝑢𝑖+1 )∣, we
𝑅𝑡 (𝜓,𝑢)
, otherwise,
would have the second constraint :
with 𝑅ˆ 𝑡 the instantaneous estimated user data rate and 𝑅𝑡 the
𝛽𝑡 (𝜓, 𝑓, 𝑢𝑖 ) ≥ 𝛽𝑡 (𝜓, 𝑓, 𝑢𝑖+1 ), (6)
long term averaged rate similarly defined in [6].
With this policy, for each RRU 𝜓 and each frequency RB following the near-far property.
𝑓 , we pick the optimal user maximizing the objective 𝜙. We Moreover, the post-processing SINR at user 𝑢𝑖 in the SIC
consider that evaluating each objective metric can be done in chain after cancellation becomes :
constant time 𝑐𝜙 . If we define the complexity as the number of 𝛽(𝜓, 𝑓, 𝑢𝑖 )
evaluations that we have to run, in order to reach a decision 𝑆𝐼𝑁 𝑅† (𝜓, 𝑓, 𝑢𝑖 ) = ∑𝜇(𝜓,𝑓 ) , (7)
to allocate the block 𝑓 , this policy is then of linear order 𝑗=𝑖+1 𝛽(𝜓, 𝑓, 𝑢𝑗 ) + 𝜂(𝜓, 𝑓, 𝑢𝑖 )

𝑂(𝑐𝜙 ∣𝑈 (𝜓)∣). This is because we consider that we use linear where the inter-cell interference and noise term divided by the
2
scans, searching for maxima (going through all the users). received power is 𝜂(𝜓, 𝑓, 𝑢𝑖 ) = ∣ℎ(𝜓,𝑓,𝑢𝑖𝜎)∣2 𝑃𝑇 (𝜓,𝑓 ) .

Authorized licensed use limited to: UNIVERSITE DE MONASTIR. Downloaded on December 10,2021 at 08:19:27 UTC from IEEE Xplore. Restrictions apply.
𝑆𝐼𝑁 𝑅† should be higher than the threshold required to 1) Non User-Centric Clusters: In the first case, if the RRU
decode the user’s signal with the lowest MCS rate as described clusters are either fixed or dynamic but not user-centric [13],
in [10]. Effectively, the optimal solution for this allocation we can distribute/parallelize the scheduling task following
policy would be to search all the combinations, in the so-called those clusters, and subsequently per frequency block 𝑓 in each
full search power allocation (FSPA), but the search space for cluster. If we assume that for each user, we have an association
this is too large, making it impractical. In order to decrease with a RRU cluster 𝐶(𝑢), ( we will
) have to perform a search
the complexity, approaches such as fixed power allocation 𝐶(𝑢)
in the sets 𝜅(𝑢) = ∪𝐾 𝑛=1 , comprised of combinations
(FPA) [7], fractional transmit power allocation (FTPA) [7], 𝑛
tree search transmit power allocation (TTPA) [11] and user of up to 𝐾 RRUs from the candidate set. 𝐾 must remain
pair power allocation (UPPA) [10] were previously proposed. small (i.e. 2, 3), since users must report back only a limited
In the proposed strategies, we would need to select the best number of measurements, to avoid control plane overhead, and
set of users to finally allocate the power to, per sub-band in due to the the fact that larger clusters may not always yield
our case. A full search for this task is prohibitive when the significant benefits [13]. With this in mind, we can limit the
number of users is large, since even for the simplest case where maximum cluster size to(𝐾, )making the RRU set selection in
∑𝐾 𝐾
we choose to select only up to a pair of users from ( the ) total the order of 𝑂( 𝑛=1 𝑛 ) = 𝑂(𝐾2(𝐾−1) − 1). If 𝑈 (𝑐)
∑2 ∣𝑈 (𝜓)∣ 𝑛
set 𝑈 (𝜓), we will have to evaluate up to 𝑛=1 = represents all the active users associated to a cluster 𝑐, the
𝑛
(∣𝑈 (𝜓)∣2 + ∣𝑈 (𝜓)∣)/2 subsets per RB per RRU. minimal coordinated scheduling policy can be written as:
Because of this, authors in [10] devised a two step strategy ∏

𝑢𝑡 , Ψ̂𝑡 ) = arg max 𝜙𝑡 (𝜓, 𝑓, 𝑢), (8)
to reduce the search to pair-wise comparisons, further reduced 𝑢∈𝑈 (𝑐),Ψ𝑡 ∈𝜅(𝑢) 𝜓∈Ψ𝑡
by around 15% when considering pairing prerequisites. Al-
though it would be interesting to extend the closed formed ˆ 𝑓) = 𝑢
𝑀𝑡 (𝜓, ˆ𝑡 , ∀𝜓ˆ ∈ Ψ̂𝑡 . (9)
solution in [10] for larger sets, the practical maximum number
of multiplexed users in PD NOMA is limited (typically a pair). In other words, we allocate the RB 𝑓 to the optimal
In fact, authors in [12], show that a large enough propagation user/sub-cluster combinations maximizing a cluster metric.
loss difference is required for candidate users in order to This policy would then have complexity in the order of
achieve practical gains, which would limit possibilities to 𝑂((𝐾2(𝐾−1) − 1)𝑐𝜙 ∣𝑈 (𝑐)∣) per cluster, per RB.
multiplex more users. 2) User-Centric Clusters: As for user-centric clusters, the
A suboptimal way that can be done linearly but still achieve complexity increases since each user reports his own cluster,
improvements would be to first find the best two users max- thus grouping users into separate sets for parallelization is not
imizing the objective metric 𝜙, while still verifying that their straightforward. Also, user-centric clustering is only feasible
SINR is higher than the minimum threshold for the lowest in a centralized architecture where the central unit constructs
MCS coding rate. This can be done with linear complexity by a global (or regional) connectivity graph. This means it would
comparing and evaluating each user’s metric and SINR. be desirable to parallelize the scheduling task on several
After finding these two users, we can then evaluate if they threads instead of distributing it per cluster (such as with fixed
can be paired and find their optimal ratios following the same clusters). Therefore, analyzing the network graph could help us
procedure as in [10]. In the case where they can’t be paired, derive properties of groups that can be used for parallelization.
the user with the highest metric is allocated the entire power as We can notice useful connectivity patterns emerging from
in the OMA strategy. This is suboptimal since the users with simulations, as can be visually observed in the clustering
the highest metric are not always possible to pair, however simulation shown in Fig. 2. In this simulation, the clustering
in the case when it is possible, it will be beneficial. If we process is based on the strongest relative reference signal (RS).
consider that the prerequisite checks and comparisons with This means that a user’s cluster is formed from the RRUs for
the two metrics in the solution set takes a constant time 𝑐𝑒𝑣𝑎𝑙 , which the estimated RS SIR is larger than 𝜖 × 𝑆𝐼𝑅𝑙𝑎𝑟𝑔𝑒𝑠𝑡 ,
the complexity will be in the order of 𝑂((𝑐𝜙 + 𝑐𝑒𝑣𝑎𝑙 )∣𝑈 (𝜓)∣). where 𝜖 is a scaling factor. The green edges in the figure are
user-RRU associations whereas the red edges represent the
major (largest) chosen RRU clusters.
C. Coordinated Orthogonal Scheduling
If we define a community as the set of users connected to
In this study, this approach is referred to as Coordinated the same cluster or a sub-cluster of the largest RRU cluster,
Orthogonal Multiple Access (CO-OMA). With coordination, several groups of users will belong to the same community.
multiple RRUs transmit to the same user simultaneously on These user communities can then be scheduled together as
the same RB 𝑓 . There are different approaches to do this in allocations would not overlap (no mixed RRU sets). Although
CoMP. However in this study we consider the simpler case updating the connectivity graph introduces more complexity in
of single-user joint-transmission (SU-JT), where RRUs jointly the process, it can be manageable considering that clustering
transmit the same user signal. Subsequently, since several RRU can be run separately and does not need to update at the
are simultaneously involved, the policy will highly depend on same frequency as the scheduling process, but on a larger time
the RRU clustering mechanism used. period. For evaluations, we use fixed clusters for simplicity.

Authorized licensed use limited to: UNIVERSITE DE MONASTIR. Downloaded on December 10,2021 at 08:19:27 UTC from IEEE Xplore. Restrictions apply.
Fig. 3: Minimal CO-NOMA connectivity graph.

and its complement at both RRUs will be 𝛽(𝜓1,2 , 𝑓, 𝑢2 ) =


1 − 𝛽(𝜓1,2 , 𝑓, 𝑢1 ). The intuition behind (10), is that the larger
fraction represents proportion to ultimately allocate to user 𝑢1 ,
from the total power to allocate to both users, and the smaller
fractions are to consider the fact that this power originates
from one of the two RRUs.
Fig. 2: Cluster communities.
The exponent term 𝛼 is an optimization term, which can
control the behavior of the power allocation as in FTPA [7].
D. Coordinated Non-Orthogonal Scheduling In order to speed up the computations, we can use 𝛼 = 1,
In the CO-NOMA scenario, we have to consider the and although it may not be the optimal value, the approach
perspectives from the previous two scenarios. This means would still yield improvements as shown in our simulations.
evaluating both (sub-)cluster selection metrics as well as PD Similarly in this case, if we consider the complexity to evaluate
user allocation constraints. The optimal solution would involve the pairing prerequisites and power allocation in constant time,
finding the combination of radio units, user sets and power the worst case allocation complexity will be upper bounded
ratios, again per RB. Obviously for a full search, the solution by 𝑂((𝐾2(𝐾−1) − 1)(𝑐𝜙 + 𝑐𝑒𝑣𝑎𝑙 )∣𝑈 (𝑐)∣) per cluster per RB.
space is exorbitantly large, however we can combine the
previously discussed strategies to decrease the complexity. It is
IV. P ERFORMANCE E VALUATION AND D ISCUSSION
however more difficult to find the optimal channel ratios in this
case since we need different power ratios for the same user.
In this situation, we have to take the different channel values We ran simulations of the discussed policies using MAT-
into consideration, since the received signal is a superposition LAB 2015a 64 bit, following the parameters summarized in
of the transmissions from the coordinating RRUs. Table 1 and observed the results shown in Fig.4. The location,
channel and traffic models are similar to [6], however the
In SU-JT, from the point of view of the user, the useful re-
allocation strategies operate per RB separately in the frequency
ceived power
∑ would consist of the sum of the different received domain. User locations are generated following a Poisson point
powers 𝜓𝑖 ∈Ψ̂ 𝑃 (𝜓𝑖 , 𝑓, 𝑢) × ∣ℎ(𝜓𝑖 , 𝑓, 𝑢)∣2 . Therefore, if we
process (PPP) with a density that is a multiple of the RRU
use the notion of fractionally allocating the power, such as in
density. In other words, 10 users/cell means the user density
FTPA, the power ratios can be constructed in an inversely pro-
is 10 times larger than the RRU density. For evaluation, we
portional fashion to the effect of their corresponding channel
measure the average user packet level throughput (discounting
values on the received power. For example, in the simplest case
dropped packets due to timeouts). Particularly for the traffic
illustrated in Fig.3 with two coordinating RRUs 𝜓1 and 𝜓2 to
model, we used the NGMN model as in [6] but have set the
serve up to two NOMA users 𝑢1 and 𝑢2 , we have the channel
proportion of users with video traffic to 30% while decreasing
values ℎ(𝜓1 , 𝑓, 𝑢1 ), ℎ(𝜓1 , 𝑓, 𝑢2 ), ℎ(𝜓2 , 𝑓, 𝑢1 ), ℎ(𝜓2 , 𝑓, 𝑢2 ).
the number of users with file (FTP) traffic accordingly.
Simplifying the notation we write ℎ𝑗𝑖 as the corresponding
channels between an RRU 𝜓𝑖 and a user 𝑢𝑗 illustrated in Fig.3. In OMA, standard OFDMA is used with the discussed
Subsequently, the suboptimal fractional power ratios at both policy per RB. As for NOMA, we used the UPPA scheme
𝜓1 and 𝜓2 for user 𝑢1 can be found following: [10] for power allocation and user-pairing, as to the best of
our knowledge so far, it has the best performance. In CO-
OMA, the discussed policy is used with fixed clusters of 2
∣ℎ11 ∣−𝛼 RRUs, based on coupling path-loss similar to the one used
∣ℎ11 ∣−𝛼 +∣ℎ12 ∣−𝛼
𝛽(𝜓1,2 , 𝑓, 𝑢1 ) = , (10) in [6]. Finally, for CO-NOMA, we used 𝛼 = 1 for all our
∣ℎ11 ∣−𝛼 ∣ℎ21 ∣−𝛼
∣ℎ11 ∣−𝛼 +∣ℎ12 ∣−𝛼
+ ∣ℎ21 ∣−𝛼 +∣ℎ22 ∣−𝛼
simulations to reduce the computation time.

Authorized licensed use limited to: UNIVERSITE DE MONASTIR. Downloaded on December 10,2021 at 08:19:27 UTC from IEEE Xplore. Restrictions apply.
TABLE I: Simulation Configuration
Parameter Simulation Model Parameters
Frequency/Bandwidth/Duplexing 2 GHz/10 MHz/FDD
Region 1500 m2
RRU Locations MHCPP-II 30.10−6 stations/m2
Inter-site Distance 500 m
Antenna Configuration 2D-Omni SISO
RRU Power 30 W
UE Locations PPP variable density
Hysteresis Threshold 3 dB
Access Scheme OMA/CO-OMA/NOMA/CO-NOMA
NOMA SIC Perfect cancellation, constraints [10]
Clustering Scheme Fixed
Maximum Cluster Size 2
Maximum NOMA Users 2
CO-NOMA Channel Exponent 1
FFT size 1024
Scheduling Interval 1 sub-frame
User motion random 0 to 6 kmph
Pathloss Model 3GPP Outdoor LOS-NLOS [14]
Fig. 4: Average user throughput per access scheme. Link Adaptation 10 % BLER target
Modulation Order QPSK, 16QAM, 64QAM
Channel Estimation Ideal
We can observe consistent improvements over different user Packet Drop Time 3GPP QoS table
densities, when using CO-NOMA. Naturally, when the number Traffic Model modified NGMN mix
of users increases, the performance drops for all strategies. HARQ retransmission Not modeled
However, the observed trend also shows that the performance
gap between strategies closes.
For NOMA, the drop in the performance gain, can be R EFERENCES
explained in part by the fact that the number of successful [1] Y. Saito, Y. Kishiyama, A. Benjebbour, T. Nakamura, A. Li, and K.
Higuchi, “Non-orthogonal multiple access (NOMA) for future radio
times users are paired while using the suboptimal linear search access,” IEEE VTC Spring, June 2013, pp. 15.
decreases. Effectively, we try to pair users maximizing the FD [2] L. Dai, B. Wang, Y. Yuan, S. Han, C.-L. I, and Z. Wang, Non-orthogonal
metric, which does not guarantee to find user pairs. With more multiple access for 5G: Solutions, challenges, opportunities, and future
research trends, IEEE Commun. Mag., vol. 53, no. 9, pp. 7481, Sep. 2015.
candidate users, we have less chances to obtain appropriate [3] A. Benjebbour, A. Li, K. Saito, Y. Saito, Y. Kishiyama, T. Nakamura,
user pairs in the resulting set. Possible ways to improve this “NOMA : from concept to standardization,” IEEE Conference on Stan-
behavior could be through user grouping into two sets (edge dards for Communications and Networking (CSCN), pp. 18-23, Oct. 2015.
[4] C. Yan, A. Harada, A. Benjebbour, Y. Lan, A. Li, and H.Jiang, “Receiver
and center users), which can be interesting to investigate in design for downlink non-orthogonal multiple access (NOMA),” IEEE
future work. VTC Spring, May 2015.
As for CO-OMA, although we benefit from spatial diversity [5] G. Li, J. Niu, D. Lee, J. Fan, and Y. Fu, “Multi-cell coordinated scheduling
and MIMO in LTE,” IEEE Commun. Surveys Tuts., vol. 16, no. 2, pp.
gain, with more users, we run out of resources quickly since 761 775, 2014.
we reserve them on multiple RRU, and cannot satisfy all the [6] A. Beylerian, T. Ohtsuki, “Multi-point fairness in resource allocation for
users as large delays lead to packet drops due to timeouts. C-RAN downlink CoMP transmission,” EURASIP Journal on Wireless
Comm. and Net., January 2016.
Finally, CO-NOMA leverages the benefits from both the [7] A. Benjebbour, A. Li, Y. Saito, Y. Kishiyama, A. Harada, T. Nakamura,
space diversity gains as well as the PD multiplexing. Although “System-Level Performance of Downlink NOMA for Future LTE En-
the discussed linear search undermines its performance for hancements,” IEEE Globecom, Dec. 2013.
[8] S. Timotheou and I. Krikidis, “Fairness for non-orthogonal multiple
larger user densities, the observed result encourages the search access in 5G systems,” IEEE Signal Process. Lett., vol. 22, no. 10, pp.
for improved strategies in this direction. 16471651, Oct. 2015.
[9] Marie-Rita Hojeij, Joumana Farah, Charbel Abdel Nour, Catherine Douil-
V. C ONCLUSION lard, “New Optimal and Suboptimal Resource Allocation Techniques for
Downlink Non-orthogonal Multiple Access,” Wireless Personal Commu-
In this paper, we discuss CO-NOMA, a downlink multiple nications, April 2016, Volume 87, Issue 3, pp 837-867.
access and resource allocation scheme for mobile networks. [10] Fei Liu, Petri Mahonen, Marina Petrova, “Proportional Fairness-Based
CO-NOMA leverages the advantages of both CoMP and User Pairing and Power Allocation for Non-Orthogonal Multiple Access”,
IEEE PIMRC, 2015.
NOMA and is shown through system level simulations to [11] A. Li, A. Harada, and H. Kayama, “A novel low computational complex-
outperform both, when considering user pairs for PD mul- ity power allocation method for non-orthogonal multiple access systems,”
tiplexing and 2 RRUs for coordinated transmission. We have IEICE Trans. Fundamentals, vol. E97-A, no. 1, pp. 57-67, Jan. 2014.
[12] T. Yazaki, Y. Sanada, “Effect of joint detection and decoding in non-
also discussed the complexity implications and show how this orthogonal multiple access”, ISPACS, Dec. 2014, pp. 245-250.
approach can scale linearly with the number of active users in [13] V. Garcia, Y. Zhou, and J. Shi, “Coordinated multipoint transmission
a suboptimal fashion but can still achieve performance gains in dense cellular networks with user-centric adaptive clustering,” IEEE
Trans. Wireless Commun., vol. 13, no. 8, pp. 42974308, Aug. 2014.
particularly with low to medium user densities. Future work [14] 3GPP TR 36.828 V11.0.0, “Further enhancements to LTE Time Division
could be to investigate user grouping and other strategies, in Duplex (TDD) for Downlink-Uplink (DL-UL) interference management
order to improve the behavior of the discussed policy. and traffic adaptation”.

Authorized licensed use limited to: UNIVERSITE DE MONASTIR. Downloaded on December 10,2021 at 08:19:27 UTC from IEEE Xplore. Restrictions apply.

You might also like