Linear Precoding for Multiuser Visible-Light

Communication With Field-of-View Diversity

Volume 8, Number 2, April 2016

Renhai Feng
Mingjun Dai
Hui Wang
Bin Chen
Xiaohui Lin

DOI: 10.1109/JPHOT.2016.2542985
1943-0655 Ó 2016 IEEE
IEEE Photonics Journal Linear Precoding for Multiuser VLC

Linear Precoding for Multiuser

Visible-Light Communication
With Field-of-View Diversity
Renhai Feng, Mingjun Dai, Hui Wang, Bin Chen, and Xiaohui Lin

Shenzhen Key Laboratory of Media Security and Shenzhen Key Lab of Advanced Communication
and Information Processing, College of Information Engineering, Shenzhen University,
Shenzhen 518060, China
The State Key Laboratory of Integrated Services Networks, Xidian University, Xi’an 710071, China

DOI: 10.1109/JPHOT.2016.2542985
1943-0655 Ó 2016 IEEE. Translations and content mining are permitted for academic research only.
Personal use is also permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Manuscript received December 31, 2015; revised March 10, 2016; accepted March 14, 2016. Date
of publication March 16, 2016; date of current version March 30, 2016. This work was supported in
part by the Postdoctoral Science Foundation of China under Grant 2015M572358; by the National
Natural Science Foundation of China under Grant 61301182, Grant 61171071, and Grant 61575126;
by the Specialized Research Fund for the Doctoral Program of Higher Education from the Ministry of
Education of China under Grant 20134408120004; by the Natural Science Foundation of Guangdong
Province under Grant S2013040016857 and Grant 2015A030313552; by Yumiao Engineering from
the Education Department of Guangdong Province under Grant 2013LYM_0077; by the Key Project
of Department of Education of Guangdong Province under Grant 2015KTSCX121; by the Open
Fund from The State Key Laboratory of Integrated Services Networks Xidian University under Grant
ISN15-06; by the Foundation of Shenzhen City under Grant KQCX20140509172609163, Grant
GJHS20120621143440025, Grant JCYJ20140418095735590, Grant JCYJ201503-24140036847, and
Grant ZDSY20120612094614154; by Natural Science Foundation of SZU under Grant 00002501
and Grant 00036107; by the Research Foundation of the Higher Education Institute of Guangdong
Province under Grant GDJ2014083; and by the Teaching Reform and Research Project of Shenzhen
University under Grant JG2015038. Corresponding author: M. Dai (e-mail: mjdai@szu.edu.cn).

Abstract: The multiple-input–multiple-output (MIMO) indoor visible-light communication

(VLC) system with multiple users, where several light-emitting diodes (LEDs) send mes-
sages to several photodiodes (PD), is studied. To enhance the diversity, each user is
assumed to be equipped with multiple PDs with a different field of view. The objective
is to minimize the LEDs’ maximum power consumption subject to multiple constraints, in-
cluding illumination constraint and capacity constraint. Under the assumption of maximum-
ratio combining at the receivers, a linear precoding method is proposed. Detailed
operation of this precoding method includes two steps: First, with matrix transformation
and semi-definite relaxation, the original nonconvex optimization problem is transformed
into solvable convex optimization subproblems. Afterward, based on the solution of the
subproblems, it is proved that a linear precoding solution can be reformulated using a
rank-1 decomposition theorem. Numerical results show that our proposed scheme with
different system parameters outperforms the traditional method.

Index Terms: Visible-light communication (VLC), multiple-input—multiple-output (MIMO),

precoding.

1. Introduction
Visible light communication (VLC) is gaining attention from both academia and industry as a
promising complement to the wireless heterogeneous network. Using emerging illumination
devices, e.g., light-emitting diodes (LEDs), as transmitters, VLC systems rapidly evolve as a viable

Vol. 8, No. 2, April 2016 7902708

IEEE Photonics Journal Linear Precoding for Multiuser VLC

and attractive solution for future communication system [1]. Besides, compared to existing radio-
frequency communication (RFC) systems [2], VLC has the following advantages: It uses the
visible light spectrum which is unregulated and license-free. It does not interfere with existing
RFC systems. In VLC system, the information waveforms are modulated directly as intensities
at the transmitter, and the signal is captured by either photo diodes (PD) or imaging sensors at
the receiver. Such process is termed as Intensity Modulated/Direct Detection (IM/DD). One of
the most significant differences between VLC and RFC systems is that the transmitted signal of
VLC has to be positive and real due to IM/DD. Besides, as for channel model, the diffuse link in
VLC is much weaker than the line-of-sight (LoS) link. In [3] and [4], the authors point out that no
multipath interference would occur at data rates below several hundred Mb/s according to a root
mean square delay spread investigation. For multiple-input-multiple-output (MIMO) VLC system,
the channel correlation is a major concern. As [5] shows, ill-conditioned channel matrix may
occur due to homogenous behaviors in different channel gains. As a result, channel precoding
becomes an important problem in VLC systems.
There are many of works along this direction that laid theoretical foundation for precoding in
VLC system [6]–[9]. In [6], the authors investigate the performance of zero forcing precoding in
the indoor VLC system. In [7], the authors investigate the erbium-doped fiber amplifier (EDFA)
effect on bit error rate (BER) performance in space optical communication. In [8], the authors
propose an approach to improve beam quality on diffractive optical elements, but it is mainly
used in inter-satellite optical communication system. In [9], the authors investigate the LED
allocation for multiuser VLC network while the MIMO case is not considered.
For our knowledge, currently all the work in precoding method is based on the minimum
mean square errors (MMSE) criteria [5], [10], [11]. In [10], the authors consider the robust pre-
coding design of the VLC system, but they fixed the receiving gain of all the detectors in a single
receiver. In [11], the authors investigate the Multi-color VLC system and propose a joint
beamformer and offset design, but they did not consider the capacity constraints of the system.
However, as [17] mentioned, illumination constraints have an effect on capacity. Therefore,
MMSE and capacity do not have equivalent effect to the VLC system. As far as we know, previ-
ous studies of precoding method based on MMSE did not consider such effect. In [5], the authors
jointly optimized the equalizer and the precoder of the VLC single user system, but they did not
consider the field of view (FoV) diversity of the receiver.
In this paper, we intend to tackle the problems existed in [5], [10] and [11], namely, we jointly
consider receivers with maximum-ratio combining (MRC) [2], capacity constraints, and FoV di-
versity, which [5], [10], [11] did not mention. We formulate a problem of minimizing the maximum
power consumption subject to multiple constraints including the illumination constraint and
the capacity constraint. We propose a precoding method based on the upper bound of achiev-
able data rate instead of MMSE criteria. The multiuser MIMO case is also considered in our
work. To make our research more realistic, a new method based on the FoV is used to achieve
diversity [13] in our system. The paper is organized as follows. The system model is presented
in Section 2. Then, we give the precoding algorithm in Section 3. Section 4 presents simulation
results. Section 5 concludes the whole work.

Notations
T denotes the transpose operation. trðÞ and rankðÞ denote the trace and rank of a matrix, re-
spectively. diagðÞ denotes extract diagonal elements of the matrix to form a vector. j  j denotes
taking absolute value itemized. A  B represents that A is larger than B element-wise.

2. System Model
See Fig. 1. We consider a downlink broadcasting scenario [12], where a room is illuminated by
M down-facing LED fixtures which are attached to the ceiling. Each fixture consists of multiple
LEDs which are modulated by the same current signal. At the receiving end, there are K re-
ceivers, and each receiver is equipped with N photodiodes (PDs). We assume K ¼ 2 for

Vol. 8, No. 2, April 2016 7902708

IEEE Photonics Journal Linear Precoding for Multiuser VLC

Fig. 1. System model for MIMO VLC system with four LED fixtures and two receivers.

simplicity. Since we are precoding on the fixtures, all the LEDs in one fixture are regarded as a
single transmitter to make the result extendable.
The signals transmitted by the LED fixtures can be given as follows:
xðtÞ ¼ v1 s1 ðt Þ þ v2 s2 ðt Þ þ b (1)

where sn ðt Þ is the desired data signal of receiver n, vn 2 RM1 is the precoding vector corre-
sponding to sn ðt Þ 2 R. Since we are using direct current (DC) biased pulse-amplitude modula-
tion (PAM), the transmit element is driven by a fixed bias current IDC 2 Rþ . Assume for fixture -i,
bi ¼ IDCi . Consequently, bias b 2 RM1þ can also be used to guarantee the LEDs work within
their linear current-light conversion region. Moreover, bias b also plays the role of adjusting illu-
mination level of the room. In order to maintain linear current-light conversion and avoid clipping
distortion, the total transmitted signal xðt Þ 2 RM1
þ must be constrained within some range IDC þ

IDC element-wise, where
2 ½0; 1 is termed as the modulation index. Thus, xðt Þ must satisfy
the amplitude constraint Pðjxðt Þj > AÞ ¼ 0 for 8 t , where A ¼
IDC .
The light received by each PD in each fixture would generally be made up of two components:
the line-of-sight (LoS) component and the diffuse component. Previous studies have shown that
the LOS component is usually much stronger than the diffuse component [14]. Inline with these
researches [15], we only consider the LoS component. We also assume the FoV is clearly de-
fined. The signals within the FoV are all received while those outside the FoV are not.
With these assumptions, the channel gain between the q-th LED and the p-th PD is given by
(
ðmþ1ÞA
cosm ðÞcosð’Þ; 0  ’ 
1
hp;q ¼ 2d 2 2 (2)
0; ’ 
12

where A is the effective area of the PDs, order m is related to the semi-angle of the Lambertian
LED 1=2 , m ¼ ln 2=lnðcos1=2 Þ,  is the angle of emission, ’ is the angle of incidence, d denotes
the distance between the LED and the PD, and
1=2 denotes the FoV of the PD according to [15].
We assume each LED fixtures can be modulated by a combination of s1 ðt Þ and s2 ðtÞ. The weight of
the mixture are determined by v1 and v2 , respectively. We thus have the constraint vn 2 RM1 , namely,
each element of vn is a real number. This constraint guarantees that the weight is a real number.
Assume the transmit signal sn ðt Þ N ð0; 1Þ. Since we are using current signal modulation in
LED, the modulated current signal xðt Þ cannot be negative. As a result, the amplitude constraint
of LED fixtures can be expressed as jv1 j þ jv2 j

b. Moreover, since the bias current must be
positive, we have b  0.

Vol. 8, No. 2, April 2016 7902708

IEEE Photonics Journal Linear Precoding for Multiuser VLC

In order to maintain linear current-light conversion and to avoid clipping distortion, the trans-
mission power should also have an upper bound p
jv1 j þ jv2 j þ
b, where p 2 RM1 þ .
In a MIMO VLC system, signals are transmitted by multiple LED fixtures and received by multiple
PDs. The MIMO channel gain matrix H can be determined by the DC gains between each LED-PD pair
[14]. Its element hmn corresponds to the channel gain between the n-th LED fixture and the m-th PD.
Due to homogenous behaviors in different channel DC gains [14], the MIMO channel gain matrix is ill-
conditioned. However, by using the method mentioned in [13], we can ensure H to be well-conditioned.
The signals received by PD n is given by
yn ðt Þ ¼ Hn xðt Þ þ zn ðt Þ (3)

where zn ðt Þ N ð0; 2n Þ is the additive noise, Hn 2 RKNM

þ is the channel gain matrix. Note that
the bias b should be known by all the receivers.
At the receivers, we use MRC [16] to combine the signals. We thus have

yn ðt Þ ¼ ðHn vn ÞT Hn xðt Þ þ ðHn vn ÞT zn ðt Þ: (4)

Since we are discussing the precoding method in a slow fading channel, regardless of the time
slot, we drop the time index ðt Þ for simplicity. The signal-to-interference-plus-noise-ratio (SINR)
of the receivers can be calculated as

vT1 HT1 H1 v1
SINR1 ðv1 ; v2 Þ ¼ (5)
vT2 HT1 H1 v2 þ zT1 z1
vT2 HT2 H2 v2
SINR2 ðv1 ; v2 Þ ¼ : (6)
vT1 HT2 H2 v1 þ zT2 z2

Since zn ðt Þ N ð0; 2n Þ is additive white Gaussian noise (AWGN), we have zTn zn ¼ 2n .

3. Precoding Method In MIMO VLC System Based on FoV Diversity

Assuming MRC in the PDs, a capacity upper bound [17] for each user in our system can be ex-
pressed as follows:
1
Cn ðv1 ; v2 Þ ¼ logð1 þ Gð
ÞSINRn ðv1 ; v2 ÞÞ (7)
2

Gð
Þ ¼ 4

ð1 
Þ 0 G
G 12 14 G
G 1: (8)

Note that Gð
Þ is quasi-convex with respect to
.

In this paper, we aimed at minimizing LED’s maximum power consumption subject to several
constraints, including the illumination constraint and the capacity constraint. The precoding
problem can be formulated as follows:

minv1 ;v2 ;
;b vT1 v1 þ vT2 v2 þ bT b (9)

s.t. C1 ðv1 ; v2 Þ  C1 (10)
C2 ðv1 ; v2 Þ  C2 (11)
jv1 j þ jv2 j

b (12)
jv1 j þ jv2 j þ
b
p (13)
v1 ; v2 2 RM1 b 2 RM1
þ
2 ½0; 1:

There are three difficulties in solving problem (9) as listed below.

1) Gð
Þ is not differentiable when
¼ 1=2.
2) Note that problem (9) is a quadratically constrained quadratic programming (QCQP) prob-
lem. Since the quadratic constraints (10) and (11) are non-convex, existing convex optimi-
zation method cannot be directly used in this problem.

Vol. 8, No. 2, April 2016 7902708

IEEE Photonics Journal Linear Precoding for Multiuser VLC

3) Constraints (12) and (13) are not differentiable when v1i ¼ 0 for 8 i 2 f1; . . . ; Mg or v2i ¼ 0
for 8 i 2 f1; . . . ; Mg.
In order to solve difficulty 1, we use binary search for
to get an optimal value with limited
computation budget.
In order to solve difficulties 2 and 3, we let V1 ¼ v1 vT1 , V2 ¼ v2 vT2 , B ¼ bbT , and P ¼ ppT .
Using some transformation technique, we can reformulate problem (9) as

minV1 ;V2 ;
;B trðV1 Þ þ trðV2 Þ þ trðBÞ (14)

s.t: trðV1 HT1 H1 Þ  D1 trðV2 HT1 H1 Þ  D1 21 (15)
trðV2 HT2 H2 Þ  D2 trðV1 HT2 H2 Þ  D2 22 (16)
2diagðV1 þ V2 Þ

diagðBÞ2
(17)
2diagðV1 þ V2 Þ þ 2
diagðBÞ
diagðPÞ
2
(18)
D2 ¼ ðe 2C2
 1Þ=Gð
Þ
D1 ¼ ðe 2C1
 1Þ=Gð
Þ
rankðV1 Þ ¼ 1 (19)
rankðV2 Þ ¼ 1 (20)
V1 ; V2 2 RMM
2 Rþ :

Note that in (15) and (16), we use trðABÞ ¼ trðBAÞ, and in (17) and (18), we use a transforma-
tion of the Cauchy-Schwarz inequality ðjv1 j þ jv2 jÞ2
2jv1 j2 þ 2jv2 j2 .
Using the transformation above, problem (9) can be turned into a Semi-Definite Programming
(SDP) problem (14) with rank constraint. Meanwhile, the variable substitution above may also
bring the following side effect:
4) We cannot guarantee b 2 RM1þ through any explicit constraint imposed on B.
5) Rank constraints (19) and (20) are also non-convex.
By dropping the non-convex rank constraints (19) and (20), problem (14) becomes a standard
SDP. It can be solved efficiently using the method proposed in [18].

Proposition 1
b in problem (9) and the optimization result B in problem (14) have the following relationship
pffiffiffiffiffiffi
bi ¼ Bii 8i 2 f1; . . . ; Mg.

Proof
Note that in optimization problem (14), we only have two functions of B, which is trðBÞ and
diagðBÞ. It means that only the diagonal part of B is useful for the optimization problem. If Bopt
is the optimal value of problem (14), we have trðBopt Þ ¼ trðdiagðBopt ÞÞ diagðBopt Þ ¼
diagðdiagðBopt ÞÞ, and hence diagðBopt Þ is also the optimal value of problem (14). Then we con-
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
struct matrix B0opt with B0optij ¼ Boptii Boptjj for 8 i; j 2 f1; . . . ; Mg, where B0opt is still the optimal
pffiffiffiffiffiffi
value of problem (14). We thus have B0opt ¼ bbT , where bi ¼ Bii for 8 i 2 f1; . . . ; Mg. h

Lemma 1
In constraint (27), there is only one m 2 f1; . . . ; Mg which satisfies
 
diag V1 m ¼ 4m : (21)

Proof
We prove by contradiction. Assuming there exist m1 ; m2 2 f1; . . . ; Mg that both satisfy (21),
we can find V01 which satisfies V01m2 m2 ¼ V1m2 m2  t and t 2 Rþ which meets constraints (25) and

Vol. 8, No. 2, April 2016 7902708

IEEE Photonics Journal Linear Precoding for Multiuser VLC

(26). Since H2 and H1 may not be square, and due to the ill condition nature of the channel ma-
trix in VLC [13], HT2 H2 and HT1 H1 is not full rank. Therefore, we can find a matrix
2 3
0    t1;m2    0
6 .. .. .. 7
6 . . .7
6 7
T¼6 t
6 m2 ;1    t m2 ;m2    0 7
7 (22)
6 . . . 7
4 .. .. .. 5
0  0  0
where

trðTHT2 H2 Þ ¼ 0; trðTHT1 H1 Þ ¼ 0: (23)

We let V01 ¼ V1  T. According to (23), V01 still satisfies constraints (25), (26). Moreover, only
the m1 -th element satisfies (21) in V01 . Since t 2 Rþ , we have trðV01 ÞGtrðV1 Þ ¼ 1 . This result contra-
dicts to the assumption that V01 is the optimal solution of problem (14). This completes the proof.

Proposition 2
Suppose Vk is an optimal solution to SDP problem (14), then there must be a rank-1 matrix
V0kthat achieves the same optimality as Vk .

Proof
We specify k ¼ 1 for simplicity. Cases k ¼ n can be proved similarly. Suppose V1 is an opti-
mal solution to SDP problem (14). We have the optimal values
 
tr V1 ¼1 (24)
  T 
tr V1 H2 H2 ¼2 (25)
  T 
tr V1 H1 H1 ¼3 (26)
 
diag V1
4 : (27)

Note that problem (14) does not directly involve V1 . If we have another rank-1 solution V01 which
achieves the same profile in (24)–(27), then V01 can achieve the same optimality as V1 .
Using Lemma 1, we can transform (27) into (21). We then define matrix

Z¼ 4
Zm;m ¼ 1Zi;j ¼ 0ði; jÞ 6¼ ðm; mÞ \ i; j 2 f1; . . . ; Mg:

Using Z, we can turn (21) into a trace form

 
tr V1 Z ¼ 4m : (28)

h
From Proposition 2, we know that if rankðV1 Þ ¼ 1, then the solution to problem (9) subject to
active constraints is given by V1 ¼ v1 v1 T . If rankðV1 Þ  2, using rank-one decomposition
theorem [19], we can find a rank-1 matrix v^1 v^1 T in polynomial-time. In [19], it also shows that
v^1 v^1 T can also satisfy the constraints form trfAi v^1 v^1 T g ¼ trfAi V1 g, i ¼ 1; 2; 3; 4, which represent
constraints (24)–(26) and (28) in our problem. Then, the solution to problem (9) subject to active
constraints can be represented as v1 .

4. Simulation Results
In order to evaluate the proposed scheme, we have numerically simulated a typical indoor VLC
scenario. As shown in Fig. 1, the volume of the room is assumed to be 3  3  3 m3 . The LED

Vol. 8, No. 2, April 2016 7902708

IEEE Photonics Journal Linear Precoding for Multiuser VLC

Fig. 2. Minimum transmit power required for the LED fixture.

Fig. 3. Feasible probability of the beamforming problem.

fixtures are deployed on top of the roof with equal distance to each other. According to [15], we as-
sume there are two receivers, and each receiver is equipped with 4 PDs with the FoV profile

1=2 ¼ 7:5 , 15 , 30 , 60 . In order to use a Monte Carlo Method for simulation, we assume the re-
ceivers move in the room randomly. Five thousand random channel matrices are generated based
on (2). The parameters in (2) are set as follows: 1=2 ¼ 15 , A ¼ 1 cm2 . Assuming the maximum il-
lumination power of LED fixture is 2.5W, our proposed algorithm is compared with the widely used
zero forcing method, which is a simplification of [5] and [6]. Our algorithm iterated for 15 times
which can guarantee a 215 accuracy for
. We plot two figures, as shown in Figs. 2 and 3, where
the former evaluates the minimum transmit power and the latter evaluates feasible probability.
In Fig. 2, the information content of both receivers varies from 0 to 1.8 (nat). The simulation
results show that our proposed algorithm always needs less power than zero forcing method. It
means that our proposed algorithm is more energy efficient than zero forcing.

Vol. 8, No. 2, April 2016 7902708

IEEE Photonics Journal Linear Precoding for Multiuser VLC

Due to the propagation nature of light, beamforming method cannot always find a feasible
solution [18]. Since the precoding problem is not always feasible, we study the probability of
finding a feasible solution while the receivers are moving in the room randomly during the
5000 simulations. We name this probability as feasible probability. In Fig. 3, it shows that with
the information content of both receivers varying, our proposed algorithm can always achieve
a higher feasible probability than the zero forcing method, which means that in some cases,
our proposed algorithm can find the feasible beamforming solution while the zero forcing
method cannot.

5. Conclusion
In this paper, we have proposed an energy efficient beamforming scheme for the indoor MIMO
VLC system which minimizes the transmit power under capacity constraint and illumination con-
straint. Simulation results have shown that our proposed scheme achieves an acceptable per-
formance under a small number of iterations. Moreover, the simulation shows that our proposed
scheme outperforms zero forcing method both in transmit power consumption and feasible
probability.

Acknowledgment
The authors would like to thank the anonymous reviewers for their valuable suggestions.

References
[1] A. Shlomi et al., Advanced Optical Wireless Communication Systems, Wiley Encyclopedia of Electrical and
Electronics Engineering. Hoboken, NJ, USA: Wiley, 2012.
[2] M. Dai et al., “Incorporating D2D to current cellular communication system,” Mobile Inf. Syst., vol. 2016, Mar. 2016,
Art. no. 2732917.
[3] A. Burton, Z. Ghassemlooy, E. Bentley, and C. Botella, “Experimental demonstration of 50-Mb/s visible light
communications using 4  4 MIMO,” IEEE Photon. Technol. Lett., vol. 26, no. 9, pp. 945–948, May 2014.
[4] N. A. Mohammed and M. A. Elkarim, “Exploring the effect of diffuse reflection on indoor localization systems based
on RSSI-VLC,” Opt. Exp., vol. 23, no. 16, pp. 20 297–20 313, Aug. 2015.
[5] Y. Kai, Q. Hua, R. J. Baxley, and S. Yao, “Joint optimization of precoder and equalizer in MIMO VLC systems,”
IEEE J. Sel. Areas Commun., vol. 33, no. 9, pp. 1949–1958, Sep. 2015.
[6] Y. Hong, J. Chen, Z. Wang, and C. Yu, “Performance of a precoding MIMO system for decentralized multiuser
indoor visible light communications,” IEEE Photon. J., vol. 5, no. 4, Aug. 2013, Art. no. 7800211.
[7] M. Li et al., “Investigation of the EDFA effect on the BER performance in space uplink optical communication under
the atmospheric turbulence,” Opt. Exp., vol. 22, no. 21, pp. 25 354–25 361, Oct. 2014.
[8] L. Tan et al., “Approach to improve beam quality of inter-satellite optical communication system based on diffractive
optical elements,” Opt. Exp., vol. 17, no. 8, pp. 6311–6319, Apr. 2009.
[9] A. Sewaiwar, S. V. Tiwari, and Y. H. Chung, “Smart LED allocation scheme for efficient multiuser visible light
communication networks,” Opt. Exp., vol. 23, no. 10, pp. 13 015–13 024, May 2015.
[10] H. Ma, L. Lampe, and S. Hranilovic, “Robust MMSE linear precoding for visible light communication broadcasting
systems,” in Proc. IEEE GC Wkshps, 2013, pp. 1081–1086.
[11] Q. Gao, R. Wang, Z. Xu, and Y. Hua, “Joint beamformer and offset design for multi-color visible light
communications,” in Proc. 6th Int. Conf. WCSP, 2014, pp. 1–6.
[12] M. Dai, K. W. Shum, and C. W. Sung, “Data dissemination with side information and feedback,” IEEE Trans.
Wireless. Commun., vol. 13, no. 9, pp. 4708–4720, Sep. 2014.
[13] C. He, T. Q. Wang, and J. Armstrong, “MIMO optical wireless receiver using photodetectors with different fields of
view,” in Proc. IEEE VTC Spring, 2013, pp. 1–5.
[14] L. Zeng et al., “High data rate multiple input multiple output (MIMO) optical wireless communications using white
LED lighting,” IEEE J. Sel. Areas Commun., vol. 27, no. 9, pp. 1654–1662, Dec. 2009.
[15] T. Fath and H. Haas, “Performance comparison of MIMO techniques for optical wireless communications in indoor
environments,” IEEE Trans. Commun., vol. 61, no. 2, pp. 733–742, Feb. 2013.
[16] K. Seung Ahn, “Performance analysis of MIMO-MRC system in the presence of multiple interferers and noise over
Rayleigh fading channels,” IEEE Trans. Wireless Commun., vol. 8, no. 7, pp. 3727–3735, Jul. 2009.
[17] A. Lapidoth, S. M. Moser, and M. A. Wigger, “On the capacity of free-space optical intensity channels,” IEEE Trans.
Inf. Theory, vol. 55, no. 10, pp. 4449–4461, Oct. 2009.
[18] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.: Cambridge Univ. Press, 2004.
[19] A. Wenbao, Y. Huang, and S. Zhang, “New results on Hermitian matrix rank-one decomposition,” Math. Programm.,
vol. 128, no. 1/2, pp. 253–283, Jun. 2011.

Vol. 8, No. 2, April 2016 7902708