You are on page 1of 5

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/TVT.2018.2831213, IEEE
Transactions on Vehicular Technology

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, MONTH 2018 1

Successive Interference Cancellation for LDPC Thus, it has recently attracted significant attention from both
Coded Non-Orthogonal Multiple Access academia and industry, e.g., [3, 4].
Systems From the information-theoretical perspective, it has been
established that NOMA not only outperforms OMA by achiev-
ing a higher sum rate, but also achieves the capacity region
Lei Yuan, Member, IEEE, Jie Pan, Nan
of a downlink broadcast channel [5]. The core idea behind
Yang, Member, IEEE, Zhiguo Ding, Senior
power-domain NOMA is to (i) simultaneously serve multiple
Member, IEEE, and Jinhong Yuan, Fellow, IEEE
users by using superposition coding with different power
at the transmitter and (ii) cancel inter-user interference by
Abstract—We design, for the first time, novel successive inter- using successive interference cancellation (SIC) at the receiver.
ference cancellation (SIC) schemes for a two-user non-orthogonal In [6], the system-level performance gain of NOMA systems
multiple access (NOMA) system using low-density parity-check
over OMA systems was investigated within macro- and micro-
(LDPC) codes. We first develop a low-complexity SIC scheme
treating interference as noise, referred to as the N-SIC scheme, cells.
and a SIC with joint detection scheme treating the modulation Apart from superposition coding and SIC, power allocation
of interference signal at the near user as side information, (PA) and user pairing are also key design problems in NOMA
referred to as the J-SIC scheme. We then propose a new extrinsic since they ensure the fairness for users. Considering a typical
information assisted SIC scheme, referred to as the E-SIC
NOMA system which consists of a base station (BS) and two
scheme, which exchanges the extrinsic decoding information of
the signals for two users and thus incurs an increase complexity. user equipments (UEs), [7] proposed two PA algorithms to
With simulations, we demonstrate that the error performance maximize the ergodic capacity under a total power constraint
advantage of the E-SIC scheme over the N-SIC and J-SIC and the minimum rate requirement of the weak user. In [8],
schemes is more significant when the power allocated to the near the benefits of user pairing for NOMA were investigated.
user is larger, e.g., higher than 45%. When this power becomes
Specifically, it showed that NOMA offers a larger sum rate
smaller, e.g., less than 10%, the low-complexity N-SIC scheme
can achieve the best error performance. In addition, we have an than OMA and this rate gain can be enlarged by selecting the
important finding that when the two users adopt different order users whose channel conditions are more distinctive.
modulation, the decoding convergence speed of the E-SIC scheme Under the assumption of perfect SIC, NOMA with capacity-
improves when the difference between the constellation sizes of achieving channel coding has been shown to approach the
two users is larger.
capacity bound of multi-user systems. However, practical wire-
Index Terms—Non-orthogonal multiple access, successive in- less systems always use finite-length channel codes, causing
terference cancellation, low-density parity-check codes. decoding failure. Indeed, the imperfect SIC introduces severe
error propagation into NOMA systems and accordingly detri-
ments the system performance. Hence, appropriate channel
I. I NTRODUCTION
codes need to be applied into NOMA to reduce the detrimental
Non-orthogonal multiple access (NOMA) has been recog- impact of the imperfect SIC. Motivated by this, in this paper
nized as a highly promising multiple access technique for the we focus on the application of low-density parity-check (LD-
fifth generation (5G) wireless networks [1, 2]. Compared to PC) codes [9] to NOMA, since LDPC codes have been chosen
traditional orthogonal multiple access (OMA), NOMA pro- as the data channel coding scheme in 5G standards [10] due
vides higher capacity and better energy efficiency, as well as to its low encoding and decoding complexity and capacity-
supports massive connectivity by enabling users utilizing the achieving performance. Therefore, the study of LDPC coded
same time and frequency resource for information conveying. NOMA system is pivotal to meet the requirement of the
enhanced mobile broadband (eMBB) scenario in 5G. We
Copyright (c) 2015 IEEE. Personal use of this material is permitted. note that LDPC codes can be decoded iteratively using the
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending a request to pubs-permissions@ieee.org. belief propagation (BP) algorithm. This algorithm is optimal
Manuscript received May 22, 2017; revised November 7, 2017; accepted for achieving the best error performance when the bipartite
April 16, 2018. The work of L. Yuan and J. Pan was supported by the graph does not contain any cycle [11]. Due to this, some
Fundamental Research Funds for the Central Universities (lzujbky-2017-188).
The work of Z. Ding was supported by the UK EPSRC under grant number studies (e.g., [12, 13]) have explored the role of LDPC codes
EP/L025272/1 and by H2020-MSCA-RISE-2015 under grant number 690750. in the Gaussian multiple-access channel (GMAC). Consid-
The review of this paper was coordinated by Dr. A. Al-Dweik. (Corresponding ering a two-user GMAC, [12] proposed near-capacity bit-
author: Jie Pan.)
L. Yuan and J. Pan are with the School of Information Science and interleaved coded modulation (BICM) using amplitude phase
Engineering, Lanzhou University, Lanzhou 730000, Gansu, China (e-mail: shift keying, based on which LDPC codes were designed and
yuanl@lzu.edu.cn; panj15@lzu.edu.cn). optimized. Concentrating on the same channel, [13] derived
N. Yang is with the Research School of Engineering, Australian National
University, Canberra, ACT 2600, Australia (e-mail: nan.yang@anu.edu.au). the achievable rate regions for two correlated binary sources
Z. Ding is with the School of Electrical and Electronic Engineer- and developed iterative joint source channel decoding by using
ing, University of Manchester, Manchester, M13 9PL, UK (e-mail: the correlation information. Despite [12] and [13] are valued
Zhiguo.ding@manchester.ac.uk).
J. Yuan is with the School of Electrical Engineering and Telecommunica- on their own merits, they have focused on the multiple-access
tions, University of New South Wales, Sydney, NSW 2052, Australia (e-mail: channel, rather than the broadcast channel. Therefore, their
j.yuan@unsw.edu.au). results cannot be used in the downlink of NOMA systems.
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. Also, the users have adopted the same modulation scheme
Digital Object Identifier xx.xxxx/TVT.2018.xxxxxxx in [12, 13], which hinders the use of their results in practical

0018-9545 (c) 2018 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/TVT.2018.2831213, IEEE
Transactions on Vehicular Technology

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, MONTH 2018 2

scenarios where the users are very likely to adopt different additive white Gaussian noises with zero-mean and variance
modulation schemes. σ 2 at UE-1 and UE-2, respectively.
In this paper we design, for the first time, new SIC schemes We assume that UE-2 directly uses a decoding algorithm
for the downlink of a two-UE NOMA system using LDPC to decode x2 since it is allocated a higher transmit power and
codes. The SIC schemes for UE-1 (i.e., the near user) include able to treat x1 as noise. Differently, UE-1 can decode x1 only
(i) SIC which treats interference as noise, referred to as N- after removing x2 through SIC. We next detail the SIC process
SIC, (ii) SIC with joint detection which treats the modulation in Section III.
of interference signal as side information, referred to as J-SIC,
and (iii) extrinsic information assisted SIC which exchanges III. S UCCESSIVE I NTERFERENCE C ANCELLATION
the extrinsic information between the decoders at UE-1, re- S CHEMES
ferred to as E-SIC, where the decoders decode the signals for
both UE-1 and UE-2 (i.e., the far user). In our design, we Without loss of generality, we assume that UE-1 employs
adopt a general assumption that the two users in the system 16-/64-QAM modulation and UE-2 employs QPSK/16-QAM
can adopt different modulation schemes, e.g., UE-1 can use modulation. We also assume that UE-1 and UE-2 perform
16- or 64-ary quadrature amplitude modulation (QAM) while coherent detection and BP decoding. Since UE-2 does not
UE-2 can use quadrature phase shift keying (QPSK). Through carry out SIC, we only consider the SIC process at UE-1 in this
simulation results, we provide several pivotal insights into the section. In this process, UE-1 subtracts x2 to achieve the best
design of LDPC coded NOMA systems. First, when the power performance. We assume that UE-1 employs two decoders,
allocated to UE-1 is large (e.g., higher than 45%), the E-SIC namely, decoder 1 for decoding x1 and decoder 2 for decoding
scheme approaches the error performance of the benchmark x2 . We next describe SIC schemes in the following.
case where UE-1 perfectly cancels the interference signal from
UE-2. Second, when this power becomes small (e.g., less than A. N-SIC: SIC with Interference as Noise
10%), the low-complexity N-SIC scheme can be used since
In the N-SIC scheme, x1 is treated as the white Gaussian
it approaches the error performance of the benchmark case.
noise. As such, this scheme is the simplest scheme. For BP
Third, the decoding convergence speed of the E-SIC scheme
decoding, the demodulator needs to generate the channel log-
becomes higher when the difference between the constellation
likelihood ratio (LLR) corresponding to each transmitted bit
sizes of UE-1 and UE-2 is larger.
and pass it to the decoder. The channel LLR for decoding x2 ,
L (bj |r1 ), is written as
II. NOMA S YSTEM W ITH LDPC C ODES ( √ )
∑ |r1 −h1 P2 s|
2

We consider a downlink NOMA system where a BS com- S:bj (s)=0 exp − |h1 |2 P1 +σ 2
municates with two UEs. The BS and each UE are equipped L (bj |r1 ) = ln ( √ ) , (1)
∑ |r1 −h1 P2 s|
2

with a single antenna. The channel state information (CSI) is S:bj (s)=1 exp − |h |2 P +σ 2 1 1
perfectly known at both the BS and UEs. We assume UE-
1 and UE-2 as the near user and the far user from the BS, where S denotes the set of all the QPSK/16-QAM symbols
respectively. We denote hi as the channel between the √ BS for UE-2, s ∈ S, and bj is the jth bit of s. After computing
and UE-i, where i ∈ {1, 2}. Here, we define hi , ℓi fi , the channel LLR, the BP decoding algorithm is carried out to
where ℓi denotes the path-loss and fi denotes the small-scale recover x2 , as follows:
fading. In the considered system, it is reasonable to assume (l)
• Variable node update: The updated message mi,o from
|h1 |2 > |h2 |2 , indicating that the near user possesses a larger variable node i to check node o in the l-th iteration is
channel gain than the far user. Thus, the near user can use a calculated as
higher order modulation and coding scheme than the far user.
(l)
∑ (l−1)
We further denote P as the total transmit power at the BS. mi,o = Zi + mo′ ,i , (2)
Accordingly, we denote P1 = αP and P2 = (1 − α) P as the o′ ̸=o
transmit power for the near and far user, respectively, where where Zi denotes the channel LLR of the i-th coded bit
α is the PA factor. As per the rules of NOMA, the transmit and mo′ ,i denotes the message from check node o′ to
(l−1)
power of the far user needs to be larger than that of the near variable node i in the (l − 1)-th iteration. The summation
user, i.e., α < 0.5. is taken over the set of neighboring check nodes except
We assume that LDPC codes are applied in the downlink the message recipient itself.
NOMA system. In the system, the blocks of k information • Check node update: The updated message mo,i from
(l)
bits are firstly encoded by LDPC codes with the rate of check node o to variable node i in the l-th iteration is
R = k/n, to produce codewords of n bits. Then the M - calculated as
ary QAM is used to translate the encoded bits into a symbol ( (l) )
stream. The resultant symbol stream is sent ∏ mi′ ,o
√ from √ the BS to (l)
mo,i = 2tanh −1
tanh , (3)
UEs. Specifically, the BS transmits x̃ , P1 x1 + P2 x2 to ′
2
i ̸=i
UEs, where xi denotes the symbol for UE-i. The received
signals at UE-1 and UE-2 are expressed as r1 = h1 x̃ + n1 and where the product is taken over the set of neighboring
r2 = h2 x̃ + n2 , respectively, where n1 and n2 denote complex variable nodes except the message recipient itself.

0018-9545 (c) 2018 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/TVT.2018.2831213, IEEE
Transactions on Vehicular Technology

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, MONTH 2018 3

• Decoding decision: After l iterations, the LLR Zitot is complexity is proportional to the size of QAM constellation.
calculated as Here, a simple approach proposed in [14] can be adopted to
∑ (l) approximately compute the channel LLR involved in the J-SIC
Zitot = Zi + mo,i , (4)
scheme. After this computation, the decoding process is the
o
same as that of the N-SIC scheme.
where the summation is taken over the set of neighboring
check nodes. We clarify that decoding is terminated either
when the maximum number of decoding iterations is C. E-SIC: Extrinsic Information Assisted SIC
reached or when the decoded bits satisfy all the check We note that the aforementioned schemes may impose
constraints of LDPC codes. an error propagation effect on the NOMA system, which is
We denote the recovered x2 by x̂2 . Once it is recovered, it caused by decoding failure. To tackle this, we propose a
is subtracted from the signal r1 for decoding x1 . The resultant novel E-SIC scheme to improve the decoding performance. In
signal r̂1 is given by the proposed scheme, the extrinsic information is exchanged
(√ √ ) between decoder 1 and decoder 2 at UE-1, as follows: In each
r̂1 = h1 P1 x1 + P2 (x2 − x̂2 ) + n1 . (5)
iteration, x2 is firstly decoded. Then the soft outputs associated
If all the check constraints of LDPC codes are satisfied, x̂2 with the variable nodes produced by decoder 2 are fed back
is the same as x2 and thus, the interference from UE-2 can to decoder 1 as the a priori information. This is followed by
be fully eliminated. Otherwise, decoding failure occurs and the decoding iterations of decoder 1. Finally, the soft outputs
error propagation happens. In the presence of error propa- associated with the variable nodes produced by decoder 1
gation, if we assume that x2 and x̂2 are independent, the are fed back to decoder 2 as the a priori information. This
interference from UE-2 cannot be eliminated and addition- completes one iteration.
2
al interference x̂2 with power P2 |h1 | is introduced, i.e., To decode x2 , we rewrite the channel LLR given by (7) as
the signal-to-interference-plus-noise ratio (SINR) increases to ∑ ∑ ( )
|r1 −ψ|2
2 2
P1 |h1 | /(2P2 |h(1 | +)σ 2 ). As such, the channel LLR for Ŝ S:bj (s)=0 Pr (ŝ) exp − σ2
L̃ (bj |r1 ) = ln ∑ ∑ ( ) , (8)
decoding x1 , L̂ b̂j |r̂1 , is written as |r1 −ψ|2
Ŝ S:bj (s)=1 Pr (ŝ) exp − σ2
( √ )
∑ |r̂1 −h1 P1 ŝ|
2

( ) Ŝ:b̂j (ŝ)=0 exp − σ2


where Pr (ŝ) denotes the probability that a symbol is selected
L̂ b̂j |r̂1 = ln ( ), (6) from the set of M -ary QAM symbols for UE-1. At the very

∑ |r̂1 −h1 P1 ŝ|
2
first iteration, we set Pr (ŝ) = 1/M . After running decoder 2
Ŝ:b̂j (ŝ)=1 exp − σ 2
for l iterations, we are able to compute the output extrinsic
∑ (l)
LLR as Zia = o mo,i .
where Ŝ denotes the set of all the M -ary QAM symbols for
Different from the N-SIC and J-SIC schemes, the E-SIC
UE-1, ŝ ∈ Ŝ, and b̂j is the jth bit of ŝ. We note that (6) can still
be used when error propagation happens, since the decoding
scheme does not implement the subtraction ( ) operator. Thus,
results would contain a large amount of correct information the channel LLR for decoding x1 , L̄ b̂j |r1 , is written as
about x1 . Finally, the BP decoding algorithm is implemented ∑ ∑ ( )
|r1 −ψ|2
to recover x1 . S Ŝ:bj (ŝ)=0 Pr (s) exp − σ 2

Since the N-SIC scheme treats x1 as Gaussian noise, it L̄ (bj |r1 ) = ln ∑ ∑ ( ) , (9)
|r1 −ψ|2
only needs to know the power of UE-1 and thus leads to a S Ŝ:bj (ŝ)=1 Pr (s) exp − σ 2

simple calculation of channel LLR. However, we note that where Pr (s) denotes the probability that a symbol is selected
UE-1 employs QAM symbols which do not follow Gaussian from the set of QPSK/16-QAM symbols for UE-2. The value
distribution. Therefore, this scheme incurs a decoding perfor- of Pr (s) can be calculated by using the extrinsic information
mance loss. Motivated by this, we propose the J-SIC scheme Zia which is provided by decoder 2. After running decoder 1
and present it in the next subsection. for l iterations, the output extrinsic LLR Zia are fed back to
decoder 2 as the a priori information and used to calculate
B. J-SIC: SIC with Joint Detection the a priori probability Pr (ŝ). We perform q such iterations
In this scheme, the QAM structure of x1 is treated as before making the final decision.
side information. Hence, the channel LLR for decoding x2 , We note that decoder 1 and decoder 2 run once only for
L̂ (bj |r1 ), is written as the N-SIC and J-SIC schemes. Differently, such decoders run
∑ ∑ ( ) q times in the E-SIC scheme. This indicates that the E-SIC
|r1 −ψ|2
Ŝ S:bj (s)=0 exp − σ 2
L̂ (bj |r1 ) = ln ∑ ∑ ( ), (7) scheme achieves a better decoding performance, while causing
|r1 −ψ|2 a higher decoding complexity. Specifically, the decoding com-
Ŝ S:bj (s)=1 exp − σ 2

(√ √ ) plexity of the E-SIC scheme is q times those of the N-SIC and


where ψ = h1 P1 ŝ + P2 s . In order to calculate the J-SIC schemes. In fact, if the BP decoder correctly decodes
channel LLR in the J-SIC scheme, both the power and the x2 , the J-SIC scheme achieves the best decoding performance
QAM structure of x1 need to be known. Compared to the N- without causing any performance loss. Therefore, in order to
SIC scheme, the J-SIC scheme improves the decoding perfor- reduce the decoding complexity of the E-SIC scheme, we
mance at the cost of increasing computational complexity. This can further improve the E-SIC scheme by combining J-SIC

0018-9545 (c) 2018 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/TVT.2018.2831213, IEEE
Transactions on Vehicular Technology

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, MONTH 2018 4

0 0
10 10
BER (benchmark)
Start FER (benchmark)
−1 −1
10 BER (N-SIC) 10
FER (N-SIC)
BER (J-SIC) BER (benchmark)
−2 −2
10 10 FER (benchmark)
FER (J-SIC)

BER/FER

BER/FER
BER (N-SIC)
BER (E-SIC-1)
FER (N-SIC)
−3 FER (E-SIC-1) −3
10 10 BER (J-SIC)
BER (E-SIC-2)
FER (J-SIC)
FER (E-SIC-2) BER (E-SIC-1)
Calculate the channel −4
10 BER (E-SIC-4) 10
−4
FER (E-SIC-1)
LLR by using (8) and Yes FER (E-SIC-4) BER (E-SIC-2)
−5
BER (E-SIC-6) FER (E-SIC-2)
−5
implement BP 10 FER (E-SIC-6) 10

decoding for UE-2 −6 −6


10 6 7 8 9 10 11 12 13 14 10 10 11 12 13 14 15 16 17
SNR (dB) SNR (dB)
q MaxIter
(a) (b)
Fig. 2. Performance of UE-1 with n = 1008 and R = 0.5, and UE-2 uses
Decoding is No
QPSK. (a) BER and FER versus SNR with α = 0.4 and 16-QAM at UE-1.
successful No (b) BER and FER versus SNR with α = 0.4 and 64-QAM at UE-1.
0 0
10 10
BER (N-SIC)
FER (N-SIC)
−1 BER (J-SIC)
10
Yes q q+1 −1
10 FER (J-SIC)
Calculate the output BER (E-SIC-2)
extrinsic LLR Z ia −2
10 FER (E-SIC-2)

BER/FER

BER/FER
−2
BER (benchmark, 16-QAM) 10
−3 FER (benchmark, 16-QAM)
10
Calculate the resultant BER (N-SIC, 16-QAM)
−3
signal by using (5) −4 FER (N-SIC, 16-QAM) 10
Calculate the channel Calculate the
10
BER (benchmark, 64-QAM)
LLR by using (9) and output extrinsic −5
FER (benchmark, 64-QAM) −4
10
10
implement BP LLR Z ia
BER (N-SIC, 64-QAM)
FER (N-SIC, 64-QAM)
decoding for UE-1 −6 −5
10 6 7 8 9 10 11 12 13 100.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
SNR (dB) α

Calculate the channel No (a) (b)


LLR by using (6) and Decoding is
implement BP Fig. 3. Performance of UE-1 with n = 1008 and R = 0.5, and UE-2 uses
successful
decoding for UE-1 QPSK. (a) BER and FER versus SNR with α = 0.3 and 16-/64-QAM at
UE-1. (b) BER and FER versus PA factor with 64-QAM at UE-1.
Yes

rate (BER) and frame error rate (FER) versus SNR when UE-
1 uses 16-QAM and α = 0.4. We first observe that the J-SIC
Output the hard
decision
scheme achieves a better decoding performance than the N-
SIC scheme. For example, when SNR = 10 dB, the J-SIC
Fig. 1. Flowchart of the E-SIC scheme. scheme achieves a BER of 1.5 × 10−3 whereas the N-SIC
scheme achieves a BER of 2.5 × 10−2 . Second, we observe
with the iteration process described in this subsection. The a large decoding performance gap between these two SIC
flowchart of our proposed E-SIC scheme is depicted in Fig. 1, schemes and the benchmark case. Third, we observe that the
where ‘MaxIter’ denotes the maximum number of decoding E-SIC scheme outperforms the other two SIC schemes. More
iterations. We initialize the number of iterations q to 1. importantly, the performance of the E-SIC scheme becomes
closer to that of the benchmark case when MaxIter increases.
For example, when SNR = 8 dB, the BER of the benchmark
IV. S IMULATION R ESULTS
case is 3 × 10−5 and the BER of the E-SIC scheme with
In this section, we evaluate the performance of three SIC MaxIter = 6 is 8 × 10−5 .
schemes through Monte Carlo simulations. Without loss of We next examine the impact of the order of modulation
generality, the channels are set as |h1 |2 = 10 dB and |h2 |2 = on the BER and FER performance. In Fig. 2(b), we plot the
0 dB in all the figures. Moreover, the parity-check matrix of BER and FER versus SNR when UE-1 uses 64-QAM and
regular LDPC codes is constructed using the progressive edge- α = 0.4. We find that the decoding convergence speed of
growth algorithm [15]. The maximum number of BP decoding the E-SIC scheme with 64-QAM is higher than that of the
iterations is 50. In addition, both UE-1 and UE-2 adopt Gray E-SIC scheme with 16-QAM. Moreover, the E-SIC scheme
mapping in modulation. with MaxIter = 2 and 64-QAM achieves almost the same
For an LDPC coded NOMA system, UE-1 achieves the performance as the benchmark case. Given the performance
best error performance if it perfectly cancels the interference gap between the E-SIC scheme with MaxIter = 6 and 16-
signal from UE-2. For the performance comparison, therefore, QAM and the benchmark case, as indicated in Fig. 2(a),
a benchmark case is built by adopting the assumption that we find that the benefit of the E-SIC scheme becomes more
the interference signal from UE-2 is perfectly cancelled. In obvious when the difference of constellation sizes between two
all the simulation results, we represent the N-SIC scheme by UE receivers becomes large.
“N-SIC”, the J-SIC scheme by “J-SIC”, and the E-SIC scheme Next, we examine the impact of a decreasing PA factor on
with MaxIter = X by “E-SIC-X”. Furthermore, we define the the BER and FER. Fig. 3(a) plots the BER and FER versus
2
signal-to-noise ratio (SNR) for UE-1 as SNR , P1 |h1 | /σ 2 . SNR for α = 0.3. In this subfigure, we consider both 16-
In Figs. 2 and 3, both UE-1 and UE-2 adopt (3, 6)-regular QAM and 64-QAM at UE-1. We observe that the decoding
LDPC codes with code length n = 1008 and rate R = 0.5, performance of the low-complexity N-SIC scheme for UE-1
and UE-2 uses QPSK modulation. Fig. 2(a) plots the bit error is almost the same as that of the benchmark case, regardless

0018-9545 (c) 2018 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/TVT.2018.2831213, IEEE
Transactions on Vehicular Technology

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, MONTH 2018 5

0 0
10 10

−1 −1
V. C ONCLUSION
10 10

−2 −2
In this paper, we proposed novel SIC schemes for an LDPC
10 BER (benchmark) 10
coded NOMA system. Specifically, the proposed extrinsic
BER/FER

FER (benchmark)

BER/FER
BER (N-SIC, α = 0.2)
−3 −3
10 FER (N-SIC, α = 0.2)
BER (J-SIC, α = 0.2)
10 BER (benchmark)
FER (benchmark)
BER (N-SIC, α = 0.45)
information assisted SIC scheme (i.e., the E-SIC scheme)
FER (J-SIC, α = 0.2)
10
−4
BER (E-SIC-4, α = 0.2)
FER (E-SIC-4, α = 0.2)
10
−4 FER (N-SIC, α = 0.45)
BER (J-SIC, α = 0.45) overcomes the disadvantage of the SIC with interference as
FER (J-SIC, α = 0.45)

10
−5
BER (E-SIC-6, α = 0.2)
FER (E-SIC-6, α = 0.2) 10
−5
BER (E-SIC-2, α = 0.45)
FER (E-SIC-2, α = 0.45)
noise scheme (i.e., the N-SIC scheme) and the SIC with
BER (N-SIC, α = 0.1) BER (N-SIC, α = 0.4)

−6
10 10
FER (N-SIC, α = 0.1)
−6
1013.5
FER (N-SIC, α = 0.4) joint detection scheme (i.e., the J-SIC scheme) by combating
10.5 11 11.5 12 12.5 13 13.5 14 14.5 15 15.5 16 16.5 17
SNR (dB) SNR (dB) the error propagation effect. Through simulations results,
(a) (b) we demonstrated that the proposed E-SIC scheme achieves
Fig. 4. Performance of UE-1. (a) BER and FER versus SNR with n = 1008 a better decoding performance than the N-SIC and J-SIC
and 16-QAM at UE-1 and UE-2. (b) BER and FER versus SNR with n = schemes when the PA factor is large (e.g., higher than 0.45).
1944, 64-QAM at UE-1, and QPSK at UE-2.
We also showed that when the maximum number of decoding
iterations increases, the decoding performance of the proposed
of the modulation adopted by UE-1. This observation is due to E-SIC scheme approaches that of the benchmark case in which
the large power difference between UE-1 and UE-2. Fig. 3(b) the interference signal from UE-2 is perfectly cancelled. We
plots the BER and FER versus PA factor when UE-1 uses 64- further showed that the decoding convergence speed of the
QAM and SNR = 12.5 dB. We observe from Fig. 3(b) that proposed E-SIC scheme is higher in the presence of larger
when α < 0.35, the N-SIC scheme achieves almost the same difference of constellation sizes between two UEs.
performance as the benchmark case, given that the BER and
FER in the benchmark case when UE-1 uses 64-QAM and R EFERENCES
SNR = 12.5 dB are 8.6 × 10−5 and 1.6 × 10−3 , respectively, [1] Z. Ding, Y. Liu, J. Choi, Q. Sun, M. Elkashlan, C.-L. I, and H. V.
as indicated in Fig. 2(b). Therefore, when α is small, the Poor, “Application of non-orthogonal multiple access in LTE and 5G
networks,” IEEE Commun. Mag., vol. 55, no. 2, pp. 185–191, Feb. 2017.
use of the N-SIC scheme is sufficient. It follows that using [2] Z. Wei, J. Yuan, D. W. K. Ng, M. Eklashlan, and Z. Ding, “A
the high-complexity J-SIC scheme or the E-SIC scheme does survey of downlink non-orthogonal multiple access for 5G wireless
not provide any considerable improvement in the decoding communication networks,” ZTE Commun., vol. 14, no. 4, pp. 17–25,
Oct. 2016.
performance. [3] Z. Ding, F. Adachi, and H. V. Poor, “The application of MIMO to non-
In order to further study the impact of the order of mod- orthogonal multiple access,” IEEE Trans. Wireless Commun., vol. 15,
no. 1, pp. 537–552, Jan. 2016.
ulation of UE-2 on the decoding performance of UE-1, we [4] Y. Yuan, Z. Yuan, G. Yu, C.-H. Hwang, P.-K. liao, A. Li, and K.
consider that both UE-1 and UE-2 adopt 16-QAM in Fig. 4(a). Takeda, “Non-orthogonal transmission technology in LTE evolution,”
In this figure, UE-2 still adopts (3, 6)-regular LDPC codes, IEEE Commun. Mag., vol. 152, no. 6, pp. 68–74, July 2016.
[5] T. Cover, “Broadcast channels,” IEEE Trans. Inf. Theory, vol. 18, no. 1,
while UE-1 adopts (3, 18)-regular LDPC codes with rate pp. 2–14, Jan. 1972.
R = 5/6. We observe that when α = 0.2, the E-SIC scheme [6] Y. Saito, A. Benjebbour, Y. Kishiyama, and T. Nakamura, “System-
with MaxIter = 6 still exhibits a worse performance than the level performance of downlink non-orthogonal multiple access (NOMA)
under various environments,” in Proc. IEEE VTC Spring 2015, Glasgow,
benchmark case in the regime of medium SNR. For example, Scotland, May 2015, pp. 1–5.
when SNR = 13 dB, the benchmark case achieves a BER of [7] Q. Sun, S. Han, C.-L. I, and Z. Pan, “On the ergodic capacity of MIMO
5.7×10−6 , while the E-SIC scheme with MaxIter = 6 achieves NOMA systems,” IEEE Wireless Commun. Lett., vol. 4, no. 4, pp. 405–
a BER of 5.3 × 10−3 . Only when α further decreases (e.g., α 408, Aug. 2015.
[8] Z. Ding, P. Fan, and H. V. Poor, “Impact of user pairing on 5G non-
decreases to 0.1), the performance of the N-SIC scheme for orthogonal multiple-access downlink transmissions,” IEEE Trans. Veh.
UE-1 becomes closer to that of the benchmark case. Therefore, Technol., vol. 65, no. 8, pp. 6010–6023, Aug. 2016.
[9] T. J. Richardson and R. Urbanke, “The capacity of low-density parity-
by comparing Fig. 4(a) with Fig. 2(b) we find that the benefits check codes under message-passing decoding,” IEEE Trans. Inf. Theory,
of three SIC schemes are more obvious for a large PA factor vol. 47, no. 2, pp. 599–618, Feb. 2001.
(or equivalently, a higher transmit power allocated to UE-1) [10] X. Wu, M. Jiang and C. Zhao, “A parity structure for scalable QC-LDPC
codes with all nodes of degree three,” IEEE Commun. Lett., vol. 21, no.
and the large difference of constellation sizes between two UE 9, pp. 1913–1916, Sep. 2017.
receivers. [11] F. R. Kschischang, B. J. Frey, and H. A. Loeliger, “Factor graphs and
the sum-product algorithm,” IEEE Trans. Inf. Theory, vol. 47, no. 2, pp.
We note that irregular quasi-cyclic (QC) LDPC codes [16] 498–519, Feb. 2001.
have been used in many practical wireless communication [12] S. Chen, K. Peng, Y. Zhang, and J. Song, “Near capacity LDPC
standards, such as IEEE 802.16e and IEEE 802.11n. Therefore, coded MU-BICM-ID for 5G,” in Proc. IEEE IWCMC 2015, Dubrovnik,
Croatia, Aug. 2015, pp. 1418–1423.
we now provide simulation results for the QC LDPC coded [13] J. He, I. Hussain, M. Juntti, and T. Matsumoto, “Transmission of
NOMA system. In Fig. 4(b), both users adopt QC LDPC codes correlated sources over non-orthogonal Gaussian MACs,” in Proc. IEEE
used in IEEE 802.11n with code length n = 1944. UE-1 ICC 2016 Wksp, Kuala Lumpur, Malaysia, May 2016, pp. 534–539.
[14] S. J. Park, “Bitwise log-likelihood ratios for quadrature amplitude
adopts 64-QAM and R = 3/4, while UE-2 adopts QPSK modulations,” IEEE Commun. Lett., vol. 19, no. 6, pp. 921–924, June
and R = 0.5. Based on Figs. 2, 3, and 4, we find that the 2015.
threshold value of α which determines the best SIC scheme, [15] X. Y. Hu, E. Eleftheriou, and D. M. Arnold, “Regular and irregular
progressive edge-growth tanner graphs,” IEEE Trans. Inf. Theory, vol.
out of three SIC schemes, depends on the modulation orders 51, no. 1, pp. 386–398, Jan. 2005.
and code parameters (i.e., code rate and code length) of two [16] A. Mahdi and V. Paliouras, “A low complexity-high throughput QC-
UEs. We clarify that this threshold can be found via simulation LDPC encoder,” IEEE Trans. Signal Process., vol. 62, no. 10, pp. 2696–
2708, May 2014.
for a given set of parameters.

0018-9545 (c) 2018 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.

You might also like