Content uploaded by Quang-Tuan Thieu
Author content
All content in this area was uploaded by Quang-Tuan Thieu on Feb 06, 2018
Content may be subject to copyright.
A Wideband Scheduling Method for Non-Orthogonal Multiple
Access in the Vienna LTE-A Downlink System-Level Simulator
Quang-Tuan Thieu∗, Chun-Hsiung Wang∗, and Hung-Yun Hsieh∗†
∗Graduate Institute of Communication Engineering
†Department of Electrical Engineering
National Taiwan University
Taipei, Taiwan 106
Email: {d96942029, r03942073, hungyun}@ntu.edu.tw
Abstract—Non-Orthogonal Multiple Access (NOMA) is a
promising direction to meet the demand of high spectral efficiency.
To fully utilize its advantage, an efficient scheduling method
for multiuser selection is required. In this paper we propose a
NOMA wideband scheduling method taking into account not only
the user set and their power allocation but also the modulation
and coding scheme. Furthermore, we introduce an extension of
the open-source Vienna LTE-A downlink system-level simulator
by implementing multi-user superposition transmission at the
transmitter and successive interference cancelation at the receiver.
Simulation results validate that the proposed scheduling method
outperforms baseline methods.
Index Terms—Multiuser superposition transmission, successive
interference cancelation, proportional fairness.
I. INTRODUCTION
Non-orthogonal multiple access (NOMA) has recently re-
ceived a lot of attention as a promising technique for the next
generation (5G) wireless network [1]. Different from orthogonal
multiple access (OMA) used by the current LTE-A system,
where radio resources are divided as the resource block (RB)
grid to be allocated orthogonally to individual users, NOMA
allows an RB to be used by multiple users when their signals
are multiplexed in the power domain at the transmitter side and
separated at the receiver side.
While NOMA has been actively investigated in both
academia and standardization bodies on its transceiver design,
there is a need for more research endeavors on system-level and
cross-layer issues such as resource allocation and scheduling.
Many scheduling methods proposed in the literature are based
on simple abstractions at the link layer. For example, among
the proposed scheduling methods for NOMA, a popular method
is based on proportional fairness (PF) as presented in [2], [3].
In principle, these algorithms perform exhaustive search for
all combinations of user sets and power allocation, and the
combination with the highest PF metric is selected. In [4],
the authors propose an improved version of the PF scheduling
method, where the mathematical characteristic of the PF metric
is exploited to reduce the number of searched user sets. The
method proposed in [5] is to update the fairness index during
the target frame allocation such that a better subband can be
allocated to a priority user through adaptive subband ordering.
In this paper, we are interested in designing an efficient
NOMA scheduler that takes into consideration not only user
pairing and power allocation, but also modulation and coding
scheme (MCS) selection. It is desired that the chosen MCS
setting results in acceptable block error rate (BLER) while
maximizing the throughput at the user. Toward this goal, we
build an MCS table that records the set of feasible MCS
settings and power split factors for any given pair of users
under consideration such that the eNodeB can have a more
accurate estimation of user performance before scheduling.
The MCS table is built in advance based on the symbol-level
interference cancelation (SLIC) receiver implemented in the
link-level simulator for use by the eNodeB in the system-level
simulator during the scheduling task.
To implement the proposed NOMA scheduling method, we
extend the well-known open-source Vienna LTE-A downlink
system-level simulator [6] that is compatible with the 3GPP
LTE-A standards. Our modification includes fundamental build-
ing blocks such as the resource block structure that was
originally designed for OMA as well as the behaviors of the
end-user and the eNodeB such that they are all compliant
for NOMA. In addition, to simulate the impact of the SLIC
receiver in the system-level simulator, we implement a link-to-
system-level mapping method that captures the characteristic
of the practical receiver used. By combining both the Vienna
LTE-A link-level and system-level simulators, we provide a
powerful tool for further investigation of NOMA in the LTE-
A communication system. Our simulation results show that,
the proposed NOMA wideband scheduling method outperforms
baseline methods in terms of the average user throughput and
spectral efficiency.
The structure of this paper is organized as follows. In
Section II, we introduce a framework in which we extend
the Vienna LTE-A downlink system-level simulator to work
for NOMA. The extension includes the introduction of the
link-to-system-level mapping for the SLIC receiver to model
the practical receiver effect. In Section III, we describe the
NOMA wideband scheduling method for the target scenario.
Simulation results are shown in Section IV, where we compare
the proposed method with baseline algorithms. Finally, we
conclude the paper in Section V.
978-1-5090-2482-7/16/$31.00 ©2016 IEEE
EĞĂƌͲh
&ĂƌͲh
ĞEŽĚĞ
Fig. 1: NOMA with a single-beam scenario
II. NOMA IMPLEMENTATION IN THE VIENNA LTE-A
DOWNLINK SYSTEM-LEVEL SIMULATOR
In this section, we first briefly introduce the theoretic model
of NOMA. Then we describe essential steps to integrate NOMA
into the existing Vienna LTE-A downlink system-level simula-
tor.
A. Theoretic Model for NOMA
We consider the downlink channel of a cellular communica-
tion system in which the eNodeB has 𝑁𝑡transmit antennas to
perform spatial multiplexing over B spatial beams. We assume
that NOMA is applied with the signal intended for two target
users called the near and far users based on their distance
to the serving eNodeB. We also assume that the near user
has better channel condition while the far user has worse
channel condition to benefit from the transmission scheme.
The transmission signal for the first beam to both users that
is multiplexed using superposition coding and precoded by a
precoding vector can be described as follows:
x=w1(𝜇𝑁𝑃1𝑠𝑁+𝜇𝐹𝑃1𝑠𝐹)+
𝐵
𝑖=2
w𝑖
𝑗
𝜇𝑖,𝑗 𝑃𝑖𝑠𝑖,𝑗 ,
where w𝑖,1≤𝑖≤𝐵,isan𝑁𝑡-dimensional unit-form
precoder applied at the 𝑖𝑡ℎ beam and 𝑃𝑖is the transmitted power
allocated at beam 𝑖. The power split factors for near and far
users are denoted by 𝜇𝑁and 𝜇𝐹=1−𝜇𝑁respectively, while
the modulated symbols for near and far users are 𝑠𝑁and 𝑠𝐹
with 𝐸[∣𝑠𝑁∣2]=𝐸[∣𝑠𝐹∣2]=1. Similarly, the 𝑗𝑡ℎ power-scaled
modulated symbol carried at beam 𝑖, with 𝑖≥2, is denoted by
𝜇𝑖,𝑗 𝑃𝑖𝑠𝑖,𝑗 with 𝑗𝜇𝑖,𝑗 =1. In this paper, we assume that
at each time slot, the two users share the whole wideband for
their transmission signals and the modulated symbols intended
for near and far users are superposed and precoded by the same
precoder. In the following, we consider the scenario where the
eNodeB performs a single-beam transmission (i.e. 𝐵=1)as
shown in Fig. 1.
B. Extending the Vienna LTE-A Downlink System-level Simu-
lator
In the Vienna LTE-A downlink system-level simulator, a
physical resource block (RB) is used by one user equipment
(UE) with homogeneous transmission power among all RBs
and all UEs. Yet NOMA allows at least two UEs to share one
RB but with different transmission powers, scaled either by 𝜇𝑁
or 𝜇𝐹as mentioned earlier. Therefore, it is required to adjust
the class “RB grid” to increase the number of allocated users
from 1to 2per RB and the power split factor (𝜇𝑁and 𝜇𝐹)
is also added as the property of the class. It is noted that, the
resource block class is made globally hence any change is both
known by all UEs and the eNodeB.
The performance of LTE-A downlink is measured based on
some statistic metrics including the average UE throughput,
peak UE throughput and cell-edge throughput. To reduce the
computational complexity, the Vienna LTE-A downlink system-
level simulator introduces the link-quality model to estimate
post-processing SINR for each UE. This model incorporates all
the macroscopic path loss, shadowing and micro-scale fading
elements given for every 6subcarriers to obtain the received
signal power. Signals from neighboring eNodeBs are also
taken into consideration as inter-cell interference. In the simple
implementation of an ideal NOMA receiver, the SIC receiver
employed by the near user is perfect such that the received
power at the near user is not impacted by the signal of the
far user. On the other hand, the far user is expected to remove
the signal from the near user by itself. In the Vienna LTE-
A downlink system-level simulator, we model the receiver at
the far user by considering the signal targeted to the near
user to be similar to interference generated by the neighboring
eNodeBs. Consequently, the simulator treats that interference
by generating the micro-scale fading elements with parameter
𝜃while the signal sent to the far user itself has micro-scale
fading elements with parameter 𝜁.
Afterwards, the post-processing SINR is passed to the
link-performance model where the transport block SINR
(TB SINR) is obtained for each layer and codeword. The
corresponding block error rate (BLER) and the acknowledge-
ment (ACK) for each codeword is estimated to incorporate
the transport block size (TB Size) given by the eNodeB to
calculate the UE throughput. In addition to the ideal NOMA
receiver model, we also consider a practical SIC receiver that
performs the estimation of practical post-processing SINR if the
user is detected as the near user. We defer the details of this
implementation in Section II-C. The process of implementing
NOMA for an UE in the system-level simulator is illustrated
in Fig. 2.
At the eNodeB side, after the scheduling process completes,
the eNodeB estimates the transport block CQI (TB CQI) such
that the BLER of the respective UE is under a predefined
threshold 𝜖.TBCQI plays an important role since the UE
will use it along with the post-processing SINR to estimate
the BLER and ACK as described above. To comply with the
NOMA scheme, the calculation of TB CQI is explained as
WŽǁĞƌƐƉůŝƚ
ĨĂĐƚŽƌ
D^ͬDKͬ
ŽĚĞZĂƚĞ
^/EZ
;ŝĚĞĂů^/Ϳ
dŚƌŽƵŐŚƉƵƚ
>ŝŶŬͲƋƵĂůŝƚLJŵŽĚĞ ů
hĞEŽĚĞ
^ĐŚĞĚƵůĞƌ
>ŝŶŬͲƉĞƌĨŽƌŵĂŶĐĞŵŽĚĞů
/ĨŶĞĂƌͲƵƐĞƌ
dͺ^/EZ
^/EZ
;ƉƌĂĐƚŝĐĂů^>/Ϳ
>Z
<
ƐƚŝŵĂƚĞĚ
Y/
dͺY/
D^ͬDKͬ
ŽĚĞZĂƚĞ
dͺ^ŝnjĞ
ůůŽĐĂƚĞĚh
ĂŶĚƉŽǁĞƌ
ƐƉůŝƚĨĂĐƚŽƌ
Fig. 2: NOMA implementation in the Vienna LTE-A
downlink system-level simulator
follows: Firstly, the estimated SINR under the perfect SIC
assumption is given by:
SINR𝑁=𝜇𝑁∗SINR1;SINR
𝐹=𝜇𝐹∗SINR2
𝜇𝑁∗SINR2+1,(1)
where SINR1and SINR2are the SINRs of the two users se-
lected as near and far users, respectively. The required TB CQI
is then estimated given the average CQI converted from either
SINR𝑁or SINR𝐹for BLER ≤𝜖(in the system-level simulator,
𝜖is set to 10−1). After that, the modulation order (e.g. for
QPSK, 16QAM or 64QAM) and coding rate are derived and
sent alongside with the TB CQI to the allocated UE.
C. Modeling Practical Receivers in the System-level Simulator
As presented previously, to extend the Vienna LTE-A down-
link system-level simulator to support NOMA, one implemen-
tation is to assume that the signal targeted to the far user can
be perfectly removed at the near user using the SIC receiver. In
fact, however, it is not realistic to have such an ideal receiver.
Moreover, the assumption could lead to an over-estimation of
performance gain. Therefore, it is more important to have “real”
but not “ideal” SINR of the near user if we want to investigate
the real performance of any scheduling algorithm designed for
NOMA.
In general, the purpose of the system-level simulator is to
focus on network-related issues such as resource allocation
and scheduling. In system-level simulation, the physical layer
is abstracted from link-level results to reduce the computation
complexity. In the following, we introduce how we model the
practical receiver at the system-level simulator by estimating
the effective SINR of the near user using the link-to-system
W
ƌĞŵŽǀĞ &&d ƋƵĂůŝnjĞƌ ĞͲ
ŵĂƉƉĞƌ
ĞŵŽĚƵͲ
ůĂƚŝŽŶ ZdžͲdĞŵƉ
DŽĚƵůĂͲ
ƚŝŽŶ
DĂƉƉĞƌ
ĞͲ
ƋƵĂůŝnjĞƌ
^ƵďƚƌĂĐƚ
ƋƵĂůŝnjĞƌ ĞͲ
ŵĂƉƉĞƌ
ĞŵŽĚƵͲ
ůĂƚŝŽŶ ĞĐŽĚĞƌ
ZdžͲĂƚĂ
ZĞĐĞŝǀĞĚ
^ŝŐŶĂů
Fig. 3: Symbol-level IC receiver
mapping method [7]. The method depends on the performance
of the receiver in the link-level simulator, and hence we start
by briefly introducing the design of the receiver.
1) Receiver Design at the Link Level: We illustrate in Fig. 3
the SLIC receiver to detect signal at the near user in the link-
level simulator. Particularly, at each time-frequency resource
element, we use the SIC receiver to detect the desired symbol
𝑠𝑁. Firstly, interference is canceled by detecting and decoding
the signal from the far user when treating its own signal as
noise. After that, the receiver subtracts the determined far user
signal from the received signal and extracts its own signal. The
following introduces the mapping method from the link-level
simulator to the system-level simulator to achieve the practical
SINR of the near user.
2) Mapping Practical SINR from the Link Level: The key
idea of the mapping method is to estimate the mutual infor-
mation per bit (MIB) of the SLIC receiver described earlier.
The MIB on each time-frequency resource element (RE) is
estimated with the calibration factor 𝛽as a weighting of the
bit-interleaved coded modulation (BICM) capacity, given as
follows:
MIBSLIC =𝛽∗BICM𝑐.(2)
The calibration factor 𝛽with 0<𝛽<1is found by solving
the optimization problem:
𝛽=argmin max
1≤𝑖≤𝑁∣log 𝑔(𝑓−1(𝛽∗BICM𝑐))−log BLER(𝑖)∣,
(3)
where the BLER of the near user is obtained by running the
link-level simulation with the SLIC receiver. Function 𝑓(⋅)in
Eq. (3) maps an SNR value to received bit mutual information
rate (RBIR), and 𝑔(⋅)maps an SNR value to BLER, both in the
AWGN channel. We generate a look-up table for 𝛽depending
on three parameters: MCS index of the near user (MCS𝑁),
modulation order of the far user (MOD𝐹) (QPSK, 16QAM or
64QAM) and the power split factor 𝜇𝐹. The BICM capacity
is also obtained given the SNR of the near user, MOD of both
near and far users and 𝜇𝐹.
The steps of estimating the practical SINR of the near user
are presented in Algorithm 1. The BICM capacity is found
by looking up the BICM table given the power split factor of
the far user 𝜇𝐹and the SINR∗
𝑁of the near user (noted that,
Algorithm 1 Determining practical SINRs
1: Determine BICM𝑐from the given (𝜇𝐹,SINR
∗
𝑁);
2: Determine 𝛽from the given (MCS𝑁,MOD
𝐹);
3: if nCodeWord == 1 then
4: MCS𝑁(cw0) &MOD𝐹(cw0) →𝛽SLIC(0) ;
5: else if nCodeWord == 2 then
6: MCS𝑁(cw0) &MOD𝐹(cw0) →𝛽SLIC(0) ;
7: MCS𝑁(cw1) &MOD𝐹(cw1) →𝛽SLIC(1) ;
8: end if
9: MIB𝑆𝐿𝐼𝐶 =𝛽SLIC ∗BICM𝑐
10: Convert MIB𝑆𝐿𝐼𝐶 →˜
SINR𝑁;
SINR∗
𝑁is the SINR value before being scaled by the power split
factor 𝜇𝑁. i.e. SINR∗
𝑁=SINR1). The calibration factor 𝛽is
looked up next, depending on the MCS and MOD of the paired
users. Depending on whether there is one layer/one codeword
or two layers/two codewords, the corresponding 𝛽has one or
two elements, as shown from Lines 3to 7. Lastly, we get MIB
and convert it to the practical ˜
SINR.
III. WIDEBAND SCHEDULING METHOD FOR NOMA
A. Construction of the MCS table
Before introducing the proposed scheduling method, we
illustrate in this section how all feasible MCS settings are
recorded in a table called the MCS table. An MCS table is
built by recording the power split factors and MCS settings
that satisfy the constraint on BLER of all given user pairs.
Therefore, the input of the algorithm include the SNRs of
the two candidate users, the set of power split factors under
consideration for superposition and the number of subframes
Nframes.
As presented in Algorithm 2, for the given SNRs of two
users, we initially use their highest MCS settings while variable
𝑀best saves the best MCS settings which have ever been
found. Subsequently, the corresponding modulation order and
coding rate are specified. Next in Line 5, we determine the
set Aof power split factors suitable for the two modulation
orders. The principle for choosing the set Ais mainly based
on the constellation points and the BICM capacity. That is, a
power split factor is selected if the minimum distance between
constellation points is maximum for the purpose of reducing
the detection error [7], [8].
From Lines 6to 12, for each 𝑎𝑖=(𝜇𝑁,𝜇𝐹)belonging to
set A, we run the link-level simulator for 𝑁frames and obtain
BLER𝑁, BLER𝐹, and the corresponding throughput 𝑟𝑁,𝑟𝐹.
If BLER is less than the predefined threshold 𝜖(implying that
the constraint is satisfied while MCS𝑁,MCS
𝐹, or both is also
better than 𝑀best), we save the feasible setting (MCS𝑁,MCS
𝐹,
𝑟𝑁,𝑟𝐹,𝑎𝑖) into the MCS table and update 𝑀best. The algorithm
stops when all the MCS settings and power split factors are
verified.
Algorithm 2 Construction of the MCS table
1: Input: SNR𝑁,SNR
𝐹;
2: Initilization: set MCS𝑁=MCS
Nmax,MCS
𝐹=MCS
Fmax,
𝑀best =(0,0);
3: while MCS𝑁≥1&MCS𝐹≥1do
4: Determine MOD𝑁,MOD
𝐹;
5: A←power split factor set given MOD𝑁,MOD
𝐹;
6: for 𝑎𝑖∈Ado
7: Run link-level simulation for 𝑁frames;
8: Obtain BLER𝑁, BLER𝐹,𝑟𝑁,𝑟𝐹;
9: if (BLER𝑁,BLER𝐹)<𝜖&(MCS𝑁,MCS𝐹)>
𝑀best then
10: MCS table ←(MCS𝑁,MCS𝐹,𝑟
𝑁,𝑟
𝐹,𝑎
𝑖);
11: 𝑀best ←(MCS𝑁,MCS𝐹);
12: end if
13: end for
14: if MCS𝐹∕=1then
15: MCS𝐹=MCS𝐹−1;
16: else if MCS𝑁∕=1then
17: MCS𝑁=MCS𝑁−1;
18: MCS𝐹=MCS𝑁;
19: end if
20: end while
B. NOMA Wideband Scheduling Algorithm
In the enhanced multiuser superposition coding scheme for
NOMA, selecting users with appropriate transmission power
settings plays a vital role to whether the received signal can
be extracted and decoded correctly at both near and far users.
We present in this section a multiuser scheduling algorithm
employing the proposed MCS table. The methodology of the
approach is to maximize the proportional fairness (PF) metric
for all scheduled users with the consideration of MCS selection.
The approach is used for both orthogonal multiplexing access
(OMA) based on OFDMA and NOMA based on multiuser
superposition transmission.
Denote 𝜔=(U,𝜇)as a user pair combination, where U
denotes the candidate user set and 𝜇denotes the allocated power
split factor set. According to [9], the wideband PF scheduler
will select 𝜔∗if it satisfies:
𝜔∗=argmax
𝜔
𝑘∈𝜔
𝑟𝑘
(𝑡𝑐−1)𝑇𝑘
,(4)
where 𝑟𝑘is the instantaneous throughput, 𝑇𝑘is the average
throughput, and 𝑡𝑐is the average window.
The algorithm shown in Algorithm 3 to be executed at the
eNodeB for every TTI is described in details as follows. For
each user 𝑢𝑖, the scheduling algorithm evaluates the PF metric
for the OMA scheme (Line 6). It continues to assess other user
𝑢𝑗with channel condition worse than 𝑢𝑖, represented by the
condition CQI(𝑢𝑖)>CQI(𝑢𝑗)in Line 8. The PF metric for the
NOMA scheme is then calculated. For any pair of users (𝑢𝑖,𝑢𝑗),
we take their reported CQI as the MCS table’s input and return
their instantaneous throughput (𝑟𝑖,𝑟𝑗), respectively. The best PF
Algorithm 3 NOMA wideband scheduling algorithm
1: //Initialization
2: PF N=-Inf(𝑁𝑈𝐸,𝑁
𝑈𝐸);
3: PF O=-Inf(𝑁𝑈𝐸);
4: for 𝑢𝑖∈Udo
5: Estimate 𝑟𝑜(𝑢𝑖)from CQI(𝑢𝑖);
6: PF O(𝑢𝑖)= 𝑟𝑜(𝑢𝑖)
(𝑡𝑐−1)𝑇(𝑢𝑖);
7: for 𝑢𝑗∈U,𝑗 ∕=𝑖do
8: if CQI(𝑢𝑖)>CQI(𝑢𝑗)then
9: 𝑟𝑛(𝑢𝑖),𝑟
𝑛(𝑢𝑗)←MCS(CQI(𝑢𝑖),CQI(𝑢𝑗));
10: PF N(𝑢𝑖,𝑢
𝑗)= 𝑟𝑛(𝑢𝑖)
(𝑡𝑐−1)𝑇(𝑢𝑖)+𝑟𝑛(𝑢𝑗)
(𝑡𝑐−1)𝑇(𝑢𝑗);
11: end if
12: end for
13: end for
14: Find (𝑢∗
𝑖,𝑢
∗
𝑗)s.t. PF N(𝑢∗
𝑖,𝑢
∗
𝑗)=max(PF N);
15: Find 𝑢∗
𝑘s.t. PF O(𝑢∗
𝑘)=max(PF O);
16: if max(PF N)>max(PF O)then
17: Allocate 𝑢∗
𝑖,𝑢
∗
𝑗;
18: 𝜇𝑁,𝜇
𝐹←MCS(CQI(𝑢∗
𝑖),CQI(𝑢∗
𝑗));
19: else
20: Allocate 𝑢∗
𝑘;
21: end if
metric for both OMA and NOMA as well as the corresponding
users are found. We allocate the whole bandwidth to the pair
(𝑢∗
𝑖,𝑢∗
𝑗) as near and far users, respectively if their maximum
PF metric is higher than the maximum PF metric in OMA
case. Otherwise, only user 𝑢∗
𝑘can use all the resource blocks.
In NOMA case, the power split factors (𝜇𝑁and 𝜇𝐹)arealso
identified from the MCS table.
IV. SIMULATION RESULTS
We evaluate the performance of the proposed scheduling
method for NOMA under the extension of the Vienna LTE-A
downlink system-level simulator in this section. Table I gives
the simulation parameters. The scenario includes a hexagonal-
grid of 19 sites with 3cells per sites (only 7center sites are
active). The macroscopic path loss model follows the 3GPP
specifications TS 36.942, while the shadow fading is generated
as a log-normal-distributed 2D-space-correlated map.
The baseline method is the multiuser PF scheduling method
proposed by NTT Docomo [3]. We compare our proposed
NOMA wideband scheduling method with the baseline for
both the ideal and the practical receiver including maximum-
likelihood (ML) and SLIC receiver. Table II shows that the
simulation result obtained by Docomo’s method is similar to
the one presented in [10], where NOMA can gain around
17% for the average UE throughput compared to the OMA
scheme. On the other hand, the proposed NOMA wideband
scheduling method can even achieve better gain up to 44%
(i.e. approximately twice the existing baseline scheduler). The
reason for this advancement predominantly comes from the
better estimation of instantaneous throughput 𝑟𝑘in Eq. (4)
based on the MCS table.
TABLE I: Simulation parameters
Parameters Value
Cell layout Hexagonal grid, 19-cell sites
with tri-sectorization
Bandwidth 10MHz
Number of Resource Block 50
No. of Antennas (Tx/Rx) 2x2
eNodeB’s max Tx power 46dBm
Tx Mode TM4 (CLSM)
Number of UEs per cell 10
Macroscopic pathloss model TS 36.942
Shadow fading type claussen
Small scale fading model Winner II+
Traffic model Full buffer
Average window 𝑡𝑐100
Simulation time 10,000 TTI
Average UE throughput [Mb/s]
024681012
Cummulative Distribution Function
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
OMA-PF Scheduler
NOMA-PF Scheduler-Docomo method
NOMA-PF Scheduler-MCS-based method
Fig. 4: CDF of the average UE throughput
For further understanding the performance of the proposed
scheduling method, we plot the empirical CDF of the average
UE throughput for different schemes in Fig. 4. We can see that
the average UE throughput of less than 2Mb/s occupies over
85% for Docomo’s method. In the meantime, the corresponding
number for the proposed method utilizing the MCS table is
only about 55%. Both scheduling methods have 94% of average
UE throughput less than 4Mb/s. Note that while only 11% of
the average UE throughput ranges from 2to 4Mb/s under
Docomo’s method, up to 39% be achieved by the proposed
method. These numbers can explain why the proposed method
outperforms the baseline in terms of average UE throughput.
In Fig. 5, we plot the mean of UE wideband-SINR mapping
to the average UE throughput. The proposed scheduling method
is seen to obtain highest throughput for wideband SINR falling
in the range [5,20] dB compared to Docomo’s method and
OMA. For SINR>20dB, the throughput drops to lower values.
It is because the near user (with very high SINR) does not need
to select the highest modulation order and coding rate; instead it
selects an appropriate setting such that the best total throughput
for both near and far users can be achieved. Fig. 6 supports
the argument when we see that the spectral efficiency (bits/cu)
TABLE II: Comparison among NOMA scheduling methods
Scheme Average UE
throughput (Mb/s)
Edge UE
throughput (Mb/s)
Average UE
throughput gain (%)
Edge UE
throughput gain (%)
OMA 1.47 0.44
Docomo (ideal receiver) 1.76 0.93 19.73 111.36
Docomo (ML-receiver) 1.73 0.79 17.69 79.55
MCS-based (ML-receiver) 2.2 0.77 49.66 75
MCS-based (SLIC-receiver) 2.12 0.64 44.22 45.45
UE wideband SINR [dB]
-10-5 0 5 1015202530
average UE throughput [Mb/s]
0
2
4
6
8
10
12
NOMA-PF Scheduler-MCS-based method
NOMA-PF Scheduler-Docomo method
OMA-PF Scheduler
Fig. 5: UE wideband SINR-to-throughput mapping
UE wideband SINR [dB]
-10-5 0 5 1015202530
average UE specctral efficiency [bit/cu]
0
1
2
3
4
5
6
NOMA-PF Scheduler-MCS-based method
NOMA-PF Scheduler-Docomo method
OMA-PF Scheduler
Fig. 6: UE wideband SINR-to-spectral-efficiency mapping
obtained from the proposed method increases steadily when
the UE wideband SINR increases. In contrast, the Docomo’s
method has lower spectral efficiency for SINR>12dB.
V. C ONCLUDING REMARKS
In this paper, we have extended the Vienna LTE-A downlink
system-level simulator to support NOMA based on multiuser
superposition transmission and successive interference cancela-
tion techniques. The practical receiver model mapped from the
link-level simulator to the system-level simulator enables us
to investigate the resource allocation and scheduling algorithm
under NOMA more accurately. We have proposed a multi-users
NOMA wideband scheduling method by jointly considering
user pairing, power allocation, and MCS selection in the Vienna
LTE-A downlink system-level simulator. Simulation results
show that our algorithm outperforms the baseline approach in
terms of achieving higher average user throughput and better
spectral efficiency. Our future work is to extend for sub-band
NOMA scheduling methods.
ACKNOWLEDGMENT
This work was supported in part by funds from the Min-
istry of Science and Technology, National Taiwan University
and MediaTek Inc. under Grants MOST-104-2622-8-002-002,
MOST-104-2221-E-002-076-MY2, and NTU-ERP-105R8300.
REFERENCES
[1] Y. Saito, Y. Kishiyama, A. Benjebbour, T. Nakamura, A. Li, and
K. Higuchi, “Non-Orthogonal Multiple Access (NOMA) for Cellular
Future Radio Access,” in IEEE 77th Vehicular Technology Conference
(VTC Spring), June 2013, pp. 1–5.
[2] N. Otao, Y. Kishiyama, and K. Higuchi, “Performance of Non-Orthogonal
Access with SIC in Cellular Downlink using Proportional Fair-based
Resource Allocation,” in 2012 International Symposium on Wireless
Communication Systems (ISWCS), Aug 2012, pp. 476–480.
[3] NTT Docomo, “3GPP TSG RAN WG1 Meeting #81: Evaluation
Methodologies for Downlink Multiuser Superposition Transmissions, R1-
153332,” Fukuoka, Japan, Tech. Rep., May 2015.
[4] T. Seyama, T. Dateki, and H. Seki, “Efficient Selection of User Sets
for Downlink Non-Orthogonal Multiple Access,” in IEEE 26th Annual
International Symposium on Personal, Indoor, and Mobile Radio Com-
munications (PIMRC), Aug 2015, pp. 1062–1066.
[5] E. Okamoto, “An Improved Proportional Fair Scheduling in Downlink
Non-Orthogonal Multiple Access System,” in IEEE 82nd Vehicular
Technology Conference (VTC Fall), Sept 2015, pp. 1–5.
[6] Vienna LTE-A Downlink System-Level Simulator. [Online].
Available: https://www.nt.tuwien.ac.at/research/mobile-communications/
vienna-lte- a-simulators//
[7] MediaTek Inc, “3GPP TSG RAN WG1 Meeting #82: Link-abstraction
Method for ML Receiver in MUST, R1-154458,” Fukuoka, Japan, Tech.
Rep., May 2015.
[8] M. Mezzavilla, M. Miozzo, M. Rossi, N. Baldo, and M. Zorzi, “A
Lightweight and Accurate Link Abstraction Model for the Simulation
of LTE Networks in Ns-3,” in 15th ACM International Conference on
Modeling, Analysis and Simulation of Wireless and Mobile Systems
(MSWiM). New York, NY, USA: ACM, 2012, pp. 55–60.
[9] Z. Sun, C. Yin, and G. Yue, “Reduced-Complexity Proportional Fair
Scheduling for OFDMA Systems,” in 2006 International Conference on
Communications, Circuits and Systems, vol. 2, June 2006, pp. 1221–1225.
[10] 3GPP, “Study on Downlink Multiuser Superposition Transmission
(MUST) for LTE; (Release 13), TR 36.859 V1.0,” Tech. Rep., 2015.
