Вы находитесь на странице: 1из 14

138 J. OPT. COMMUN. NETW./VOL. 10, NO. 3/MARCH 2018 Hassan et al.

Delay-QoS-Aware Adaptive Modulation


and Power Allocation for Dual-Channel
Coherent OWC
Md. Zoheb Hassan, Md. Jahangir Hossain, Julian Cheng, and Victor C. M. Leung

Abstract—Statistical-delay quality of service (QoS) to design fading mitigation solutions by taking into account
provides bounded link-layer delay over wireless fading
the delay-bound requirements of the transmitted traffics.
channels with a certain delay-bound violation probability.
We propose statistical-delay-QoS-aware adaptive modula- Because atmospheric turbulence fading is a quasi-static
tion (AM) and power allocation for a dual-channel coherent channel, reliable estimates of the channel state informa-
optical wireless communication system over the atmos-
pheric turbulence fading channels. For given statistical- tion (CSI) can be obtained at the transmitter through
delay constraints and target bit-error-rate requirements, feedback from the receiver. Based on such CSI, the trans-
our proposed AM and power allocation maximize the mitter can adapt the transmission parameters over the tur-
effective spectral efficiency subject to the transmit-power bulence fading channels. Moreover, due to large channel
constraints. We develop delay-QoS-aware adaptive trans-
coherence time of atmospheric turbulence fading, an
mission schemes by employing independent and joint
channel optimizations subject to average transmit-power OWC system does not require the frequent adaptation of
constraints. We also consider independent, joint, and transmission parameters. Consequently, adaptive trans-
successive channel optimizations for developing delay- mission is a feasible fading mitigation solution for OWC
QoS-aware adaptive transmission schemes subject to systems. Constant-power variable-rate (CPVR) adaptive
peak transmit-power constraints. Numerical results dem-
onstrate that our proposed AM and power allocation transmission schemes were investigated by adapting the
significantly outperform the conventional adaptive trans- modulation orders over the lognormal and gamma–gamma
mission schemes in the strict statistical-delay constraints. turbulence fading channels [2,3]. Due to the stochastic
Numerical results also depict superiority of the joint nature of turbulence fading and link attenuation, received
channel optimization in the strong turbulence fading and
optical power also fluctuates. Consequently, for an average
strict statistical-delay constraints.
transmit-power-constrained OWC, transmit-power adapta-
Index Terms—Adaptive modulation; Effective spectral tion (along with rate adaptation) offers another degree of
efficiency; Power allocation. freedom for adaptive transmission over atmospheric turbu-
lence fading channels. Considering continuous rate adap-
tation and the average transmit-power constraint,
I. INTRODUCTION different transmit-power adaptation schemes were pro-
posed for an OWC system [4]. Performance of transmit-
power and rate adaptation algorithms were studied for
A tmospheric-turbulence-induced optical signal fading
is a dominant performance degrading factor for any
outdoor optical wireless communication (OWC) system,
an OWC system subject to dynamic link attenuation via
experiment [5]. Considering practical modulations and
and it severely degrades the error rate and data rate of average/peak transmit-power constraints, joint transmit-
an OWC system. In addition, atmospheric turbulence fad- power and rate adaptation was proposed over the atmos-
ing affects the link-layer delay, also known as the queuing pheric turbulence fading channels [6]. By adapting the
delay, of an OWC system [1]. A deep fade can exist up to transmission rate and power, the aforementioned works
1–100 ms, and such a deep fade can cause link-layer improved spectral efficiency of an OWC system while main-
delay-bound violation for a large number of transmitted taining a desired bit-error rate (BER) at the receiver.
frames. For real-time traffics, such as video conferencing However, the aforementioned transmit-power and rate
and live streaming, the link-layer delay bound is a key adaptation schemes did not consider delay-QoS require-
quality-of-service (QoS) metric. Therefore, it is imperative ments of the transmitted traffics. Consequently, our moti-
vation is to design adaptive transmission for OWC systems
by jointly considering atmospheric turbulence fading and
Manuscript received October 3, 2017; revised December 17, 2017; delay-QoS requirements of the transmitted traffics.
accepted December 17, 2017; published February 13, 2018 (Doc. ID 308354).
Md. Z. Hassan (e-mail: mdzoheb@ece.ubc.ca) and V. C. M. Leung are with Recent works on delay-QoS-aware adaptive transmis-
the Department of Electrical and Computer Engineering, University of sions for OWC systems are reviewed as follows. Adaptive
British Columbia, Vancouver, British Columbia V6T 1Z4, Canada. modulation (AM) was considered to maximize the through-
Md. J. Hossain and J. Cheng are with the School of Engineering,
University of British Columbia, Kelowna, British Columbia V1V 1V7,
put of an OWC system supporting only the latency-
Canada. tolerant traffics [7]. By integrating AM and an automatic
https://doi.org/10.1364/JOCN.10.000138 retransmission request protocol, cross-layer performance

1943-0620/18/030138-14 Journal © 2018 Optical Society of America


Hassan et al. VOL. 10, NO. 3/MARCH 2018/J. OPT. COMMUN. NETW. 139

analysis of an OWC system was performed, and delay QoS ESE similar to the average spectral efficiency (ASE) of the
of an OWC system was investigated in terms of the number conventional variable-power–variable-rate (VPVR) adap-
of required retransmissions [8]. However, the authors in tive transmission scheme, and (2) at a stringent statistical-
Ref. [8] did not provide any guideline for designing the delay constraint, ICO approaches a fixed-rate channel
adaptive transmissions based on the delay-QoS require- inversion scheme. We also develop delay-QoS-aware AM
ments. Queuing-delay-aware adaptive link allocation pro- and power allocation for dual-channel coherent OWC sub-
tocols were investigated for hybrid radio-frequency (RF)/ ject to a peak transmit-power constraint. We consider ICO,
OWC backhaul links [9]. However, the authors in Ref. [9] JCO, and successive channel optimization (SCO) for a peak
considered only two extreme cases, namely delay-tolerant transmit-power-constrained coherent OWC system. In ad-
transmission where traffics can tolerate arbitrary long dition, we investigate the computational complexity of our
delay, and delay-limited transmission where traffics do proposed adaptive transmission schemes. Numerical re-
not tolerate any delay. By using the concept of effective sults demonstrate that our proposed AM and power alloca-
capacity (EC) and the average transmit-power constraint, tion achieve significantly larger ESE compared to the
we proposed transmit-power adaptation schemes to pro- conventional adaptive transmission schemes in the strict
vide statistical-delay-bound guarantee to the transmitted statistical-delay constraints. Numerical results also con-
traffics in a coherent OWC system [10]. However, we con- firm that JCO outperforms both ICO and SCO in strong
sidered a continuous transmission rate adaptation with turbulence fading and strict statistical-delay constraints.
an ideal AM and coding. Consequently, the adaptive trans- However, the performance gap among our proposed chan-
mission schemes proposed in Ref. [10] only provide a theo- nel optimizations decreases as turbulence fading becomes
retical upper bound of the statistical delay-QoS-aware weak and/or the statistical-delay constraint becomes loose.
throughput of a coherent OWC system. The organization of the rest of the paper is as follows.
In this work, we propose statistical delay-QoS-aware The system model is described in Section II. Sections III
AM and power allocation for a dual-channel coherent and IV provide detailed analysis of ICO and JCO subject
OWC system over the gamma–gamma turbulence chan- to an average transmit-power constraint, respectively.
nels. Different practical coherent OWC systems, such as Section V provides delay-QoS-aware AM and power alloca-
coherent polarization multiplexing and 2 × 2 coherent tion subject to a peak transmit-power constraint.
multiple-input–multiple-output OWC, can be modeled as Section VI presents some selected numerical results, and
dual-channel coherent OWC systems [11]. For given stat- finally Section VII provides concluding remarks.
istical-delay constraints and target BER requirements,
we formulate the proposed AM and power allocation as
an optimization problem of maximizing effective spectral
II. SYSTEM MODEL
efficiency (ESE) subject to transmit-power constraints.
Although the sources of energy for OWC systems are not A. Atmospheric Turbulence Fading and
limited, due to laser lifetime (from a device safety point Assumptions
of view) and eye safety standards for outdoor OWC com-
munications, transmission power of an OWC system is The gamma–gamma distribution recently emerged as a
not unbounded [12]. Consequently, average and/or peak useful turbulence model because it has an excellent fit with
transmit-power constraints are employed for outdoor OWC measured data over a wide range of turbulence conditions
communications [6]. Unlike RF communications, the goal from weak to strong turbulence regimes [13]. In this paper,
of our power allocation scheme is not to reduce energy con- we model the optical irradiance I using a normalized
sumption at the transmitter by minimizing the transmit gamma–gamma random variable (RV) with probability
power. Our objective is to efficiently allocate the total avail- density function (PDF):
able output power of a laser transmitter between two op-
tical channels, based on the channel fading gains and 2 αβ αβ
 p
delay-QoS requirements of the transmitted traffics, so that f I I  αβ 2 I 2 −1 K α−β 2 αβI ; (1)
ΓαΓβ
the considered OWC system can achieve an improved ESE
over the atmospheric turbulence channel. We first present where Γ· is the gamma function {[14], Eq. (8.310)} and
independent and joint channel optimizations by consider- K α−β · is the modified Bessel function of the second kind
ing average transmit-power constraints. In an independent of order α − β {[14], Eq. (8.432.9)}. Here, α and β are the
channel optimization (ICO), modulation order and trans- shaping parameters of the gamma–gamma distribution.
mit power of each channel are adapted based on the CSI Under an assumption of plane wave propagation with neg-
of individual channels. In a joint channel optimization ligible inner scale (which corresponds to long propagation
(JCO), modulation order and transmit power of both chan- distance and small detector area), the values of α and β are
nels are proposed to adapt jointly if both channels have in- determined from [15] [Eqs. (2a) and (2b)]. We have the
stantaneous signal-to-noise ratios (SNRs) larger than a following assumptions:
certain threshold. We develop closed-form ESE expres-
sions for both ICO- and JCO-based adaptive transmission • A1: The data traffic has a constant arrival rate at the
schemes subject to average transmit-power constraints. transmitter. The assumption of a constant traffic arrival
Our analysis reveals the following insights: (1) at a loose rate is justified when the rate of channel variation is
statistical-delay constraint, both ICO and JCO achieve much faster compared to the source rate variation. For
140 J. OPT. COMMUN. NETW./VOL. 10, NO. 3/MARCH 2018 Hassan et al.

example, the channel coherence time of OWC communi- modulated parallel optical carriers. Finally, adaptive power
cations is on the order of milliseconds, and the typical control is applied to adjust the transmit power of the modu-
time scale of a video source rate adaptation is on the lated optical carriers. At the receiver, coherent detection
order of seconds [16]. Hence, the traffic arrival rate and demodulation are applied on each optical carrier.
remains constant over a large number of independent Following this stage, the detected information bits are
channel fading realizations. stored in a receive buffer and data sink. The modulation
• A2: The accumulated phase noises due to the atmos- order and the transmit power for each optical carrier are
pheric turbulence and/or lasers can be tracked and cor- adapted at each TS based on the statistical-delay-QoS re-
rected almost perfectly following the photodetection. Such quirements of the transmitted traffics, target BER require-
an assumption ensures mathematical tractability of the ments, and CSI obtained from the receiver. In our system
ensuing AM and power allocation analysis. In particular, model, we consider that the duration of a typical TS is same
carrier phase error (CPE) is present during the detection as the channel coherence time (on the order of millisec-
of transmitted symbols with imperfect phase noise com- onds). Consequently, the channel gains remain constant
pensation. CPE results in mathematically involved BER over a given TS, and independently vary from one TS to
expressions for higher-order two-dimensional coherent another TS. We denote the total average transmit power of
modulations [17]. Due to such involved BER expressions, the considered system as Ps , and without loss of generality,
it is challenging to develop statistical delay-QoS-aware we assume that the average transmit power is equally
joint AM and power allocation with imperfect phase noise divided between both channels. Following [18], the instan-
compensation at the receiver. We also emphasize that the taneous SNR (per symbol) of the ith optical channel at the
assumption of perfect phase noise compensation can be tth TS is written as
satisfied in practical coherent OWC systems [18].
• A3: The parallel optical channels experience ergodic and γ i ≜ γ i t  γ c Ii t; ∀i ∈ f1; 2g; (2)
independent and identically distributed (i.i.d.) fading. In where γ c is the average SNR defined as γ c  RΔ 2qB Ps . Here, R
L

addition, the complete knowledge of the channel statis- is the photodetector sensitivity, q is the electronic charge, B
tics and CSI is available at the transmitter so that the is the bandwidth of the optical filter at the receiver, ΔL is
transmitter can adapt the transmission parameters. the path loss coefficient, and Ii t is the atmospheric turbu-
lence fading coefficient of the ith channel at the tth TS. In
B. System Overview an OWC system, the duration of a typical frame in the
data link layer is on the order of microseconds [8].
Consequently, a given TS contains a large number of
Figure 1 shows the block diagram of our system model. frames, and similar modulation order and transmit power
From a given data source, the traffic arrives in the trans- are selected for all the frames. Therefore, all the frames in a
mitter at a constant arrival rate. Such traffics are grouped given TS are transmitted using the same transmission
into multiple frames at the data link layer and stored at a rate; however, transmission rate independently varies
transmit buffer. Each frame is decomposed into multiple for different TSs. Based on [19] [Eq. (14)], we consider the
parallel bit streams that are transmitted over the parallel following conditional BER expression for M-ary modula-
optical channels. At a given transmission slot (TS), AMs tion over the ith optical channel, ∀i ∈ f1; 2g:
are applied to the parallel bit streams. The modulated
 
symbols drive an optical modulator to yield coherent c γ μ θ; γ
BERt ≈ c1 exp − 2c3 i i : (3)
M i θ; γ − 1

In Eq. (3), θ is the QoS exponent; BERt is the target BER;


μi θ; γ is the statistical delay-QoS-aware transmit-power
adaptation factor (ratio of the allocated power to average
transmit power) for the ith optical channel; M i θ; γ is
the selected modulation order based on instantaneous
SNR and statistical-delay-QoS requirements for the ith
optical channel; and c1 , c2 , and c3 are the modulation-
dependent parameters. In particular, we have c1 ; c2 ; c3  
0.2; 1.6; 1 {[19], Eq. (9)}, c1 ; c2 ; c3   0.05; 6; 1.9 {[19],
Eq. (11)}, c1 ; c2 ; c3   0.2; 1.85; 2.19 {[4], Eq. (7)} for
M-ary quadrature amplitude modulation (M-QAM),
M-ary phase shift keying (M-PSK), and M-ary pulse ampli-
tude modulation (M-PAM), respectively. Equation (3) pro-
vides approximate BER expression; however, such a BER
expression is invertible and is useful for designing delay-
QoS-aware AM and power allocation. Moreover, such a
BER expression is highly accurate for BER smaller than
10−3 . Therefore, we consider BERt < 10−3 in our system
Fig. 1. System block diagram. model. From Eq. (3), M ci 3 θ; γ  1  Kμi θ; γγ i where the
Hassan et al. VOL. 10, NO. 3/MARCH 2018/J. OPT. COMMUN. NETW. 141

c2
factor K is given by K  − logBER t ∕c  . At the tth TS, the link-layer delay-bound violation probability is approxi-
1
the transmission rate of a frame (in bits/frame unit) of mated as {[26], Eq. (6)}
the ith OWC channel by considering continuous rate adap- PrD ≥ Dmax  ≈ exp−θζDmax : (6)
tation and BER constraint is obtained as {[20], Eq. (16)}
To quantify the relationship between the transmission
Tf B rate and the statistical-delay-QoS constraint, EC was pro-
Ri t  log2 1  Kμi θ; γγ i ; ∀i ∈ f1; 2g; (4)
c3 posed in the literature. EC is a link-layer capacity, and it is
where T f is the frame duration. Due to discrete rate defined as the maximum constant traffic arrival rate that
adaptation in practice, Eq. (4) provides an upper bound can be supported by the transmission rate subject to a cer-
of the practical transmission rate. We will use Eq. (4) to tain statistical-delay constraint specified by θ. By using [27]
develop JCO-based AM and power allocation in Section IV. [Eq. (12)], the mathematical expression of EC for the ith
(∀i ∈ f1; 2g) channel is obtained as1
" !#!
1 XT
C. Statistical-Delay QoS and EC i
Ec θ  − lim log E exp −θ Ri t ; (7)
T→∞ θT
t1
Statistical-delay QoS is employed in recent literature for where E· is the mathematical expectation operator.
QoS-aware resource allocations over wireless networks Because Ri t is an ergodic and i.i.d. random process, from
(see Ref. [21] and references therein). For delay-sensitive [20] [Eq. (4)], we obtain
traffics, delay violation and buffer overflow at the link layer
are critical QoS parameters. Due to the time-varying 1
Ei
c θ  − log Eexp−θRi ; ∀ i ∈ f1; 2g: (8)
nature of the wireless fading channels, maintaining θ
deterministic delay and buffer overflow at the link layer
is challenging for wireless fading channels. An alternative When the queue is in steady state, the average arrival rate
strategy is to maintain delay-QoS requirements sta- is equal to the average departure rate [28]. Accordingly, EC in
tistically over the fading channels. In a statistical- Eq. (8) can also be considered as the maximum achievable
delay-QoS provisioning scheme, the link-layer delay is throughput of a communication system subject to the link-
maintained less than a certain delay-bound subject to a layer delay-QoS constraint. From Eq. (6), for a target delay-
small delay-bound violation probability. Real-time traffics bound violation probability, PrD ≥ Dmax  ≤ ϵ, the required
statistically tolerate certain delay-bound violation at the QoS exponent should satisfy θ ≥ θo ≜ ln1∕ϵ
ζDmax . Equation (8) is
link layer. For example, it is recommended that the data a monotonically decreasing function of θ. Therefore, consider-
traffics for real-time voice over IP in the long-term evolu- ing the statistical-delay constraint at the link layer, the maxi-
tion advanced standard should experience maximum 2% mum achievable throughput for the ith channel in the
probability of having link-layer delay more than 50 ms considered system is obtained as Ei − lnϵ
c  ζDmax , ∀ i ∈ f1; 2g.
[22]. Similarly, delay-sensitive applications for machine-
type communications and device-to-device communications
in 5G networks also have delay-bound requirements, and III. ICO WITH AVERAGE TRANSMIT-POWER CONSTRAINT
such delay-bound requirements can be statistically satis-
fied [23,24]. Therefore, the analysis of statistical-delay
QoS is important for practical wireless communications. In ICO for a dual-channel coherent OWC, transmit
parameters of each channel are independently adapted sub-
Statistical QoS is maintained by satisfying a target ject to the average transmit power per channel constraint.
buffer-overflow probability at the link layer. We denote In this section, we develop AM and power allocation for a
Q as a steady-state buffer length, and Qth as the threshold dual-channel coherent OWC system by employing an ICO
value of buffer length. For an asymptotically large value of approach. We assume that for each optical channel, a total
Qth , the buffer-overflow probability is given by [25] N constellations are available, and we denote M i θ; γ as the
PrQ ≥ Qth  ≈ exp−θQth ; (5) assigned constellation for the ith channel. We divide the
range of received SNR for each channel into N nonoverlap-
where θ > 0 provides the decaying rate of the tail probabil- ping regions, and we associate each region with a unique
ity of buffer length. The values of θ imply the nature of QoS constellation. For a given target BER, the constellation
requirements of the data traffics. A large value of θ implies selection and transmit-power adaptation rule is given as2
strict statistical-QoS requirements, and a small value of θ
implies loose statistical-QoS requirements. Statistical-
1
delay QoS at the link layer is modeled by two parameters, Due to space limitation, a detailed discussion for the derivation of the
mathematical model of EC is beyond the scope of this paper. Interested
namely, the maximum delay-bound and the delay-bound
readers are referred to [27] for the detailed derivation of Eq. (7).
violation probability. For stationary and constant traffic 2
In Eq. (9), we consider discrete rate adaptation. In this case, the transmis-
arrival rate, the buffer-overflow probability given by sion rate of a frame over the ith OWC channel is given by Ri  T f Bni, where
Eq. (5) also implies a link-layer delay-bound violation prob- ni ∈ f1; 2; ·; ·; Ng. To satisfy the target BER constraint for discrete rate adap-
ni c3
−1
ability. We denote D as the link-layer delay, Dmax as the tation, we consider μi θ; γ  2 Kγ i
, ∀i ∈ f1; 2g. Note that fμi θ; γg depends
on the selected constellations. However, selection of a constellation is deter-
maximum tolerable link-layer delay, and ζ as the constant mined by jointly considering Eqs. (9) and (12). Consequently, the value of the
traffic arrival rate. From Little’s formula, it can be shown QoS exponent, θ, influences the selection of the constellation and the trans-
that in steady state, Q  ζD and Qth  ζDmax . Hence, mit power adaptation factor.
142 J. OPT. COMMUN. NETW./VOL. 10, NO. 3/MARCH 2018 Hassan et al.

select M i θ; γ  2n and X


N−1
2nc3 − 1
Λα; β; γ i;n  − Λα; β; γ i;n1 
2 −1
nc3 KΓαΓβ
μi θ; γ  if γ i;n ≤ γ i < γ i;n1 ; (9) n1
Kγ i 2Nc3 − 1
n  1; 2; ·; ·; N; ∀ i ∈ f1; 2g:  Λα; β; γ i;N   1; (14)
KΓαΓβ

j 1; 2
αβγ i;n
In Eq. (9), fγ i;n g are the region boundaries. We also select where Λα; β; γ i;n   γ1i;n G3;0
0;3  α; β , and where
γc
γ i;N1  ∞, and μi θ; γ  0 if γ i < γ i;1, ∀ i ∈ f1; 2g. The m;n
Gp;q · is the Meijer’s G-function {[14], Eq. (9.301)}.
region boundaries for each channel are determined by Equation (14) is numerically solved to obtain the value
maximizing ESE of the channel subject to the average of fλi g, ∀i ∈ f1; 2g. For discrete-rate adaptation, ESE pro-
transmit power per channel constraint. The region boun- vides maximum achievable spectral efficiency subject to
dary selection problem for the ith channel is given by a certain statistical-delay-QoS constraint specified by θ.
We evaluate ESE with respect to the QoS exponent as a
N Z
X γ i;n1
performance metric for our proposed AM and power alloca-
min exp−θT f Bnf γ i γdγ tion schemes. Evaluation of the ESE against the QoS expo-
fγ i;n g γ i;n
n0 nent illustrates the change of achievable spectral efficiency
N Z
X of our proposed adaptive transmission scheme as the
γ i;n1 2nc3 − 1
s:t: f γ i γdγ  1; statistical-delay-QoS requirements of the transmitted traf-
n1 γ i;n kγ
fics change. Because an OWC system supports different
γ i;0 < γ i;1 < γ i;2 <    < γ i;N < ∞; (10) classes of traffics having different link-layer delay require-
ments, it is practically important to evaluate ESE against
where γ i;0  0 and f γ i γ is the PDF of the instantaneous QoS exponents for OWC systems. We obtain ESE (in bits/s/
Hz) of a dual-channel coherent OWC with an ICO-based
SNR of the ith channel. The Lagrangian function of
AM and transmit-power allocation scheme as
Eq. (10) is given as
X
Z N Z γ SE θ 
EInd SE;i θ;
EInd (15)
γ i;1 X i;n1 i∈f1;2g
L f γ i γdγ  exp−θT f Bnf γ i γdγ
0 n1 γ i;n

N Z
! SE;i θ is the ESE of the ith channel, and it is
where EInd
X γ i;n1 2nc3 − 1 given as
 λi f γ i γdγ − 1
γ i;n kγ
n1
 X
N−1
X
N−1 1
 νn γ i;n1 − γ i;n  − ϵγ i;1 ; SE;i θ  −
EInd log F γi γ i;1   exp−θT f BnF γ i γ i;n1 
(11) θT f B n1
n1 
− F γ i γ i;n   exp−θT f BN1 − F γ i γ i;N  : (16)
where λi , fνn g, and ϵ are the Lagrangian multipliers. By
using the complementary slackness of the Karush–
Khun–Tucker (KKT) condition, we can show that ϵ  0 In Eq. (16), F γ i · is the cumulative distribution function
and fνn g  0. The local minimizer of Eq. (10) can be of the received SNR of the ith channel. By using
obtained by the following KKT condition: Eq. (2), Eq. (1), [29] [Eq. (07.34.03.0605.01)], and [14]
[Eq. (7.811.2)], we obtain

∂L  
0 1 αβγ o  1
∂γ i;n F γ i γ o   G2;1 : (17)
ΓαΓβ 1;3 γ c  α; β; 0
λi 2nc3 − 2n−1c3 
⇒ γ i;n  ; n  1; 2; …; N; (12)
K2−ηn−1 − 2−ηn  Note that if γ c → ∞, F γ i γ o  → 0. Therefore, for 0 < θ < ∞,
ESE of a dual-channel coherent OWC approaches 2N
θT B bits/s/Hz at asymptotically high average SNR. From
where η  c3 lnf 2 . The value of λi is obtained by satisfying
Eq. (12), we can show that for a given average SNR, the
the average transmit power per channel constraint, which
region boundaries for selecting higher-order modulations
is given as
increase with θ. As a result, with the increase of θ (or with
the decrease of required delay-bound violation probability),
XN
2nc3 − 1 the transmitter tends to pick lower-order modulations.
Eγ i;n γ −1  − Eγ i;n1 γ −1   1; (13) Consequently, ESE of the considered system decreases
n1
K
as the statistical-delay constraint becomes strict.
R Special Case 1: We first consider a loose statistical-
where Eγ o γ −1   γ∞o 1γ f γi γdγ i . By using Eq. (2), Eq. (1), [29] delay constraint, i.e., θ → 0. ESE of the ith channel (in
[Eq. (07.34.03.0605.01)], and [14] [Eq. (7.811.4)], we can the unit of bits/s/Hz) at a loose statistical-delay constraint
write Eq. (13) as is obtained as
Hassan et al. VOL. 10, NO. 3/MARCH 2018/J. OPT. COMMUN. NETW. 143

PN−1
logF γ i γ˜i;1   exp−θT f BnF γi γ˜i;n1  − F γi γ˜i;n   exp−θT f BN1 − F γ i γ˜i;N 
limEInd θ  lim n1
θ→0 SE;i θ→0 −θT f B
PN−1
n1 n exp−θT f BnF γ i γ˜i;n1  − F γ i γ˜i;n   N exp−θT f BN1 − F γ i γ˜i;N 
 lim P
θ→0 F γi γ˜i;1   N
n1 exp−θT f BnF γ i γ˜i;n1  − F γ i γ˜i;n   exp−θT f BN1 − F γ i γ˜i;N 

X
N X
N
 nF γi γ˜i;n1  − F γi γ˜i;n   N − F γ i γ˜i;n : (18)
n1 n1

In Eq. (18), fγ˜i;n g are region boundaries at a loose select M i θ; γ  2n if b


γ i;n ≤ γ i <b
γ i;n1 ;
statistical-delay constraint, and fγ˜i;n g are obtained as n  1; 2; ·; ·; N; ∀i ∈ f1; 2g: (21)
λ0 2nc3 − 2n−1c3 
γ˜i;n  limγ i;n  ; n  1; 2; …; N; In Eq. (21), b γ i;1  0.5Q−1 BERt 2 with Q−1 · is the
θ→0 K
γ i;n  2 K3 −1 , for
nc

γ˜i;N1  ∞; (19) inverse of the Gaussian Q function [3], b


n  2; 3; ·; ·; N, and b γ i;N1  ∞. The resultant ESE of a
where λ0 is obtained by applying Eq. (19) to Eq. (14). The CPVR transmission scheme is obtained as
total ESE of a dual-channel coherent OWC in the loose X 1
ECPVR θ  − logF γi b
γ i;1 
statistical-delay
P SE θ 
constraint is obtained as limθ→0 EInd SE
θT fB
2N − 2 N n1 γ i i;n  bits∕s∕Hz. Consequently, at loose
F γ˜ i∈f1;2g

statistical-delay constraints, an ICO achieves ESE similar X


N−1

to the ASE of a conventional VPVR adaptive transmission  exp−θT f BnFγ i b


γ i;n1  − Fγ i b
γ i;n 
n1
scheme [19]. Such a result is expected because the VPVR
adaptive transmission scheme does not consider any delay  exp−θT f BN1 − F γi b
γ i;N : (22)
constraint. When the region boundaries given by Eq. (19)
In a CPFR scheme, we also have μi θ; γ  1, ∀i ∈ f1; 2g.
are used for AM and transmit-power allocation with any
For the ith (i ∈ f1; 2g) channel, M nϕ  2nϕ (where
arbitrary QoS exponent, the resultant ESE is obtained as
nϕ ∈ f1; 2; ·; ·; Ng) order modulation is selected if
X
nϕ c 3
1 γ i ≥ γ ϕ ≜ 2 K −1 , and transmission is suspended if γ i < γ ϕ.
EVPVR θ  − log F γ i γ˜i;1  The ESE of a CPFR transmission scheme is obtained as
SE
i∈f1;2g
θT f B
X 1
X
N−1 ECPFR θ  − logF γ i γ ϕ 
 exp−θT f BnF γ i γ˜i;n1  − F γi γ˜i;n 
SE
i∈f1;2g
θT fB
n1
 exp−θT f Bnϕ 1 − F γ i γ ϕ : (23)
 exp−θT f BN1 − F γi γ˜i;N : (20)
IV. JCO WITH AVERAGE TRANSMIT-POWER CONSTRAINT
Special Case 2: We now consider the stringent
statistical-delay constraint, i.e., θ → ∞. For the stringent
statistical-delay constraint and continuous rate adapta- A. Continuous Rate Adaptation and Power
tion, transmit-power allocation approaches a fixed-rate Allocation
channel inversion scheme, i.e., limθ→∞ μi θ; γ  Kγσ i , where
σ is a constant [20]. We assume that there is a modulation We first summarize the transmit-power allocation assum-
order, M nd  2nd , such that M nd ≤ 1  σ < M nd 1 . ing continuous rate adaptation in a dual-channel coherent
Consequently, at the stringent statistical-delay constraint, OWC system [10]. Based on such results, we will derive
M nd -ary modulation is selected with a transmit power modulation selection and power adaptation rules for dis-
nd c 3
−1
μi θ; γ  2 Kγ i
, ∀i ∈ f1; 2g. The value of M nd is selected crete constellations. By using Eq. (4), the EC maximization
from the average transmit power per channel constraint, problem for a dual-channel coherent OWC system subject to
nd c 3
−1
i.e., Eγ i 2 Kγ i
 ≤ 1. Consequently, at the stringent delay con- JCO and the average transmit-power constraint is given by
straint, nd bits will be transmitted over each channel, " #
where nd  minfbc13 log2 1  EγK−1 c; Ng. 1 Y
−η
max − log E 1  Kμi θ; γγ i 
For performance comparison, we also consider two fμi θ;γg θ i∈f1;2g
conventional ICO-based transmission technologies, " #
X
namely, CPVR and constant-power fixed-rate (CPFR) s:t: E μi θ; γ  2: (24)
transmission schemes. In a CPVR transmission scheme, i∈f1;2g
the transmit power over each channel is kept constant
(i.e., Ps ∕2), and we have μi θ; γ  1, ∀i ∈ f1; 2g. The modu- We denote γ ϕ as the cutoff SNR. Following [10], the
lation orders are selected based on the following rule: power adaptation rules are summarized as follows.
144 J. OPT. COMMUN. NETW./VOL. 10, NO. 3/MARCH 2018 Hassan et al.

• Both channels are active (minγ 1 ; γ 2  > γ ϕ ): factors over the first and second channels are respectively
2 3 determined as
1 1
μi θ; γ  4 1 − 5 ; (25) 2n1 c3 − 1 2n2 c3 − 1
γ 12η K
2η η
12η γ γ 12η
Kγ i μ11 θ; γ  and μ1
2 θ; γ  : (30)
ϕ 1 2 Kγ 1 Kγ 2

where i ∈ f1; 2g and x  maxx; 0. Note that when both channels are active, each channel
• Only one channel is active (minγ 1 ; γ 2  < γ ϕ ≤ will be able to carry at least binary constellation.
maxγ 1 ; γ 2 ): Therefore, both channels will be active when minγ 1 ; γ 2  ≥
γ
 Kϕ 22η1c3 . The probability of jointly transmitting n1 and
1 1 n2 bits over the first and second channels, respectively, is
μi θ; γ  1 η
− ; μj θ; γ  0; (26)
γ ϕ Kγ i 
1η 1η
Kγ i given as

where i; j ∈ f1; 2g, i ≠ j, and γ i > γ j . Fn1 ; n2   F γ 1 Γ1 1


n1 1;n2  − F γ 1 Γn1 ;n2 
• No channel is active: maxγ 1 ; γ 2  < γ ϕ and
× F γ2 Γ1 1
n2 1;n1  − F γ 2 Γn2 ;n1 : (31)
μi θ; γ  0, ∀i ∈ f1; 2g.

In the case of continuous rate adaptation, the value of γ ϕ


is obtained by applying Eqs. (25) and (26) to the average 2) Only the Channel With Maximum Instantaneous
γ
transmit-power constraint given by [10] [Eq. (23)]. SNR Is Active: If minγ 1 ; γ 2  <  Kϕ 22η1c3 , only the
channel with maximum instantaneous SNR can remain
active. We need to consider the following cases:
γ γ
B. AM and Transmit-Power Allocation (1) γ m ≜ maxγ 1 ; γ 2  ≥  Kϕ 22η1c3 and (2)  Kϕ 2η1c3 ≤
γ ϕ 2η1c3
γ m <  K 2 where m  arg maxγ 1 ; γ 2 .
γ
We consider the following scenarios: (1) both channels We first consider γ m ≥  Kϕ 22η1c3. By using Eq. (26), we
are active and modulation orders are jointly selected, obtain the relation between the assigned constellation size
(2) only the channel with maximum instantaneous SNR and maximum instantaneous channel SNR as
is active, and (3) no channel is active.
 1
Kγ m 1η
1) Both Channels Are Active and Modulation Orders Are M cm3 θ; γ  : (32)
γϕ
Jointly Selected: By using Eq. (25), the relations between
the assigned constellation sizes and instantaneous SNRs
in a JCO approach are obtained as By using Eq. (32), we obtain the modulation order
selection and power adaptation rule for the channel with
Kγ 1 Kγ 2 maximum SNR as
M c13 θ;γ  1 2η η
; M c23 θ;γ  1 2η η
:
γϕ K
12η 12η γ 1 γ 2 
12η γϕ K
12η 12η γ 1 γ 2 12η 2nc3 − 1
select M m θ; γ  2n and μ2
m θ; γ 
(27) Kγ m
if Γ2 2
n ≤ γ m < Γn1 ; n  1; 2; ·; ·; ·; N: (33)
From Eq. (27), we obtain
  γ γ
γϕ In Eq. (33), Γ2
1   K 2
ϕ 2η1c3
, Γ2 nc η1
n   K 2 3
ϕ
for
γ1  M c13 θ; γM 1 θ; γM 2 θ; γηc3 ;
K n  2; 3; ·; ·; ·; N and Γ2
N1  ∞.
 
γϕ
γ2  M c23 θ; γM 1 θ; γM 2 θ; γηc3 : (28) We now consider that maximum instantaneous channel
K SNR satisfies the following condition:
   
We denote M ni  2ni as the modulation order selected for γ ϕ η1c γ ϕ 2η1c
2 3 ≤ γ < 2 3: (34)
the ith channel where ni ∈ f1; 2; ·; ·; ·; Ng and i ∈ f1; 2g. We K m
K
use Eq. (28) for developing the modulation order selection
rule in a JCO approach. M n1 -ary and M n2 -ary modulations
If Eq. (34) is satisfied, only binary constellation will
will be jointly selected for the first and second channels,
be transmitted over the channel having maximum
respectively, if γ 1 and γ 2 jointly satisfy the following
instantaneous SNR with a transmit-power adaptation
conditions:
factor μ2 2c3 −1
m θ; γ  Kγ m . The following cases can happen:
γ γ γ
Γ1 1
n1 ;n2 ≤ γ 1 < Γn1 1;n2 and Γ1 1
n2 ;n1 ≤ γ 2 < Γn2 1;n1 ; (29) (1) fγ 1 ; γ 2 g ∈  Kϕ 2c3 η1 ;  Kϕ 2c3 2η1  and (2) γ i ∈ 0;  Kϕ 
γ γ
2c3 η1 , γ j ∈  Kϕ 2c3 η1 ;  Kϕ 2c3 2η1 , ∀i; j ∈ f1; 2g, and
γ
where Γ1
n;m   K 2 3
ϕ ηc nmnc3 and Γ1
N1;m  ∞, i ≠ j. We obtain the probability of transmitting binary
∀fn; mg ∈ f1; 2; ·; ·; ·; Ng. To satisfy the target BER require- constellation over the channel having maximum instanta-
ments, the corresponding transmit-power adaptation neous SNR as
Hassan et al. VOL. 10, NO. 3/MARCH 2018/J. OPT. COMMUN. NETW. 145

Z
γϕ 2c3 2η1 Z γ2 M cn3 − 1
P
K

γϕ
γϕ f γ1 ;γ2 γ 1 ; γ 2 dγ 1 dγ 2 An γ a ; γ b  
γ2  2c3 η1 γ1  2c3 η1
KΓαΓβ
K K    
Z
γϕ 2c3 2η1 Z αβγ a  2  2
3;0 αβγ b 
γ1 × G3;0 − G
γ c  1; α; β γ c  1; α; β
K 0;3 0;3

γ ϕ
γϕ f γ1 ;γ 2 γ 1 ; γ 2 dγ 1 dγ 2
c3 η1
γ1  2 γ2  2c3 η1
K K
(38)
Z
γϕ 2c3 η1 Z
γϕ 2c3 2η1
K K

γ ϕ f γ 1 ;γ 2 γ 1 ; γ 2 dγ 1 dγ 2
γ 1 0 γ2  K 2c3 η1 and
Z
γϕ 2c3 η1 Z
γϕ 2c3 2η1
Z
K K

γ ϕ f γ 1 ;γ 2 γ 1 ; γ 2 dγ 1 dγ 2 : (35)
γ 2 0 γ1  2c3 η1
γb 2c3 − 1
K Gi γ a ; γ b   F γ j γ i f γi γ i dγ i ; (39)
γa Kγ i

where i; j ∈ f1; 2g; i ≠ j. Equation (37) is numerically


In Eq. (35), f γ1 ;γ2 γ 1 ; γ 2  is the joint PDF of γ 1 and γ 2 . solved to obtain the value of γ ϕ used in Eqs. (29)–(36).
We assume that γ 1 and γ 2 are i.i.d. RVs. Therefore, we The closed-form ESE (in bits/s/Hz) of a JCO approach is
have f γ 1 ;γ2 γ 1 ; γ 2   f γ 1 γ 1 f γ 2 γ 2  and F γ1 r  F γ2 r ≜ obtained as
F γ r for any r > 0. Consequently, when Eq. (34) is satisfied,
we obtain the probability of transmitting binary constella- XN X N
tion over the channel having maximum instantaneous 1
SE θ  −
EJoint log Fn1 ;n2 exp−θT f Bn1 n2 
SNR as θT f B n 1 n 1
1 2
   N
γ ϕ c 2η1 X 2
2F γ 23 F γ Γn1 
   2    2 K n1
γ ϕ 2η1c γ ϕ η1c
P  Fγ 2 3 − Fγ 2 3 : (36) −F γ Γ2
n exp−θT f Bn
K K    2    2 
γ ϕ 2η1c γ ϕ η1c
 Fγ 2 3 − Fγ 2 3
K K
   2
γ ϕ 2η1c
γ
×exp−θT f B F γ 2 3 : (40)
3) No Channel Is Active: When maxγ 1 ; γ 2  <  Kϕ 2η1c3 , K
transmission over both channels is suspended, i.e.,
μi θ; γ  0, ∀i ∈ 1; 2. We can show that ϵn ≜ limθ→0 Γ1 2
n;m  limθ→0 Γn 
γ
Considering the aforementioned scenarios, we obtain the  Kϕ 2nc3 .
After some algebraic manipulations, we obtain
average transmit-power constraint in JCO-based AM and ESE of the JCO approach in the loose statistical-delay
power allocation as constraint as

   
P PN
2 N− N F ϵ
n1 γ n  1 − F ϵ
γ 1   2F ϵ
γ 1  N − F ϵ
n1 γ n  X
N

SE θ 
limEJoint  2N − 2 F γ ϵn  bits∕s∕Hz: (41)
θ→0 1 − F γ ϵ1 2  2F γ ϵ1 1 − F γ ϵ1   F 2γ ϵ1  n1

X
N X
N
From Eq. (41), it is evident that both ICO and JCO
F γ 2 Γ1 1 1 1
n2 1;n1  − F γ 2 Γn2 ;n1 An1 Γn1 ;n2 ; Γn1 1;n2  achieve similar ESE performances at the loose statisti-
n1 1 n2 1
cal-delay constraint. Such an observation is also depicted
N X
X N
from our numerical results.
 F γ 1 Γ1 1 1 1
n1 1;n2  − F γ 1 Γn1 ;n2 An2 Γn2 ;n1 ; Γn2 1;n1 
n1 1 n2 1

X    N
γ ϕ c η1 X C. Computational Complexity of ICO and JCO
 F γj≠i23 Ani Γ2 2
ni ; Γni 1 
i∈f1;2g
K ni 1
X γ ϕ   
γ ϕ

1) Complexity of ICO: To analyze the complexity of an
 Gi 2c3 η1 ; 2c3 2η1  2: (37)
K K adaptive transmission scheme, we need to determine the
i∈f1;2g
complexity of calculating the region boundaries and the
complexity of performing AM and power allocation for
In Eq. (37), An γ a ; γ b  and Gi γ a ; γ b  are, respectively, given region boundaries. The region boundaries of ICO-
defined as based AM and power allocation are obtained by jointly
146 J. OPT. COMMUN. NETW./VOL. 10, NO. 3/MARCH 2018 Hassan et al.

solving Eqs. (12) and (14). We emphasize that Eq. (14) A. ICO
depends only on the channel statistics and statistical-
delay-QoS requirements of the traffics. Hence, Eq. (14)
For an ICO, we obtain the solution to Eq. (42) as
can be solved offline. Consequently, in an ICO-based
AM and power allocation scheme, the region boundaries 2 3
are calculated offline, and such region boundaries 4 1 1 1 5 ;
i θ; γ 
xInd η
− ∀i ∈ f1; 2g; (43)
are stored at the transmitter. Therefore, the required γ a;I
η1
Kγ i 1η Kγ i
complexity for calculating the region boundaries can be
affordable. In an online operation, ICO independently
where γ a;I is defined as
selects the modulation order and corresponding power
adaptation factor for each optical channel by using the 0 1−η1
stored region boundaries in Eq. (9). We can show that
B 1 1
 1
C
γ a;I  B η C
Kγ 1 Kγ 2
@ 1 1η 
our proposed ICO-based AM and power allocation : (44)
η A
require a simple mapping in online operation. In Kγ 1  1
Kγ 2
1η

particular, the maximum complexity of such a mapping


is ON.
For a target BER, the AM and transmit-power allocation
2) Complexity of JCO: The region boundaries for over each channel are performed according to the following
JCO-based AM and power allocation depend on the value rule:
of γ ϕ and delay-QoS requirements of the transmitted
traffics. The value of γ ϕ is obtained by numerically solving select M i θ; γ  2ni and Pi θ; γ  xi θ; γPpeak
Eq. (37). Depending on the size of available constellations,
if νni ≤ γ i < νni 1 ; ni  1; 2; ·; ·; N; ∀i ∈ f1; 2g: (45)
the numerical evaluation of Eq. (37) can be computation-
ally extensive. However, Eq. (37) can be numerically γ
solved offline because Eq. (37) depends only on the channel In Eq. (45), νni   Ka;I 2ni c3 η1 , ni  1; 2; ·; ·; N, νN1  ∞,
statistics and statistical-delay-QoS requirements of the and Pi θ; γ  0 if γ i < ν1, ∀i ∈ f1; 2g. In what follows,
traffics. As a result, the region boundaries of JCO-based we discuss the complexity for AM and power allocation
AM and power allocation are calculated offline at the trans- by using Eq. (45). For the peak transmit-power constraint,
mitter with affordable complexity. Moreover, such region region boundaries of ICO-based AM and power allocation
boundaries are stored at the transmitter for online opera- require online computation. Specifically, the region boun-
tion. In an online operation, JCO needs to find a suitable daries for using Eq. (45) require the value of γ a;I . At each
condition, from Eqs. (29), (33), and (34), that is satisfied for coherence time, the value of γ a;I can be easily obtained from
a given set of instantaneous SNRs. We can show that Eq. (44) based on the received CSI, target BER, and QoS
the online operation of JCO requires maximum ON 2  iter- exponent. The OWC transmitter requires N computations
ations to select suitable modulation orders and power to calculate the region boundaries. For given region boun-
adaptation factors for both optical channels. Because daries, each optical channel independently requires at
atmospheric turbulence fading has a large channel coher- most N  1 searches to select a suitable modulation
ent time, such a computation can be affordable within order. Hence, at each coherence time, the total computa-
channel coherence time. tional complexity of ICO is ON  2N  1 ≈ ON.
Consequently, with the peak transmit-power constraint,
ICO-based AM and power allocation have linear computa-
tional complexity.
V. ADAPTIVE TRANSMISSION SUBJECT TO THE PEAK
TRANSMIT-POWER CONSTRAINT
B. JCO
We denote the available peak transmit power as Ppeak.
The transmit power of the ith channel is given by For JCO, the solution to Eq. (42) is obtained as
Pi θ; γ  xi θ; γPpeak , where xi θ; γ ∈ 0; 1 is the peak
power allocation factor. The EC maximization problem 1 1
for a peak-power-constrained dual-channel coherent OWC xJoint
i θ; γ  1 η
− ; ∀i ∈ f1; 2g; (46)
γ a;J γ 1 γ 2  K
12η 1η

12η
Kγ i
system is given by

where γ a;J is defined as


 Y
1 0 1−2η1
max − log Eγ1 ;γ2 1  Kxi θ; γγ i −η
θT f B
fxi θ;γg 1  Kγ1 1  Kγ1 2
γ a;J  @ A
i∈f1;2g
X η 2η : (47)
s:t: xi θ; γ  1; xi θ; γ ∈ 0; 1; ∀i ∈ f1; 2g: (42) 2γ 1 γ 2 −12η K −12η
i∈f1;2g

Equation (46) is an optimal solution to Eq. (42) if


RΔL Ppeak
In Eq. (42), γ i  γ p Ii , ∀i ∈ f1; 2g where γ p  qB . xi θ; γ > 0, ∀i ∈ f1; 2g. If such a condition is not satisfied,
Hassan et al. VOL. 10, NO. 3/MARCH 2018/J. OPT. COMMUN. NETW. 147

all the available transmit power is allocated to the channel AM and power allocation also have linear computational
with maximum SNR. In such a case, we have xm θ; γ  1, complexity of ON.
xn≠m θ; γ  0, and m  arg maxγ 1 ; γ 2 . By using the
aforementioned results, we develop JCO-based AM and Pseudo Code 2: SCO-based AM and power allocation sub-
transmit-power allocation subject to the peak transmit- ject to a peak transmit-power constraint:
power constraint. We consider the following cases: (1) both γ
1: if minγ 1 ; γ 2  ≥  Ka;I 2c3 η1 then
channels are active and AMs are jointly selected, (2) both
2: Find n1 ∈ f1; 2; ·; ·; Ng: νn1 ≤ minγ 1 ; γ 2  < νn1 1 .
channels are active and AMs are independently selected,
(3) only the channel with maximum SNR is active, Select M n1  2n1 order modulation over the chan-
and (4) no channel is active. We denote Πn1 ;n2  nel having minimum instantaneous SNR with
γ
 Ka;J 2n1 c3 ηc3 n1 n2  . The key steps of JCO-based AM and Pm θ;γ  xIndm θ;γPpeak where m  arg minγ 1 ; γ 2 .
power allocation are given in pseudo code 1. Similar to 3: Find n2 ∈ fn1 ; n1  N; ·; ·; Ng: νn2 ≤ maxγ 1 ; γ 2  <
ICO, region boundaries of JCO also require online compu- νn2 1 . Select M n2  2n2 order modulation over the
tation for peak transmit-power-constrained OWC. In channel having maximum instantaneous SNR
particular, with a peak transmit-power constraint, the with Pm θ; γ  xInd m θ; γPpeak .
region boundaries of JCO-based AM and power allocation 4: else if ∃nm ∈ f1; 2; ·; ·; Ng: 2nm c3 ≤ 1  K maxγ 1 ; γ 2  <
are calculated based on the values of γ a;I and γ a;J . However, 2nm 1c3 then
both γ a;I and γ a;J require simple calculations as depicted 5: Select M nm  2nm order modulation only over
from Eqs. (44) and (47), respectively. At each coherence the channel having maximum SNR with
time, the OWC transmitter requires N 2  N computations Pm θ; γ  Ppeak .
to calculate the region boundaries. For a given set of region 6: else
boundaries, a maximum N 2  2N  1 iterations are 7: Suspend the transmission over both channels.
required to select suitable modulation orders for both 8: end if
optical channels. Hence, at each coherence time, the
total computational complexity of JCO is O2 N 2 
2N  1  N ≈ ON 2 . Consequently, with a peak trans- VI. NUMERICAL RESULTS
mit-power constraint, JCO-based AM and power allocation
have quadratic computational complexity.
In this section, we present some selected numerical
results to demonstrate the performance of the proposed
adaptive transmission schemes. We consider strong gamma–
Pseudo Code 1: JCO-based AM and power allocation gamma turbulence fading with α  2.04, β  1.10; weak
subject to a peak transmit-power constraint gamma–gamma turbulence fading with α  4.43,
γ
1: if minγ 1 ; γ 2  ≥  Ka;J 2c3 2η1 and ∃n1 ; n2  ∈ f1; 2; ·; ·; Ng: β  4.39; and K-turbulence fading (which is a special case
Πn1 ;n2 ≤ γ 1 < Πn1 1;n2 and Πn2 ;n1 ≤ γ 2 < Πn2 1;n1 then of strong gamma–gamma turbulence fading with β  1) with
2: Select M n1  2n1 and M n2  2n2 order modulations α  1.99 and α  1.11. In all the numerical results, we
over the first and second channels, with P1 θ; γ  consider that N  4 constellations are available for each
xJoint θ; γPpeak and P2 θ; γ  xJoint θ; γPpeak . channel. We assume the following system parameters:
1 2
3: else if minγ 1 ; γ 2  ≥  K 2
γ a;I c3 η1
then R  0.75 A∕W, T f  10−6 s, and B  108 Hz.
4: Find ni ∈ f1; 2; ·; ·; Ng: νni ≤ γ i < νni 1 . Select M ni  Figure 2 compares ESE of different AMs by considering
2ni order modulation over the ith channel with ICO and the average transmit-power constraint over the
Pi θ; γ  xInd gamma–gamma turbulence channels. Figure 2 shows that
i θ; γPpeak , ∀i ∈ f1; 2g.
5: else if ∃nm ∈ f1; 2; ·; ·; Ng: 2nm c3 ≤ 1  K maxγ 1 ; γ 2  < the ESE obtained from our developed expression matches
2nm 1c3 then well with the ESE obtained from Monte Carlo simulations.
6: Select M nm  2nm order modulation only over Figure 2 depicts that ESE of a given M-ary modulation de-
the channel having maximum SNR with creases as the value of θ increases. In other words, for an
Pm θ; γ  Ppeak . OWC system, maximum spectral efficiency is achieved at
7: else loose delay constraints, and the achievable spectral effi-
8: Suspend the transmission over both channels. ciency gradually decreases as the delay constraint becomes
9: end if strict. Such an observation is consistent with our analysis
in Section III. Consequently, delay-QoS requirements of
the transmitted traffics indeed influence the achievable
spectral efficiency of an OWC system. Figure 2 also
C. SCO depicts the superiority of adaptive M-QAM over adaptive
M-PSK and M-PAM in both weak and strong turbulence fad-
In SCO, the channels are ranked according to the in- ing channels. The region boundaries in an adaptive M-QAM
creasing value of their instantaneous SNRs. AM over a are smaller compared to the region boundaries in adaptive
given channel is performed based on the instantaneous M-PSK and M-PAM. Consequently, the considered coherent
SNR of that channel and the modulation order selected OWC transmitter tends to select higher-order modulation
for the previous ranked channel. The key steps of SCO without sacrificing the target BER requirements when
are given as pseudo code 2. Similar to ICO, SCO-based adaptive M-QAM-based ICO is employed. As a result, an
148 J. OPT. COMMUN. NETW./VOL. 10, NO. 3/MARCH 2018 Hassan et al.

9 constraints. However, JCO outperforms ICO as θ becomes


large, i.e., statistical-delay constraints become strict. Such a
8
performance gap is more pronounced in the strong turbu-
Effective Spectral Efficiency (bits/s/Hz)

7 lence fading. Figure 3 shows that at θ  10−1 , JCO achieves


8.5% and 38.4% higher ESE compared to ICO in weak and
6 strong gamma–gamma turbulence fading, respectively.
Figure 3 also depicts that ESE of the VPVR scheme con-
5
verges to the ESE of ICO and JCO at loose statistical-delay
4 constraints, which is consistent with our analysis. However,
ESE of the VPVR scheme significantly drops at strict stat-
3 Monte Carlo Simulation istical-delay constraints. At θ  10−1 , JCO achieves 3.5 and
Adaptive M-QAM, weak turbulence
2 Adaptive M-PSK, weak turbulence 4.7 times larger ESE compared to the VPVR scheme in weak
Adaptive M-PAM, weak turbulence and strong gamma–gamma turbulence fading, respectively.
1 Adaptive M-QAM, strong turbulence ESE of constant transmit-power-based schemes, such as
Adaptive M-PSK, strong turbulence
Adaptive M-PAM, strong turbulence CPVR and 16-QAM-based CPFR, are also plotted in
0
10-5 10-4 10-3 10-2 10-1 Fig. 3. It is evident from Fig. 3 that the proposed ICO
QoS-exponent and JCO significantly outperform both CPVR and 16-
QAM-based CPFR transmission schemes, especially in the
Fig. 2. ESE comparison among different M-ary modulations strict statistical-delay constraints. We have already ex-
with ICO, the average transmit-power constraint, γ c  30 dB, plained in Subsection IV.C that our proposed ICO- and
and BERt  10−6 . JCO-based adaptive transmissions have reasonable com-
plexity. Therefore, our proposed AM and transmit-power
allocation provide large delay-QoS-aware throughput gain
adaptive M-QAM-based ICO provides larger ESE compared without requiring significant computational complexity.
to the adaptive M-PSK and M-PAM-based ICO techniques. Such an observation justifies the use of proposed AM and
Due to superior ESE performance over the fading channels, power allocation for delay-QoS-constrained OWC instead
we consider adaptive M-QAM in the subsequent numerical of the constant transmit-power-based schemes.
results. Figure 4 compares ESE of ICO and JCO over the
Figure 3 compares ESE of ICO and JCO over the K-turbulence fading channels considering the average
gamma–gamma turbulence fading channel considering transmit-power constraint. Comparing Figs. 3 and 4, we
the average transmit-power constraint. Here, ESE of observe that the ESE gain of JCO in the strict statistical-
ICO and JCO are obtained by evaluating Eqs. (15) and delay constraint becomes more prominent as the turbu-
(40), respectively. Figure 3 depicts that both ICO and lence fading becomes stronger. In Fig. 4, JCO improves
JCO achieve similar ESE for small values of θ (i.e., in ESE of ICO by 36.8% at θ  10−1 and α  1.99. However,
the loose statistical-delay constraints). Such a result is JCO achieves almost 2.05 times larger ESE compared to
expected because we have analytically showed that ESE ICO for the same value of θ when α  1.11. Hence, our pro-
of ICO and JCO converges at loose statistical-delay posed JCO efficiently supports strict delay constraints over

9 7
Weak Turbulence

8
6
Effective Spectral Efficiency (bits/s/Hz)
Effective Spectral Efficiency (bits/s/Hz)

7
5
6 =1.99

4 =1.11
5
Strong Turbulence
4 3
JCO, =1.99, Eq. (40)
3 JCO, =1.11, Eq. (40)
JCO, Eq. (40) 2 ICO, =1.99, Eq. (15)
2 ICO, Eq. (15) ICO, =1.11, Eq. (15) =1.99
Weak Turbulence
VPVR, Eq. (20) 1 VPVR, Eq. (20)
1 CPVR, Eq. (22) CPVR, Eq. (22) =1.11
16-QAM CPFR, Eq. (23) Strong Turbulence 16-QAM CPFR, Eq. (23)
0 0
10-5 10-4 10-3 10-2 10-1 10-5 10-4 10-3 10-2 10-1
QoS-exponent QoS-exponent

Fig. 3. ESE comparison among different M-QAM-based adaptive Fig. 4. ESE comparison among different M-QAM-based adaptive
transmissions in gamma–gamma turbulence fading with the aver- transmissions in K-turbulence fading with the average transmit-
age transmit-power constraint, γ c  30 dB, and BERt  10−6 . power constraint, γ c  30 dB, and BERt  10−8 .
Hassan et al. VOL. 10, NO. 3/MARCH 2018/J. OPT. COMMUN. NETW. 149

strong turbulence fading channels. Figure 4 also depicts 8

ESE of the conventional VPVR, CPVR, and 16-QAM-based


7
CPFR over the K-turbulence channels. Our proposed adap-

Effective Spectral Efficiency (bits/s/Hz)


tive transmission schemes significantly outperform the
6
conventional adaptive transmission schemes in the strict
statistical-delay constraints, especially when α becomes 5 ICO, Target BER=10 -6
small (i.e., when severity of the K-turbulence fading in- -6
JCO, Target BER=10
creases). From Figs. 3 and 4, we obtain the following in- 4
sights on the AM and power allocation for OWC SCO, Target BER=10 -6
-6
systems: (1) only channel-aware adaptive transmission 3 EPA, Target BER=10
-8
schemes, such as VPVR and CPVR, do not support strict ICO, Target BER=10
delay-QoS constraints, and (2) to achieve suitable spectral 2
JCO, Target BER=10 -8
efficiency in strict delay-QoS requirements, AM and power SCO, Target BER=10
-8
1
allocation for an OWC system should jointly consider the -8
EPA, Target BER=10
variation of optical transmission link as well as statistical- 0
delay-QoS requirements of the transmitted traffics. 10-5 10-4 10-3 10-2 10-1 100
QoS-exponent
Due to the intractability of the BER expressions, it is
challenging to investigate the impact of the uncompen-
sated phase noise on the proposed adaptive transmission Fig. 6. ESE comparison among M-QAM-based ICO, JCO, SCO,
schemes with M-PSK and M-QAM. However, such an and EPA in weak gamma–gamma turbulence fading subject to a
peak transmit-power constraint (Ppeak  100 mW.)
intractability is not present when M-PAM is used. Due
to space limitations, we omit the detailed analysis of the
ESE of M-PAM-based ICO and JCO with uncompensated
within 10°. Figure 5 also depicts that JCO outperforms
phase noise. In Fig. 5, we compare ESE of ICO and JCO
ICO even in the presence of uncompensated phase noise,
employing adaptive M-PAM over the K-turbulence fading
especially in the strict statistical-delay constraints.
in the presence of uncompensated phase noise. We consider
that the transmitter has perfect knowledge of the instan- Figure 6 illustrates ESE of the proposed ICO-, JCO-, and
taneous channel gains and channel statistics. However, the SCO-based adaptive transmissions in the presence of a peak
transmitter does not have knowledge about the phase noise transmit-power constraint and different target BER require-
at the receiver. Consequently, the transmitter adapts ments. For a peak transmit-power constraint, we evaluate
modulation orders and transmit powers without consider- ESE of ICO by simulating Eq. (45), and ESE of JCO and
ing phase noise at the receiver. Figure 5 depicts that both SCO by simulating the pseudo code 1 of Subsection V.B
ICO- and JCO-based adaptive transmission schemes expe- and pseudo code 2 of Subsection V.C, respectively.
rience ESE degradation as the standard deviation of the Figure 6 also depicts ESE of an equal power allocation
uncompensated phase noise increases. However, the del- (EPA)-based adaptive transmission scheme. In an EPA,
eterious impact of phase noise on ESE is little when the the available peak transmit power is equally allocated be-
standard deviation of the uncompensated phase noise is tween both optical channels, i.e., xi θ; γ  0.5, ∀i ∈ f1; 2g.
The AMs in an EPA are selected according to Eq. (45), where
ni c 3
−1
the region boundaries are given by νni  20.5 K , ni 
3.5
1; 2; ·; ·; N, ∀ i ∈ f1; 2g. Figure 6 depicts that EPA achieves
optimal ESE only at loose statistical-delay constraints. At
Effective Spectral Efficiency (bits/s/Hz)

3 strict statistical-delay constraints, ESE of EPA-based adap-


tive transmission becomes almost zero regardless of the tar-
2.5 get BER requirements. Consequently, an EPA does not
provide strict statistical-delay-QoS guarantee to a peak
ICO, Without Phase Noise transmit-power-constrained dual-channel coherent OWC
2
ICO, With Phase Noise =10
0 system. Figure 6 also depicts that the proposed ICO, JCO,
and SCO achieve similar ESE at the loose statistical-delay
1.5 ICO, With Phase Noise =30 0 constraints. However, JCO outperforms both ICO and SCO
JCO, Without Phase Noise at strict statistical-delay constraints. Moreover, in Fig. 6 we
0 observe that SCO outperforms ICO at strict statistical-delay
1 JCO, With Phase Noise =10
0 constraints. The reason is that in an ICO, transmit power is
JCO, With Phase Noise =30
always allocated to both channels regardless of the channel
0.5
10-5 10-4 10-3 10-2 10-1 fading conditions. In an SCO, the transmission of a given
QoS-exponent channel is suspended when the fading of that channel be-
comes severe, and all the transmission power is allocated
Fig. 5. ESE comparison between coherent M-PAM-based to the other channel. Consequently, SCO offers an efficient
ICO and JCO in K-turbulence fading (α  1.99) with the average utilization of the available transmit power and improved
transmit-power constraint, imperfect phase noise compensation, ESE compared to ICO. Note that as the target BER
γ c  30 dB, and target BER  10−8 . decreases, the transmitter tends to pick lower-order
150 J. OPT. COMMUN. NETW./VOL. 10, NO. 3/MARCH 2018 Hassan et al.

modulations to satisfy the BER requirements subject to 100


the same peak transmit-power constraint. Consequently,

] max
it is depicted in Fig. 6 that ESE of the proposed adaptive 10-2

Delay-bound violation probability, Pr[D>D


transmission schemes decreases as the target BER is
reduced. 10-4

Figure 7 illustrates ESE versus delay-bound (i.e., maxi-


mum tolerable delay for a transmitted frame) trade-off of 10-6

JCO, ICO, and SCO considering the peak transmit-power


constraint. Figure 7 shows that ESE of the proposed adap- 10-8
JCO, =2.04, =1.10, Target BER=10 -6
tive transmission schemes decreases as the delay-bound
-6
gets smaller and/or target delay-bound violation probabil- 10-10 ICO, =2.04, =1.10, Target BER=10
ity becomes tighter. We can show that the values of θ in- JCO, =1.11, =1.00, Target BER=10 -8
-12
crease when the delay-bound and/or target delay-bound 10 ICO, =1.11, =1.10, Target BER=10 -8
violation probability decrease. Moreover, the region boun-
daries in our proposed adaptive transmission schemes 10-14 -5
10 10-4 10-3 10-2 10-1
increase as the values of θ increase. Therefore, the trans- QoS-exponent
mitter tends to pick lower-order modulations when the de-
lay-bound and/or target delay-bound violation probability Fig. 8. Delay-bound violation probability of M-QAM-based ICO
decrease. As a result, ESE of the proposed adaptive trans- and JCO subject to the average transmit-power constraint.
mission schemes decreases for smaller delay-bound and/or
target delay-bound violation probability. However, ESE of a same delay-bound violation probability for small values
given channel optimization becomes independent of the of θ. Such an observation is expected because ESE of both
target delay-bound violation probability when the delay JCO and ICO converges for small values of θ. However,
bound becomes sufficiently large, as depicted in Fig. 7. delay-bound violation probability of both ICO and JCO de-
Figure 7 shows that JCO outperforms both ICO and creases as θ increases. For larger values of θ, we have the
SCO regardless of the delay-bound and/or target delay- following two observations: (1) for a given set of physical
bound violation probability. SCO also outperforms ICO system parameters (i.e., turbulence fading and target
in both strict and loose delay bounds as depicted in Fig. 7. BER), JCO achieves smaller delay-bound violation proba-
Figure 8 illustrates the change of delay-bound violation bility compared to ICO, and (2) compared to ICO, JCO
probability (i.e., the probability of exceeding link-layer experiences less increase of the achievable delay-bound
delay beyond Dmax ) with respect to the change of QoS ex- violation probability as the turbulence fading gets strong
ponents. Without loss of generality, we assume that total and/or the target BER requirement becomes more strict.
tolerable wait time at transmit buffer in a typical TS is Both observations can be explained by the following
1 ms. We also assume that a typical TS contains 1000 arguments. In our previous numerical results, we have
frames, and all the frames experience the same queuing demonstrated that JCO achieves larger ESE compared to
delay. Consequently, in Fig. 8 we have Dmax  10−6 s. ICO in large QoS exponents. Moreover, in large QoS expo-
Figure 8 illustrates that both JCO and ICO achieve the nents, the ESE performance gap between JCO and ICO
increases as the turbulence fading gets strong and/or
6 the target BER requirement becomes more strict. Due to
larger ESE, JCO has a better probability of satisfying
5.5
the delay bound even for strong turbulence fading and/or
Effective Spectral Efficiency (bits/s/Hz)

5 strict target BER requirements. Accordingly, for a delay-


constrained dual-channel OWC system, it is beneficial to
4.5
jointly adapt the transmit parameters of both optical chan-
4 nels. JCO has higher implementation complexity compared
3.5
to ICO. Therefore, JCO should be employed if the consid-
-4
ICO, Delay-bound violation probability=10 ered OWC system can afford the implementation complex-
3 -4 ity. However, ICO is preferable when the considered OWC
JCO, Delay-bound violation probability=10
2.5 SCO, Delay-bound violation probability=10
-4 system requires simple implementation complexity and/or
-7 only supports applications with nonstrict delay-QoS
2
ICO, Delay-bound violation probability=10
-7 constraints.
JCO, Delay-bound violation probability=10
1.5 -7
SCO, Delay-bound violation probability=10
1
2 4 6 8 10 12 14 16 18 20
VII. CONCLUSION
Frame Delay-Bound, D max ( s)
We developed statistical-delay-QoS-aware AM and
Fig. 7. ESE versus delay-bound performance comparison of power allocation for a dual-channel coherent OWC sys-
M-QAM-based ICO, JCO, and SCO in strong gamma–gamma tur- tem over the gamma–gamma turbulence fading channels.
bulence fading subject to Ppeak  100 mW, 1 Gbps constant traffic Our proposed adaptive transmission schemes improve
arrival rate, and BERt  10−6 . the statistical-delay-QoS requirements versus achievable
Hassan et al. VOL. 10, NO. 3/MARCH 2018/J. OPT. COMMUN. NETW. 151

spectral efficiency trade-off by considering the average and atmospheric turbulence channels,” Opt. Express, vol. 18,
peak transmit-power constraints. Our numerical results no. 19, pp. 20445–20454, Sept. 2010.
provided the following two observations: (1) our proposed [13] A. Al-Habash, L. C. Andrews, and R. L. Phillips,
AM and power allocation achieve significantly larger “Mathematical model for the irradiance probability density
ESE compared to the conventional adaptive transmission function of a laser beam propagating through turbulent
media,” Opt. Eng., vol. 40, no. 8, pp. 1554–1562, Aug. 2001.
schemes in the strict statistical-delay constraints, and
(2) the proposed JCO-based adaptive transmission pro- [14] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series,
and Products, 7th ed. San Diego, CA: Academic, 2007.
vides improved ESE and delay-bound violation probability
[15] N. Wang and J. Cheng, “Moment-based estimation for the
compared to the proposed ICO- and SCO-based adaptive
shape parameters of the gamma–gamma atmospheric
transmissions in strong turbulence fading.
turbulence model,” Opt. Express, vol. 18, no. 12, pp. 12824–
12831, June 2010.
ACKNOWLEDGMENT [16] A. A. Khalek, C. Caramanis, and R. W. Heath, “Delay-
This work was supported by an NSERC Discovery Grant constrained video transmission: Quality-driven resource allo-
cation and scheduling,” IEEE J. Sel. Top. Signal Process.,
and in part by the UBC Four-Year Fellowship.
vol. 9, no. 1, pp. 60–75, Feb. 2015.
[17] X. Song, F. Yang, J. Cheng, N. Al-Dhahir, and Z. Xu,
REFERENCES “Subcarrier phase-shift keying systems with phase errors
in lognormal turbulence channels,” J. Lightwave Technol.,
[1] T. Rakia, F. Gebali, H.-C. Yang, and M.-S. Alouini, “Cross layer vol. 33, no. 9, pp. 1896–1904, Feb. 2015.
analysis of P2MP hybrid RF/FSO network,” J. Opt. Commun. [18] M. Niu, J. Schlenker, J. Cheng, J. F. Holzman, and R. Schober,
Netw., vol. 9, no. 3, pp. 234–243, Mar. 2017. “Coherent wireless optical communications with predetection
[2] N. D. Chatzidiamantis, A. S. Lioumpas, G. K. Karagiannidis, and postdetection EGC over gamma–gamma atmospheric
and S. Arnon, “Adaptive subcarrier PSK intensity modulation turbulence channels,” J. Opt. Commun. Netw., vol. 3,
in free space optical systems,” IEEE Trans. Commun., vol. 59, no. 11, pp. 860–869, Nov. 2011.
no. 5, pp. 1368–1377, May 2011. [19] A. J. Goldsmith and S. Chua, “Variable-rate variable-power
[3] M. Z. Hassan, M. J. Hossain, and J. Cheng, “Performance of MQAM for fading channels,” IEEE Trans. Commun., vol. 45,
nonadaptive and adaptive subcarrier intensity modulations no. 10, pp. 1218–1230, Oct. 1997.
in gamma–gamma turbulence,” IEEE Trans. Commun., [20] J. Tang and X. Zhang, “QoS-driven power and rate adaptation
vol. 61, no. 27, pp. 2946–2957, July 2013. over wireless links,” IEEE Trans. Wireless Commun., vol. 6,
[4] I. B. Djordjevic, “Adaptive modulation and coding for free no. 8, pp. 3058–3068, Aug. 2007.
space optical channels,” J. Opt. Commun. Netw., vol. 2, [21] Y. Li, L. Liu, H. Li, J. Zhang, and Y. Yi, “Resource allocation
no. 5, pp. 221–229, May 2010. for delay-sensitive traffic over LTE-advanced relay net-
[5] O. Barsimantov and N. N. Nikulin, “Adaptive optimization works,” IEEE Trans. Wireless Commun., vol. 14, no. 8,
of a free space laser communication system under dynamic pp. 4291–4303, Aug. 2015.
link attenuation,” J. Opt. Commun. Netw., vol. 3, no. 3, [22] 3rd Generation Partnership Project Access, “Further ad-
pp. 215–222, Mar. 2011. vancements for E-UTRA physical layer aspects,” 3GPP TR
[6] M. Karimi and M. Uysal, “Novel adaptive transmission algo- 36.814, Tech. Rep., 2010.
rithms for free space optical links,” IEEE Trans. Commun., [23] A. Aijaz, M. Tshangini, M. R. Nakhai, X. Chu, and A.-H.
vol. 60, no. 12, pp. 3808–3815, Dec. 2012. Aghvami, “Energy-efficient uplink resource allocation in
[7] M. Czaputa, S. Hranilovic, S. S. Muhammad, and E. Leitgeb, LTE networks with M2M/H2H co-existence under statistical
“Free space optical links for latency tolerant traffics,” IET QoS guarantees,” IEEE Trans. Commun., vol. 62, no. 7,
Commun., vol. 6, no. 5, pp. 507–513, June 2012. pp. 2353–2365, July 2014.
[8] V. V. Mai and A. T. Pham, “Cross-layer analysis of adaptive [24] X. Mi, L. Xiao, M. Zhao, X. Xu, and J. Wang, “Statistical
rate transmission and ARQ for free space optical communica- QoS-driven resource allocation and source adaptation for
tions,” IEEE Photonics J., vol. 8, no. 1, pp. 1–15, Feb. 2016. D2D communications underlying OFDMA-based cellular
networks,” IEEE Access, vol. 5, pp. 3981–3999, Apr. 2017.
[9] V. Jamali, D. S. Michalopouls, M. Uysal, and R. Schober,
“Link allocation for multiuser systems with hybrid RF/FSO [25] C. S. Chang, “Stability, queue length, and delay of determin-
backhaul: Delay-limited and delay-tolerant designs,” IEEE istic and stochastic queueing networks,” IEEE Trans. Autom.
Trans. Wireless Commun., vol. 15, no. 5, pp. 3281–3295, Control, vol. 39, no. 5, pp. 913–931, May 1994.
May 2016. [26] Q. Du and X. Zhang, “QoS-aware base-station selections for
[10] M. Z. Hassan, M. J. Hossain, J. Cheng, and V. C. M. Leung, distributed MIMO links in broadband wireless networks,”
“Effective capacity of coherent POLMUX OWC impaired by IEEE J. Sel. Areas Commun., vol. 29, no. 6, pp. 1123–1138,
atmospheric turbulence and pointing errors,” J. Lightwave June 2011.
Technol., vol. 34, no. 21, pp. 5007–5022, Nov. 2016. [27] D. Wu and R. Negi, “Effective capacity: A wireless link model
[11] N. Cvijetic, D. Y. Qian, J. J. Yu, Y. K. Huang, and T. Wang, for support of quality of service,” IEEE Trans. Wireless
“Polarization-multiplexed optical wireless transmission with Commun., vol. 2, no. 4, pp. 630–643, July 2003.
coherent detection,” J. Lightwave Technol., vol. 28, no. 8, [28] M. C. Gursoy, “MIMO wireless communications under statis-
pp. 1218–1227, Mar. 2010. tical queuing constraints,” IEEE Trans. Inf. Theory, vol. 57,
[12] A. García-Zambrana, B. Castillo-Vázquez, and C. Castillo- no. 9, pp. 5897–5917, Sept. 2011.
Vázquez, “Average capacity of FSO links with transmit laser [29] The Wolfram functions site [Online]. Available: http://
selection using non-uniform OOK signaling over exponential functions.wolfram.com.

Вам также может понравиться