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/LCOMM.2016.2642921, IEEE
Communications Letters
1
User Grouping and Power Allocation for NOMA
Visible Light Communication Multi-cell Networks
Xiaoke Zhang, Qian Gao, Chen Gong and Zhengyuan Xu
Abstract—To design an efficient multiple access scheme for
visible light communication (VLC) multi-cell networks, this letter
leverages the non-orthogonal multiple access (NOMA), which
has received significant attention in the 5th generation wireless
communication. A user grouping based on user locations is
proposed to reduce interference for VLC multi-cell networks.
With the residual interference from the successive interference
cancellation in NOMA taken into account, we optimize the power
allocation within each cell to improve the achievable user rate
under user quality of service constraint. The performance of the
proposed approaches is evaluated by the numerical results.
Index Terms—visible
light
orthogonal multiple access (NOMA), multi-cell networks.
communication (VLC), non-
I. INTRODUCTION
With a rapid growth in portable information terminals, the
demand for high rate wireless data communication in local
area networks keeps increasing. Although radio frequency
(RF) has been widely commercialized for communication
purposes because of the wide area coverage and little interfer-
ence in frequency band [1], the limited spectrum cannot well
accommodate the increasing communication rate requirement.
Visible light communication (VLC) has received significant
interest due to its advantages in unlicensed spectrum, natu-
ral confidentiality, convenient deployment, low energy con-
sumption, etc. VLC networks is also considered to be an
environmental-friendly solution for smart home networking.
Consider the realistic network application where multiple
VLC access points (VAPs) are placed in the room ceiling
and multiple mobile users are randomly distributed within
the circular coverage area underneath. Distinctive from RF
communications,
in VLC is significantly smaller
for the sharp signal attenuation and restricted user field of
view (FOV), which brings a big challenge for dealing with
the interference in multi-cell networks. A few terminal ac-
cess approaches have been proposed. A graph theory based
scheduling method is proposed in [2] where the users are
the cell
This work was supported by National Key Basic Research Program of China
(Grant No. 2013CB329201), Key Program of National Natural Science Foun-
dation of China (Grant No. 61631018), National Natural Science Foundation
of China (Grant No. 61501420), Key Research Program of Frontier Sciences
of CAS (Grant No. QYZDY-SSW-JSC003), Key Project
in Science and
Technology of Guangdong Province (Grant No. 2014B010119001), Shenzhen
Peacock Plan (No. 1108170036003286), and the Fundamental Research Funds
for the Central Universities.
The authors are with Key Laboratory of Wireless-Optical Communications,
Chinese Academy of Sciences, University of Science and Technology of
China, Hefei, Anhui 230027, China. Z. Xu is also with Shenzhen Grad-
uate School, Tsinghua University, Shenzhen 518055, China. Email: zxi-
aoke@mail.ustc.edu.cn, {qgao, cgong821, xuzy}@ustc.edu.cn.
selected to access the network according to proportional fair-
ness priority factor. Furthermore, [3] proposes a user-centric
cluster formation technique employing vectored transmission
to allow each multiple-access-point cell serving multiple users
simultaneously.
On the other hand, non-orthogonal multiple access (NOMA)
has been recently suggested as a promising solution in the
5th generation (5G) wireless networks [4]. Multiple users
are multiplexed in the power domain on the transmitter side
by superposition coding and multi-user signal separation is
accomplished on the receiver side by successive interference
cancellation (SIC) at the receiver side. NOMA is demonstrated
to outperform the orthogonal multiple access in terms of the
ergodic sum rate and the user outage probability [5].
Due to the limited coverage which well controls limited user
number in each cell, the computational complexity of the SIC
can be well controlled. Other advantages of NOMA for the
VLC network include high signal-to-noise ratio (SNR), easily
adjusted optical gains and the slow-fading channel such that
the channel state information is available at the transceiver
side. With fixed power allocation, the performance of NOMA
for a single VAP is investigated in [6]. The channel dependent
gain ratio power allocation is proposed in [7] to ensure
efficiency and fairness compared to static power allocation
approach, where the scenario of two VAPs are considered.
In this letter, we propose a user grouping based on user
locations for NOMA VLC multi-cell networks, as high precise
positioning is feasible in VLC [8]. We also propose two types
of quality of service (QoS) guaranteed power allocation within
each cell to iteratively optimize the sum user rate or the max-
min user rate. The residual interference during the SIC in
NOMA is taken into account in the formulated optimization
problem.
II. NOMA VLC MULTI-CELL NETWORKS
A. User grouping in NOMA VLC Multi-cell Networks
A typical NOMA VLC network is shown in Figure 1.
The red dots denote the VAPs in the network. The coverage
area consists of four types. Area type Li; i 2 f1; 2; 3; 4g
represents that the corresponding hatched area can receive
line of sight (LOS) signal of i VAPs. The neighboring cells
are distinguished by solid line and dashed line. The cell size
is determined by the user FOV, vertical distance, attenuation
coefficient and so on. The interference is influenced by the
following factors:
1) User FOV: When the user FOV is sufficiently small, no
more than one VAP exists in the FOV which will cer-
tainly eliminate the inter-cell interference. Meanwhile,
1089-7798 (c) 2016 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/LCOMM.2016.2642921, IEEE
Communications Letters
2
g„ k” with refractive index n is given by n2sin2„c” if
0 k c and 0 otherwise, where c is the concentrator
FOV semiangle [1].
Without loss of generality, assume that the K users in a
particular cell are sorted based on the link gain as h1 h2
hK. Let sk denote the message that the VAP delivers
to the k th user. According to the NOMA protocol, fsk; k =
1; 2; : : : ; Kg is superposed and transmitted by the VAP as
Pelecsi + IDC;
(2)
K∑
√
x =
ai
i=1
Fig. 1. NOMA VLC network model
this would cause communication failure for the absence
of LOS optical signal in coverage holes.
2) User Distribution: While the inter-cell interference can
be avoided without co-frequency VAPs working, this
condition depends on the time-varying user distribution.
3) Frequency Reuse (FR) Factor: Although FR = 1 can
achieve the highest spectrum efficiency, users located
in areas L2; L3; L4 in Figure 1 suffer from severe in-
terference where sophisticated interference cancellation
or user scheduling technique is in need. This situation
would be significantly improved when FR = 2 since
only users in area type L4 can receive co-frequency
interference.
To balance the interference management requirements and
the frequency efficiency, we adopt FR = 2 in this letter.
Based on the user positions, the user grouping for NOMA
VLC networks is designed as follows. The users in L1 and
L3 are assigned to the VAP without
interference. As the
probability of users existing in L4 is reasonably small, the
users in L4 are allocated with special bandwidth Bi to avoid
the interference. Since a user in L2 can select either VAP1
or VAP2 for access without interference, they are scheduled
for load balancing among the VAPs and 1d„V APi” may be
regarded as the priority, where d„V APi” denotes the number
of users connected to VAPi.
B. System Model
The power of the reflected signal is usually much weaker
than that of the LOS signal and thus can be neglected. The
VAP is placed at height Lk above the users. The k th user,
denoted as Uk, is located on a polar coordinate plane at the
distance dk from the light-emitting diode (LED), and the LED
irradiance angle and the photodiode (PD) incidence angle are
given by ϕk and k, respectively. According to the Lambertian
emission model, the channel gain of the optical link between
the VAP and the k th user, denoted as hk, is given by
cosm„ϕk”T„ k”g„ k” cos„ k”;
(1)
where A denotes the detection area of the PD, T„ k” represents
the gain of the optical filter, m is the order of Lamber-
tian emission relying on the transmitter semiangle 12 by
m = ln 2ln„cos 12”. The gain of nonimaging concentrator
A„m + 1”
2d2
k
hk =
where IDC is the DC bias added to ensure the positive instan-
taneous intensity and ai is the power allocation coefficient for
ith user. Therefore, the observation at the k th user is given by
K∑
√
yk = hk
ai
i=1
Pelecsi + nk;
(3)
where nk denotes the additive real-valued Gaussian noise with
zero mean and variance 2
k including the shot noise and the
thermal noise [9].
In NOMA, users with lower channel gain will be allocated
more power, i.e. a1 a2 aK. At the receiver side,
the user would perform SIC [5]. Note that the SIC requires
highly accurate channel and signal estimation otherwise the
non-negligible residual interference remains. Although optical
wireless channel is practically stationary, the channel estima-
tion error can still exist due to the feedback delay and user
mobility which leads to the residual interference within the
process of SIC. Let " represent the fraction of user’s power
not cancelled [10] and in the process of SIC, the k th user is
supposed to detect the message for the j th (j k) user with
the observed signal yk!j, given by
„hkai” + "
„hkai” + nk :
K∑
j1∑
yk!j =
(4)
i=j
i=1
To guarantee the success of SIC, the k th user is supposed to
decode the message for the j th (j k) user. The achievable
rate of the j th user’s message at user k, denoted as ˜Rk!j, is
given by
)
˜Rk!j =
K
1 +
∑
i= j +1„hk ai”2+"
j k; j , K;
∑ j1
„hK a j”2
„hK ai”2+1
∑ j1
„hk a j”2
)
1 +
i=1
"
i=1
B
2 log2
B
2 log2
„hk ai”2+1
Tj;
Tj;
j = k = K;
(5)
where Tj denotes the targeted data rate satisfying the j th
user’s QoS requirements, = 2 TSN R, TSN R = PelecN0B
denotes the transmitted signal-to-noise ratio (TSNR) [5][6], B
denotes the transmission bandwidth, denotes the photoelec-
tric conversion efficiency. The scaling factor 12 indicates the
spectral efficiency is lost due to the Hermitian symmetry.
Note that in Eq. (5), if k1 k2 i, we have ˜Rk1!i
˜Rk2!i ˜Ri where ˜Ri ≜ ˜Ri!i since a1 a2 aK.
Therefore, Eq. (5) can be further simplified as
k = 1; 2; ; K:
˜Rk Tk;
(6)
(
(
8>>>>>><>>>>>>:
1089-7798 (c) 2016 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.
VAP1VAP2VAP3VAP4VAP1VAP2VAP3VAP4L3L1L4L2
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/LCOMM.2016.2642921, IEEE
Communications Letters
3
k = 1; 2; : : : ; K;
(7)
= 1;
k = 1; 2; : : : ; K:
:
˜Rk is a function of fak; ak+1; : : : ; aKg, which
Notice that
leads to the objective parameters coupling in the original
optimization problem and brings difficulty to analysis. Perform
i=k a2
variable substitution as pk ≜
; k = 1; 2; : : : ; K and Eq.
i
(5) can be equivalently expressed as
K
maximize
s.t.
k
K
k=1
∑
˜Rk;
˜Rk Tk;
k=1 a2
ak 0;
K∑
8>>><>>>:
∑
( „1 "”pk + mk
(„1 "”pK + mK
)
(
∑
h2
k
log2
pk+1 "pk + mk
k = 1; 2; : : : ; K 1;
log2
"pK + mK
k = K;
)
)
8>>>>>>>><>>>>>>>>:
B
2
B
2
˜Rk =
It can be seen from Eq. (5) and (6) that the power alloca-
tion parameters fa1; a2; : : : ; aKg jointly determine each user’s
achievable rate and thus may nontrivially affect the corre-
sponding modulation and coding scheme for data transmission
of each user. As the power allocation plays a key rule in
NOMA, we investigate the QoS-guaranteed power allocation
strategies for NOMA in Section III.
III. QOS-GUARANTEED NOMA POWER ALLOCATION
A. QoS-guaranteed Max-Sum Rate Criterion
NOMA can support a flexible management of user rate
and provide an efficient way to ensure fairness by adjusting
power allocation coefficients. We aim to maximize the sum
of user achievable rate by optimizing the power allocation
while satisfying the basic QoS requirements. With Eq. (6), the
corresponding optimization problem is formulated as follows,
≜ Gk„pk; pk+1”;
≜ GK„pK”;
;
(8)
; k 2 f1; 2; : : : ; Kg. Then the achiev-
where mk = " + 1
able sum rate of users is denoted as ˜Rtot al and transformed
k=1 Gk„pk; pk+1” + GK„pK”.
K1
into separable form as ˜Rtot al =
The constraints in problem Eq. (7) form the feasible region
D. Clearly, these constraints are linear and are hence convex.
However, the objective function is not convex which is difficult
to solve directly using standard optimization solvers. We de-
velop a gradient projection (GP) algorithm [11] which includes
a gradient descending process and a projection process.
Let p denote the variable p = „p2; ; pK”. The gradient
descending process iteratively takes steps in the direction of
the gradient of the objective function at a given position
yielding
where the superscript „”„i” denotes the iteration time,ep denotes
variableep
the variable with step added and i denotes the step size which
can be chosen by backtracking line search [11]. When the
„i+1” steps out of D, it is mapped into D by finding
the nearest feasible point in D. The corresponding projection
process is described as a convex optimization problem whose
„i” + i @ ˜Rtot al„p”
„i+1” = p
p=p„i”;
ep
(9)
@p
TABLE I
SIMULATION PARAMETERS
Parameter name, notation
VAP height, H
User height, z
Semi-angle at half power, ϕ12
Information signal power, Pel e c
Signal bandwidth, B
Noise power spectral density, N0
PD detection area, A
PD responsivity,
PD FOV, f ov
Optical filter gain, T„ ”
Refractive index, n
Value
3 m
0:85 m
60◦
1:25 mW
20 MHz
1021 A2/Hz
0:28 A/W
1 cm2
32◦
1
1:5
solution can be efficiently obtained utilizing the standard
solver such as CVX [12] embedded with MATLAB R⃝.
B. QoS-guaranteed Max-min Rate Criterion
Distinct from the fairness criterion investigated in [13]
where the minimum of user achievable rate is maximized,
we consider additional QoS requirement of each user and
take the residual interference into account. We propose the
QoS-guaranteed max-min rate criterion with the associated
optimization problem formulated as follows
max
fa1;a2;:::;aK g min
˜Ri;
k = 1; 2; : : : ; K;
8>>>>><>>>>>:
K∑
i2f1;2;:::;Kg
˜Rk Tk;
a2
ak 0;
k=1
k
s.t.
= 1;
k = 1; 2; : : : ; K;
(10)
(11)
With the additional QoS constraints, this problem can still
be solved following the searching algorithm in [13] which is
time-consuming to obtain a solution with desired accuracy.
Instead, we adopt GP algorithm to dynamically adjust the
power allocation parameters in a short interval. To overcome
the non-differentiability of objective function in Eq. (10), we
adopt the following approximation [14],
)
i2f1;2;:::;Kg Ri ≃ lim
min
!+1
1
ln
exp„Ri”
:
(12)
(
K∑
i=1
IV. NUMERICAL RESULTS
In this section, we evaluate the performance of our proposed
NOMA VLC multi-cell network parameterized as in Table I.
Assume that the K users simulated are uniformly distributed
in the 1:8m1:8m square area illustrated in the upper right
corner of Fig. 1.
Fig. 2 shows a higher achievable sum user rate can be
obtained with NOMA compared to OMA for either FR = 2 or
FR = 4. We select 30 random user distributions for each case
and take the average value. Adopting FR = 2 will drastically
improve the sum user rate for doubled available bandwidth
in each cell. Although the residual interference during the
SIC can degrade the performance of NOMA, NOMA can still
outperform OMA when " = 0:1. A higher user rate can be
achieved with smaller " which emphasizes the importance
1089-7798 (c) 2016 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/LCOMM.2016.2642921, IEEE
Communications Letters
4
Fig. 2. The maximized sum user rate comparison for the NOMA VLC
network under different user number K when Tk = 3.
Fig. 4. The change of the maximized minimum user rate in the process of
algorithm iteration when Tk = 3 and K = 5.
QoS-guaranteed power allocation within each cell on either
max-sum rate criterion or max-min rate criterion. By virtue
of the GP algorithm, the power allocation coefficients can be
dynamically adjusted.
REFERENCES
Fig. 3. The effect of parameters on the sum user rate for the NOMA VLC
network when Tk = 1 and K = 5.
of the channel estimation accuracy. When the user number
increases, the sum user rate first grows because the probability
of more VAPs in service is improved.
Fig. 3 investigates the effect of parameters on the sum user
rate of the NOMA VLC multi-cell network. Due to larger
FOV, the more overlapping area of cells will enhance the
interference which results in the decrease of sum user rate. The
sum user rate is drastically increased when N0 is improved by
one order of magnitude.
We also verify the efficiency of our proposed QoS-
guaranteed max-min user rate criterion. Assume 3 users
are distributed in a particular cell with channel gains h =
»3:683; 4:877; 4:966… 105. The maximized minimum user
rate is achieved at R1 = »10:85; 11:26; 11:02… for " = 0:1
and T1 = »10; 10; 10…. When one of the users requests higher
transmission rate as T2 = »20; 5; 5…,
the power allocation
solution achieves a user rate of R2 = »20:00; 6:52; 6:76….
Obviously, the minimum user rate changes from 10:85 to 6:52
due to the QoS constraints. Fig. 4 illustrates the max-min user
rate is optimized in the process of iteration. Our adopted Eq.
(12) has a good approximation performance.
V. CONCLUSION
In this letter, the user grouping and the power allocation
for NOMA VLC networks are proposed. With residual in-
terference during SIC taken into account, we investigate the
[1] D. Karunatilaka, F. Zafar, V. Kalavally, and R. Parthiban, “LED based
indoor visible light communications: State of the art,” IEEE Commu-
nications Surveys Tutorials, vol. 17, no. 3, pp. 1649–1678, thirdquarter
2015.
[2] Y. Tao, X. Liang, J. Wang, and C. Zhao, “Scheduling for indoor visible
light communication based on graph theory,” Optics express, vol. 23,
no. 3, pp. 2737–2752, Feb 2015.
[3] X. Li, F. Jin, R. Zhang, J. Wang, Z. Xu, and L. Hanzo, “Users first:
User-centric cluster formation for interference-mitigation in visible-light
networks,” IEEE Transactions on Wireless Communications, vol. 15,
no. 1, pp. 39–53, Jan 2016.
[4] Z. Ding, Y. Liu, J. Choi, Q. Sun, M. Elkashlan, C. I, and H. V.
Poor, “Application of non-orthogonal multiple access in LTE and
5G networks,” To appear in IEEE Communications Magazine, 2017.
[Online]. Available: http://arxiv.org/abs/1511.08610
[5] Z. Ding, Z. Yang, P. Fan, and H. V. Poor, “On the performance of
non-orthogonal multiple access in 5G systems with randomly deployed
users,” IEEE Signal Processing Letters, vol. 21, no. 12, pp. 1501–1505,
Dec 2014.
[6] L. Yin, X. Wu, and H. Haas, “On the performance of non-orthogonal
multiple access in visible light communication,” in Personal, Indoor,
and Mobile Radio Communications (PIMRC), 2015 IEEE 26th Annual
International Symposium on, Aug 2015, pp. 1354–1359.
[7] H. Marshoud, V. M. Kapinas, G. K. Karagiannidis, and S. Muhaidat,
“Non-orthogonal multiple access for visible light communications,”
IEEE Photonics Technology Letters, vol. 28, no. 1, pp. 51–54, Jan 2016.
[8] H. S. Kim, D. R. Kim, S. H. Yang, Y. H. Son, and S. K. Han, “An
indoor visible light communication positioning system using a RF carrier
allocation technique,” Journal of Lightwave Technology, vol. 31, no. 1,
pp. 134–144, Jan 2013.
[9] J. M. Kahn and J. R. Barry, “Wireless infrared communications,”
Proceedings of the IEEE, vol. 85, no. 2, pp. 265–298, Feb 1997.
[10] J. G. Andrews and T. H. Meng, “Optimum power control for succes-
sive interference cancellation with imperfect channel estimation,” IEEE
Transactions on Wireless Communications, vol. 2, no. 2, pp. 375–383,
Mar 2003.
[11] S. Boyd and L. Vandenberghe, Convex optimization.
Cambridge
[13] S. Timotheou and I. Krikidis, “Fairness for non-orthogonal multiple
access in 5G systems,” IEEE Signal Processing Letters, vol. 22, no. 10,
pp. 1647–1651, Oct 2015.
[14] R. J. Drost and B. M. Sadler, “Constellation design for channel prec-
ompensation in multi-wavelength visible light communications,” IEEE
Transactions on Communications, vol. 62, no. 6, pp. 1995–2005, June
2014.
[12] M. Grant, S. Boyd, and Y. Ye, “CVX: Matlab software for disciplined
University Press, 2004.
convex programming,” 2008.
1089-7798 (c) 2016 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.
6810122122232425262728User NumberSum User Rate [Mbps](a) FR=4 OMANOMA, ε=0.05NOMA, ε=0.16810124244464850525456User NumberSum User Rate [Mbps](b) FR=2 OMANOMA, ε=0.05NOMA, ε=0.12468101214161820020406080100User Random Distribution CountSum User Rate [Mbps] FOV=32°, N0=10−21A2/HzFOV=40°, N0=10−21A2/HzFOV=32°, N0=10−20A2/Hz05101520253044.555.566.5Iteration TimeUser Rate [Mbp/s] min−function approximation ε=0min−function real ε=0min−function approximation ε=0.1min−function real ε=0.1