Published in IET Optoelectronics

Received on 3rd November 2010

Revised on 20th July 2011
doi: 10.1049/iet-opt.2010.0100
ISSN 1751-8768
Improved bit error rate evaluation for optically
pre-amplied free-space optical communication
systems in turbulent atmosphere
A.O. Aladeloba A.J. Phillips M.S. Woolfson
Division of Electrical Systems and Optics, Faculty of Engineering, University of Nottingham, University Park,
Nottingham NG7 2RD, UK
E-mail: andrew.phillips@nottingham.ac.uk
Abstract: The use of an optical pre-amplier at the receiver of free-space optical (FSO) communication systems necessitates the
consideration of both turbulence and amplied spontaneous emission (ASE) noise in performance modelling. Until now, this
problem has been typically approached by the use of a Gaussian approximation (GA) for the conditional probability of error
with averaging then performed using the received irradiance probability density function (PDF) (governed by scintillation).
However, the GA is decient as it only uses the rst two moments. Moment generating function (MGF) techniques, notably
the Chernoff bound (CB) and modied Chernoff bound (MCB), are used to evaluate the bit error rate performance of an
optically pre-amplied FSO system in the presence of both atmospheric turbulence and ASE noise. The MGF-based methods
incorporate a fuller statistical description of the signal and noise processes encountered in the optically pre-amplied case
when compared to the GA. The lognormal, gammagamma, K and negative exponential distributions have been used to
characterise the weak, moderate, strong and saturated turbulence regimes. The MCB gives the tightest bound upon the BER
compared to the CB, particularly at lower gains, and, as it also can be exceeded by the GA at higher gains, it is a logical
method to use in general.
1 Introduction
In recent years, there has been a renewed interest in
understanding and exploiting free-space optical (FSO)
communication, mainly because of the large potential
bandwidth available compared to radio frequency (RF) and
its exibility compared to optical bre. FSO communication
typically entails transmitting information-bearing near-
infrared light through the air between two or more
transceivers. FSO has emerged as a commercially viable
complementary technology to RF and millimetre-wave
systems [13]. Compared to the conventional RF systems,
FSO systems have low mass, volume and power
requirement, no frequency license requirements and
narrower beam size [14], while their relatively low overall
cost and ease of deployment are the advantages FSO
systems offer over the optical bre systems which have
additional costs from optical bre cable, right of way and
trenching. Some of the applications of FSO
communications include last mile access, backup link for
optical bre, cellular communication backhaul, high-
denition television, difcult terrains, multi-campus
communication networks and ad hoc networks [3, 5, 6].
Despite their attractions, FSO systems are faced with a
number of challenges such as beam attenuation (scattering
particularly from fog, and absorption) and turbulence
(which is the major impairment considered in this paper)
[14, 7]. Turbulence in FSO links arises as a result of
variation in temperature and pressure of the atmosphere
which leads to refractive-index uctuations. This
phenomenon, commonly known as scintillation, causes
amplitude, phase and intensity uctuation of a laser beam
as it propagates along the transmission path and
signicantly affects the performance of the link for
distances greater than 1 km [2, 7].
The application of optical pre-amplication to overcome
the impact of receiver thermal noise is one way of
improving the receiver sensitivity of FSO systems. Aside
from the optical gain, the optical pre-amplier also
generates amplied spontaneous emission (ASE) noise
which, in turn, generates additional beat noises
(spontaneous-spontaneous and signal-spontaneous) in the
electrical domain at the receiver. The overall electrical
domain noise is non-Gaussian, although it has often been
approximated as Gaussian in probability density functions
(PDFs) used for describing binary signals dominated by
ASE noise [8, 9]. The inadequacy of this approach (despite
a fortuitous near cancelling of erroneous Gaussian tails that
gives the approach some credibility [10]) led to more
advanced techniques being developed in bre system
performance calculations [11, 12]. However when FSO
reception in the presence of ASE and turbulence has been
considered so far it has typically been with the Gaussian
approximation (GA) assumed for the conditional error
26 IET Optoelectron., 2012, Vol. 6, Iss. 1, pp. 2633
& The Institution of Engineering and Technology 2011 doi: 10.1049/iet-opt.2010.0100
www.ietdl.org
probability for a given irradiance [13, 14]. A non-central chi
square PDF approach has also recently been used [15].
The moment generating function (MGF) represents a
convenient statistical way of describing the signal plus ASE
noise in a system employing an optically pre-amplied
receiver while Chernoff bound (CB) and modied Chernoff
bound (MCB) are techniques that use this description to
obtain upper bounds upon the bit error rate (BER) [12, 16].
Like the PDF, the MGF can provide a full statistical
representation of the signal plus noise [12]. This paper
presents an MGF-based approach for modelling the
performance of an onoff keying (OOK) intensity-
modulated turbulent FSO system with an optically pre-
amplifed receiver. Unlike in OOK-based optical bre
systems where the decision threshold is steady, the OOK-
based FSO systems threshold is dependent on the
instantaneous irradiance. The decision circuit at the receiver
thus uses an adaptive threshold because of the near-optimal
performance that is achievable. This, however, implies that
the threshold level varies in sympathy with the uctuating
average incident optical signal though it should be noted
that turbulence uctuations are much slower (1 kHz [14])
than the bit rates to be used. The lognormal (LN), gamma
gamma (GG), K (KD) and negative exponential (NE)
distributions are employed in this work and the results
obtained are compared with the customary GA approach.
2 Receiver system
Fig. 1 shows the schematic diagram of an optically pre-
amplied FSO receiver. A laser source with operating
wavelength l of 1.55 mm is assumed in this work. The
optical beam spreads out as it approaches the receiver with
beam pattern characterised by its planar angular cross
section u l/D
TX
[2, 4] (assuming a diffraction-limited
optical system), where D
TX
is the transmitter aperture
diameter. Non-return-to-zero (NRZ) OOK modulation is
assumed. At the receiver, the receiver collecting lens (RCL)
is assumed to be perfectly aligned with the transmitter lens
in a pointing and tracking scheme. It collects the incident
laser beam which is coupled to a bre using a collimator
(following [17]) and then optically amplied. The process
of optical pre-amplication produces ASE noise whose eld
is statistically described as Gaussian. An optical band-pass
lter (OBPF) is placed after the pre-amplier to
signicantly reduce the ASE noise in the incident optical
signal. Another OBPF (not shown) can be placed before the
pre-amplier to reduce the ambient light. The use of an
additional OBPF before the optical amplier is neglected in
the current paper because the ambient light (that
accompanies the optical signal) even after optical
amplication is typically small compared to the ASE and
can be ltered by the OBPF after the optical amplier
which must be retained to control ASE-ASE beat noise. A
pi n photodiode with quantum efciency h is placed
after the optical amplier to convert the information-bearing
light into electrical signal. An avalanche photodiode could
also be used which is advantageous compared to a pi n
photodiode with no optical pre-amplier but of only limited
value with the optical pre-amplier. This electrical signal is
then electrically pre-amplied and ltered before being
passed to the decision device where the threshold is
applied. The process of photodetection can be described as
a square law detection in which the signal beats with ASE
noise, causing signal-spontaneous beat noise and also the
ASE beats with itself causing spontaneousspontaneous
beat noise. Typically these beat noises, and particularly the
signal-spontaneous beat noise, mean that the receiver is no
longer dominated by receiver thermal noise. As stated
earlier the threshold is assumed to adapt to the
instantaneous irradiance at the receiver. Consequently, an
optimal threshold for each instantaneous irradiance level is
assumed. This can be realistically approached for example
in the Kalman lter-based method of [18].
3 Atmospheric channel models
In clear air conditions, optical beam propagation through the
atmosphere is particularly affected by turbulence-induced
scintillation which signicantly reduces the performance of
the link [1, 4, 7]. The major consequence of scintillation is
uctuations in the irradiance at the receiver which results in
high BERs [1]. Several mathematical models have been
proposed to characterise different turbulence regimes using
PDFs for the randomly varying irradiance [1, 3, 4, 7, 14].
The modied Rician, LN distribution and more recently the
GG distribution are the commonly reported models for
characterising the weak turbulence regime, although the
modied Rician PDF which does not agree with
experimental measurements is less popular [1, 3]. In the
strong turbulence regimes, several models such as the K,
GG and NE distributions have been proposed with the latter
best suited for saturated regimes [1, 4]. However, it should
be noted that the GG distribution gives better t to weak
irradiance uctuation measured data when compared with
the LN distribution [1, 3, 19, 20] while both K and GG
distributions show excellent ts with strong turbulence
measured data [1, 21]. Table 1 shows the PDFs of all
atmospheric turbulence models considered in this paper. I is
the instantaneous irradiance and is greater than zero for all
PDFs, and kIl is the average received irradiance. The a and
b parameters will be dened in Table 1. The Rytov
variance s
2
1
is a parameter commonly used to classify weak
(s
2
1
, 1), moderate (s
2
1
1), strong (s
2
1
. 1), and saturated
Fig. 1 Optically pre-amplied FSO receiver
IET Optoelectron., 2012, Vol. 6, Iss. 1, pp. 2633 27
doi: 10.1049/iet-opt.2010.0100 & The Institution of Engineering and Technology 2011
www.ietdl.org
(s
2
1
1) turbulent optical links. It is given by [1, 3]
s
2
1
= 1.23C
2
n
k
7/6
l
11/6
(1)
where C
2
n
is the refractive-index structure constant (whose
value is typically within range 10
217
m
22/3
C
2
n

10
213
m
22/3
[1, 3]) is taken to be constant for horizontal
path communication link and modelled as a function of
altitude for uplinks and downlinks [1, 3]. Here k 2p/l
and l represent the optical wavenumber and length of the
optical link, respectively. Typical values of s
2
1
, a and b
used for modelling weak, moderate, strong and saturated
turbulence regimes are shown in Table 2. s
2
I
, used for LN
distribution, is the variance of the natural logarithm
irradiance (normalised to its mean).
In the LN distribution model, while mathematically simple,
the PDF peak and tail values do not correspond to
experimental data [1, 3]. This implies that accuracy of
statistical analysis such as detection and fade probabilities
arrived at using this model will be signicantly affected.
Another shortcoming of LN distribution is that the Rytov
variance on which it depends will not be able to account
for multiple scattering caused by turbulence eddies as the
optical link length increases [1, 3]. The GG distribution was
developed for atmospheric turbulence modelling by
Andrews and Phillips [1] although the PDF was highlighted
in related works of Nakagami [22], Lewinski [23] and of
Teich and Diament [24]. This model takes into
consideration both the large-scale and small-scale effects on
optical beam traversing a turbulent atmospheric channel.
This implies that the GG distribution model is valid for
both weak and strong turbulence regimes. The GG PDF
thus depends on the effective number of large-scale eddies
of the scattering process a and the effective number of
small-scale eddies of the scattering process b. In weak
turbulence conditions, a and b 1. In strong turbulence
conditions, a and b decrease substantially such that in the
saturation regime, b 1 as GG approaches NE [1] though
a will increase again as saturation deepens. Note that the
GG distribution reduces to the K distribution when b is
unity, while the K distribution tends to the negative
exponential distribution as a 1. The KD is a widely
accepted model for characterising the strong turbulence
regime, although it was originally proposed to model non-
Rayleigh sea echo [1, 21]. The KD scintillation index
S.I. 1 +2/a always exceeds unity and has strong
agreement with experimental data in similar conditions. In
very strong irradiance uctuations, that is, saturation
turbulence regime, where the number of independent
scatterings becomes large and the optical link length spans
several kilometres, the value of the scintillation index tends
to unity (from above). The optical eld traversing this
medium is experimentally veried to follow the negative
exponential statistics for irradiance [1, 4].
4 Moment generating function for
optical signal
For a given irradiance I the optical power at the optical
amplier input during OOK NRZ bit of data j [ {0, 1} is
expressed mathematically as
P
j
(I) = a
j
IA (2)
where a
1
2r/(r +1), a
0
2/(r +1), A is the area of the
receiver aperture and r is the extinction ratio (typically 10).
Clearly I is the mean irradiance for the bit stream at a
particular time.
The MGF (conditional on I ) can then be obtained from, for
example [12], under the assumption of an integrating
response over bit period T (which has a noise equivalent
Table 1 Commonly used atmospheric turbulence models
Distribution type Probability density function Notes
Lognormal [1, 3, 6]
p
LN
(I, kIl) =
1
Is
I
....
2p
exp
ln(
I
kIl
) +
1
2
s
2
I
_ _
2s
2
I
2

For plane wave propagation

s
2
I
= exp
0.49s
2
1
(1 +1.11s
12/3
1
)
7/6
+
0.51s
2
1
(1 +0.69s
17/5
1
)
5/6
_ _
1
Gammagamma
[1, 3, 5, 2024]
p
GG
(I, kIl) =
2(ab)
(a+b)/2
G(a)G(b)
I
kIl
_ _
a +b
2
_ _
1
K
ab
2
..........
ab
I
kIl
_ _
_ _ _
For plane wave propagation
a = exp
0.49s
2
1
(1 +1.11s
12/5
1
)
7/6
_ _
1
_ _
1
K [1, 20, 24]
p
KD
(I, kIl) =
2a
G(a)
a
I
kIl
_ _
(a1)/2
K
a1
2
.....
a
I
kIl
_ _ _
b = exp
0.51s
2
1
(1 +0.69s
12/5
1
)
5/6
_ _
1
_ _
1
Negative exponential [1, 4] p
NE
(I, kIl) =
1
kIl
exp
I
kIl
_ _
Table 2 Typical parameters for characterising weak-to-
saturated turbulence regimes [1, 25]
Parameter Turbulence regimes
Weak
(s
2
1
, 1)
Moderate
(s
2
1
1)
Strong
(s
2
1
. 1)
Saturated
(s
2
1
1)
s
2
1
0.2 1.6 3.5 25
a 11.651 4.027 4.226 8.048
b 10.122 1.911 1.362 1.032
28 IET Optoelectron., 2012, Vol. 6, Iss. 1, pp. 2633
& The Institution of Engineering and Technology 2011 doi: 10.1049/iet-opt.2010.0100
www.ietdl.org
bandwidth B
e
1/2T)
M
Y
j
(s/I ) =
exp[R

GsqP
j
(I)/(1 (R

N
0
sq/T))]
[1 (R

N
0
sq/T)]
L
(3)
where q is the electron charge, s is the standard parameter in
the transform domain for the MGF, L B
0
m
t
T is the product
of spatial and temporal modes, B
0
is the OBPF bandwidth in
Hz, m
t
2 is number of polarisation modes, N
0

n
sp
(G21)hf is the ASE power spectral density (PSD) in
W/Hz (in single polarisation), n
sp
is the spontaneous
emission factor, R

h/hf, G is the optical amplier gain, h

is the Plancks constant and f is the optical frequency in Hz.
On introducing the Gaussian receiver thermal noise, a new
overall conditional MGF for the signal at the decision device
is obtained [12]
M
Z
j
(s|I) = M
th
(s)M
Y
j
(s|I) (4)
where M
th
(s) = exp(s
2
th
s
2
/2) is the thermal noise MGF and
s
2
th
is the thermal noise variance at the decision circuit. The
conditioning of the MGF on I will be removed in the BER
calculation in the next section.
5 Bit error rate analysis
In this section, the application of MGF methods, specically
the CB and MCB for the BER evaluation is presented for
weak-to-strong turbulence regimes, using the LN, GG, KD
and NE atmospheric turbulence models.
The BER for a given irradiance I is
BER(I ) =
1
2
[P(1|0, I) +P(0|1, I )] (5)
where P(1|0, I) represents the probability of receiving a 1
given that 0 was transmitted and P(0|1, I) represents the
probability of receiving a 0 given that 1 was transmitted.
On applying the CB separately to each conditional
probability, (6) and (7) are obtained
P(1|0, I ) =P(i
0
(I) .i
D
(I)) exp(s
0
i
D
(I ))M
Z
0
(s
0
|I), s
0
.0
(6)
P(0|1, I) =P(i
1
(I) ,i
D
(I )) exp(s
1
i
D
(I))M
Z
1
(s
1
(I )), s
1
.0
(7)
where i
D
(I ) is the decision threshold.
The CB therefore gives the upper bound on the BER as
BER
CB
(I ) =
1
2
[ exp ( si
D
(I))M
Z
0
(s|I )
+exp (si
D
(I))M
Z
1
( s|I )] s = s
0
= s
1
. 0
(8)
where M
Z
0
(s|I) and M
Z
1
(s|I ) are given by (4). The setting of
s
0
s
1
s is a computational convenience that incurs a
very small accuracy penalty (as s
0
and s
1
can of course be
optimised separately) [12].
The MCB involves a similar approach to the CB except that
it typically provides a tighter upper bound on the BER in non-
turbulent systems [12, 16]. The MCB BER is thus given by [12]
BER
MCB
(I)
=
1
2
exp(si
D
(I))M
Z
0
(s|I )
ss
th
....
2p
+
exp(si
D
(I ))M
Z
0
(s|I)
ss
th
....
2p

_ _
s .0
(9)
The optimum threshold for the CB is obtained by differentiating
BER
CB
(I ) with respect to i
D
and setting the result to zero. The
resultant optimum decision threshold for a particular I is given
by [12]
i
Dopt
CB
(I) =
ln(M
Y
1
(s|I)/M
Y
0
(s|I ))
2s
(10)
The same value of i
Dopt
CB
(I) is obtained for the MCB.
On substituting (10) into (8) and (9), the bounds upon the
BER with optimal threshold can be written as [12]
BER
CB
(I ) = M
th
(s)
....................
M
Y
1
(s|I)M
Y
0
(s|I )
_
, s . 0 (11)
BER
MCB
(I) =
M
th
(s)
ss
th
....
2p

....................
M
Y
1
(s|I )M
Y
0
(s|I )
_
, s . 0
(12)
For the GA, in which at sampling instant, the noise
experienced by a 0 or 1 is zero mean Gaussian with
variance s
2
d0
or s
2
d1
and the mean signal level is i
0
(I ) or
i
1
(I ), the BER is given by [8, 9, 12, 14]
BER
GA
(I) =
1
2
erfc
Q(I)
..
2

_ _
(13)
Q(I ) =
i
1
(I) i
0
(I)
s
d1
(I) +s
d0
(I )
(14)
where s
2
dj
(I) = s
2
ssp,j
(I) +s
2
spsp
+s
2
th
represents the total
noise variance for j [ {0, 1} noise components.
s
2
ssp,j
(I) = 4R
2
q
2
GP
j
(I )N
0
B
e
is the signal-spontaneous
emission beat noise variance and s
2
spsp
= 2m
t
q
2
R
2
N
2
0
B
0
B
e
is the spontaneous-spontaneous emission beat noise
variance. As shot noise is not included in the MGF (though
it can be adapted to do so) it is also neglected here.
In the case of the CB (11) and MCB (12) the tightest bound
is obtained by nding the optimum value for s for each
irradiance I. For adaptive threshold OOK, facilitated by the
slow fading of the irradiance, where it is assumed that the
appropriate optimum threshold can be realised, the overall
BER is given as
BER
X,Y
(kIl) =
_
1
0
BER
X
(I)p
Y
(I, kIl) dI (15)
where BER
X
(I ) represents the BERs shown in (11), (12) and
(13) (X CB, MCB, GA with CB and MCB understood to
refer to the optimum s (which varies with I )) while p
Y
(I, kIl)
represents the atmospheric turbulence models shown in
Table 1 (Y LN, GG, KD or NE).
IET Optoelectron., 2012, Vol. 6, Iss. 1, pp. 2633 29
doi: 10.1049/iet-opt.2010.0100 & The Institution of Engineering and Technology 2011
www.ietdl.org
6 Results and discussion
In Table 3, the system parameter values used in this section
are listed. For simplicity coupling losses are neglected. To
obtain the results for more realistic coupling between air
and optical amplier input it is simply necessary to shift the
curves by the appropriate number of dB. The typical
coupling loss value is about 810 dB [13], although the
optical amplication process (when G . 10 dB) helps to
compensate for these losses. The receiver thermal noise is
7 10
27
A
2
, chosen to give receiver sensitivity (with no
turbulence or ASE) of 223 dBm at a BER of 10
212
. The
choice of RCL diameter of 4 mm approximately gives a
point receiver (so aperture averaging is neglected) because
it is less than the spatial coherence width
r
0
= (1.46C
2
n
k
2
l)
3/5
[1] at the receiver for typical link
lengths and C
2
n
values. For example, the calculated
minimum (for C
2
n
= 10
13
m
2/3
) and maximum (for
C
2
n
= 10
17
m
2/3
) values for r
0
at a typical optical link
length, l 1000 m are 0.0094 and 2.36 m, respectively.
The a and b values used in this analysis (where required)
were adopted from [1, 25] (see Table 2) as they have the
closest t to measured turbulence data (Rytov variance s
2
1
).
Fig. 2 shows the BER curves for CB, MCB, and GA at low-
gain optical amplier (G 8.8 dB), using the parameters in
Table 2 to model weak, moderate, strong and saturated
turbulence regimes. The CB is clearly seen to exceed both
the MCB and GA which give relatively similar BERs for all
employed atmospheric turbulence models. The BER curves
obtained for LN distribution differ from the GG distribution
by about 3 dB at target BER of 10
212
as shown in Fig. 2a.
The discrepancy is well known in non-amplied systems
[1, 3]. The discrepancy is clearly lower at worse BERs and
so the LN approach (which is easier to calculate) can be
more appropriate when forward error correction (FEC) is
available. The NE distribution is mainly used for
characterising fully saturated turbulence condition whereas
Table 3 Parameter values used for the numerical results
Parameter Description Value
R
b
bit rate 2.5 Gb/s [14]
l optical wavelength 1550 nm [14]
B
0
optical-lter
bandwidth
76 GHz
G optical amplier gains 30.6 and 8.8 dB [12, 14]
n
sp
spontaneous emission
parameter
1.5 (equivalent to noise
gure of 4.77 dB)
h quantum efciency 1 [14]
r extinction ratio 10 dB
D
RX
RCL diameter 4 mm [17]
Fig. 2 BER against normalised average received irradiance at RCL input [dB] using G 8.8 dB and D
RX
4 mm for
a No turbulence, and weak turbulence using LN and GG distributions
b No turbulence, weak, moderate, strong, and saturated turbulences using GG distribution
c Strong turbulence using GG and K distributions
d Saturated turbulence using NE distribution
30 IET Optoelectron., 2012, Vol. 6, Iss. 1, pp. 2633
& The Institution of Engineering and Technology 2011 doi: 10.1049/iet-opt.2010.0100
www.ietdl.org
the GG and K distributions are more appropriate for
characterising strong to saturated turbulence regime as
a 1 (and as b 1 for GG distribution only). Therefore
in Fig. 2d the NE distribution results are given and these
would almost coincide with the saturation regime results for
the GG and K distributions using the appropriate
parameters from Table 2, and which are thus not shown.
Fig. 2c shows the discrepancy between the K and the GG
distributions for the strong turbulence regime, where it can
be seen that the KD curves are almost the same as the fully
saturated NE in Fig. 2d. The GG by contrast differs by
virtue of using a value of b that is not unity.
Fig. 3 shows the BER curves for the high-gain
(G 30.6 dB) case using the same parameters as before to
characterise the atmospheric turbulence regimes. Here the
CB and MCB BER curves are almost matching, while
the GA differs from both CB and MCB, even more as the
turbulence strength increases. The similarity of the CB and
MCB is owing to the dominance of the signal-dependent
noise. Consideration of both Figs. 2 and 3 indicates that the
MCB with GG distribution is probably the most sensible
approach for modelling optically pre-amplied FSO
receiver in all atmospheric turbulence regimes. This is
because the MCB gives a tighter bound than the CB
especially when the contribution of the thermal noise is
relatively high and because the GG distribution is
reasonable over a whole range of turbulence conditions. It
is also noteworthy that all strong and saturated theoretical
BER curves continue almost linearly for higher powers than
shown but these would of course ultimately overload the
receiver.
Although the GA falls below the bounds in these
calculations, the uncertainty regarding whether it is above
or below the real BER is well reported in related MGF
analyses [12, 16], arising ultimately because of its moment
deciency. This can also be illustrated by specic cases,
using some different parameters than previously as shown
in Fig. 4 which gives the BER curves for the high-gain
(G 30.6 dB) case using (a) r 1, B
0
76 GHz,
(b) r 1, B
0
20 GHz. It can be seen from Fig. 4 that the
GA can exceed the MCB (which is almost matching the
CB) for the no turbulence regime for both cases, while as
the turbulence increases to the weak regime, the GA can
still slightly exceed the MCB (and CB) for both the LN and
GG cases in Fig. 4b, while in Fig. 4a this is only true for
the LN case. Generally, in the high-gain cases severe
turbulence conditions will move the GA-MCB (and CB)
crossover points to better BER (below the range plotted). It
should be further noted that the choice of different receiver
lters, other than the integrating response used here, with
the possibility of inter-symbol interference and also the use
of RZ modulation (of various pulse shapes) will also impact
Fig. 3 BER against normalised average received irradiance at RCL input [dB] using G 30.6 dB and D
RX
4 mm for
a No turbulence, and weak turbulence using LN and GG distributions
b No turbulence, weak, moderate, strong and saturated turbulences using GG distribution
c Strong turbulence using GG and K distributions
d Saturated turbulence using NE distribution
IET Optoelectron., 2012, Vol. 6, Iss. 1, pp. 2633 31
doi: 10.1049/iet-opt.2010.0100 & The Institution of Engineering and Technology 2011
www.ietdl.org
on the relative merit of the GA and the CB/MCB, as, for
example, investigated in [12] for the non-turbulent case.
Returning to the original parameters of Table 3, power
penalty plots are shown in Fig. 5 for the following BER
levels: 10
23
, 10
26
, 10
29
and 10
212
. In this work, the
power penalty refers to the additional power needed in the
presence of impairments (turbulence and ASE noise) to
return the FSO system to the BER achievable without the
impairments. As might be expected, the penalty increases as
the turbulence strength (described by the Rytov variance)
increases and as the BER decreases. Beyond s
2
1
= 0.2, the
penalty increases very sharply until very high fading
conditions where the change in penalty gradually falls, for
instance, at high gain G 30.6 dB and BER of 10
26
using
MCB, when the Rytov variance rises from s
2
1
= 0.1 to
s
2
1
= 0.2, the penalty only rises from 3.5 to 5 dB. But when
it increases to s
2
1
= 1.6, the power penalty rises to 27 dB. It
should be noted that the power penalties shown in Fig. 5
are theoretical values. In practice, for example, BER of
10
212
will not be obtainable under high Rytov variance
conditions.
7 Conclusion
BER modelling for optically pre-amplied OOK FSO system
operating over atmospheric turbulence is investigated using
MGF-based techniques such as CB and MCB for the rst
time. The results obtained were compared with the GA
approach for both high and low gains using the main
candidate atmospheric turbulence models to characterise the
weak, moderate, strong and saturated turbulence regimes.
Overall, it can be seen that the MCB gives the tightest
bound upon the BER compared to the CB, particularly at
lower gains, and that it also can be exceeded by the GA at
higher gains, hence it is a logical method to use. The GG
distribution is further seen to be the most exible model for
characterising atmospheric turbulence across a whole range
of conditions.
8 References
1 Andrews, L.C., Phillips, R.L.: Laser beam propagation through random
media (SPIE Press, Bellingham, Washington, 2005, 2nd edn.)
Fig. 4 BER against normalised average received irradiance at RCL input [dB] for no turbulence and weak turbulence, using LN and GG
distributions G 30.6 dB, D
RX
4 mm
a r 1, B
0
76 GHz
b r 1, B
0
20 GHz
Fig. 5 Atmospheric turbulence induced power penalty against Rytov variance for OOK NRZ FSO modelled using GG distribution with
a Low-gain optical amplier (G 8.8 dB)
b High-gain optical amplier (G 30.6 dB)
Note that these are theoretical values and in practice BER of 10
212
will not be obtainable under high Rytov variance conditions
32 IET Optoelectron., 2012, Vol. 6, Iss. 1, pp. 2633
& The Institution of Engineering and Technology 2011 doi: 10.1049/iet-opt.2010.0100
www.ietdl.org
2 Caplan, D.O.: Laser communication transmitter and receiver design,
J. Opt. Fibre Commun. Rep., 2007, 4, pp. 225362
3 Majumdar, A.K.: Free-space laser communication performance in the
atmospheric channel, J. Opt. Fiber Commun. Res., 2005, 2,
pp. 345396
4 Karp, S., Gagliardi, R.M., Moran, S.E., Stotts, L.B.: Optical channels:
bers clouds water and the atmosphere (Plenum Press, New York,
1988)
5 Andrews, L.C., Phillips, R.L.: Free space optical communication link
and atmospheric effects: single aperture and arrays. Proc. SPIE in
Free-Space Laser Communication Technologies XVI, San Jose, CA,
vol. 5338, no. 16, pp. 265275
6 Kiasaleh, K.: Performance analysis of free-space on-off-keying optical
communication systems impaired by turbulence. Proc. SPIE in Free-
Space Laser Communication Technologies XIV, San Jose, CA,
January 2002, vol. 4635, pp. 150161
7 Zhu, X., Kahn, J.M.: Free-space optical communication through
atmospheric turbulence channels, IEEE Trans. Commun., 2002, 50,
(8), pp. 12931300
8 Olsson, N.A.: Lightwave systems with optical ampliers, J. Lightwave
Technol., 1989, 7, (7), pp. 10711082
9 Yamamoto, Y.: Noise and error rate performance of semiconductor
laser ampliers in PCM-IM optical transmission systems, IEEE
J. Quantum Electron., 1980, 16, (10), pp. 10731081
10 Dlubek, M.P., Phillips, A.J., Larkins, E.C.: Optical signal quality metric
based on statistical moments and Laguerre expansion, Opt. Quantum
Electron., 2008, 40, (8), pp. 561575
11 Chan, B., Conradi, J.: On the non-gaussian noise in erbium-doped ber
ampliers, J. Lightwave Technol., 1997, 15, (4), pp. 680687
12 Ribeiro, L.F.B., Da Rocha, J.R.F., Pinto, J.L.: Performance evaluation
of EDFA preamplied receivers taking into account intersymbol
interference, J. Lightwave Technol., 1995, 13, (2), pp. 225232
13 Cao, Q.L., Brandt-Pearce, M., Wilson, S.G.: Free space optical MIMO
system using PPM modulation and a single optical amplier. Second
Int. Conf. on Communications and Networking, Shanghai, China,
August 2007, vol. 1, pp. 11131117
14 Razavi, M., Shapiro, J.H.: Wireless optical communications via
diversity reception and optical preamplication, IEEE Trans. Wirel.
Commun., 2005, 4, (3), pp. 975983
15 Ulmer, T.G., Henion, S.R., Walther, F.G.: Power penalty from
amplied spontaneous emission in spatial diversity links for
fade mitigation, IEEE Photonics Technol. Lett., 2009, 21, (3),
pp. 170172
16 OReilly, J., Da Rocha, J.R.F.: Improved error probability evaluation
methods for direct detection optical communication systems, IEEE
Trans. Inf. Theory, 1987, 33, (6), pp. 839848
17 Abtahi, M., Lemieux, P., Mathlouthi, W., Rusch, L.A.: Suppression of
turbulence-induced scintillation in free-space optical communication
systems using saturated optical ampliers, J. Lightwave Technol.,
2006, 24, (12), pp. 49664973
18 Chen, C., Yang, H., Jiang, H., Fan, J., Han, C., Ding, Y.: Mitigation of
turbulence-induced scintillation noise in free-space optical
communication links using kalman lter. IEEE Congress on Image
and Signal Processing, Hainan, China, May 2008, vol. 5, pp. 470473
19 Akella, J., Yuksel, M., Kalyanaraman, S.: Multi-channel
communication in free-space optical networks for the last-mile. Proc.
15th IEEE Workshop on Local and Metropolitan Area Networks,
Princeton, NJ, June 2007, pp. 4348
20 Ghassemlooy, Z., Popoola, W.O., Leitgeb, E.: Free-space optical
communication using subcarrier modulation in gammagamma
atmospheric turbulence. Ninth Int. Conf. on Transparent Optical
Networks, Rome, Italy, July 2007, vol. 13, pp. 156160
21 Al-Habash, M.A., Andrews, L.C., Phillips, R.L.: Mathematical model
for the irradiance probability density function of a laser beam
propagating through turbulent media, Opt. Eng., 2001, 40, (8),
pp. 15541562
22 Nakagami, M.: The m-distribution a general formula of intensity
distribution of rapid fading, in Hoffman, W.C. (Ed.): Statistical
methods in radio wave propagation (Pergamon, New York, 1960),
pp. 336
23 Lewinski, D.J.: Nonstationary probabilistic target and clustter scattering
models, IEEE Trans. Antenna Propag., 1983, 31, (3), pp. 490498
24 Teich, M.C., Diament, P.: Multiple stochastic representations for K
distributions and their Poisson transform, J. Opt. Soc. Am. A, 1989,
6, (1), pp. 8091
25 Popoola, W.O., Ghassemlooy, Z.: BPSK Subcarrier intensity
modulated free-space optical communications in atmospheric
turbulence, J. Lightwave Technol., 2009, 27, (8), pp. 967973
IET Optoelectron., 2012, Vol. 6, Iss. 1, pp. 2633 33
doi: 10.1049/iet-opt.2010.0100 & The Institution of Engineering and Technology 2011
www.ietdl.org