Академический Документы
Профессиональный Документы
Культура Документы
The performance of lasercom systems operating in the atmosphere is reduced by optical turbulence, which
causes irradiance fluctuations in the received signal. The result is a randomly fading signal. Fade
statistics obtained from experimental data were compared to theoretical predictions based on the log-
normal and gamma gamma distributions. The probability of fade, the expected number of fades per
second, and the mean fade time were calculated from the irradiance fluctuations of a Gaussian beam wave
propagating through the atmosphere along a horizontal path, near ground, in the moderate-to-strong
optical turbulence regime. Irradiance data were collected simultaneously at three receiving apertures,
each with a different size. Atmospheric propagation parameters were inferred from the measurements
and were used in calculations for the theoretical distributions. It was found that fade predictions made
by the gamma gamma and lognormal distributions provide an upper and lower bound, respectively, for
the probability of fade and the number of fades per second for the irradiance data collected in the
moderate-to-strong fluctuation regime. What is believed to be a new integral expression for the expected
number of fades based on the gamma gamma distribution was developed. This new expression tracked
the gamma gamma distributed data more closely than the existing approximation and resulted in a
higher number of fades. 2007 Optical Society of America
OCIS codes: 010.1300, 010.1330, 290.5930.
propagation paths [9 12]. Most recent work involves IT
the theoretical comparisons of fade statistics based Pfa PrI IT pIIdI, (1)
on the LN and GG distributions [79,11]. Some stud-
0
ies contain comparisons between theoretical predic-
tions of fade statistics and experimental data [10,12]. where IT is the threshold level of the irradiance and
One challenge when comparing experimental data pII is the PDF of the irradiance fluctuations. Rather
with theory is to accurately characterize the propa- than expressing the probability of fade as a function
gation path. In this work, the path average values of of IT, the fade threshold parameter, FT, is introduced.
the structure parameter, Cn2, as well as the inner, l0, FT is defined as the number of decibels below the
and outer, L0, scales of turbulence, were inferred from mean irradiance [13]
the experimental measurements by using scintilla-
tion theory combined with a numerical algorithm [1].
I0, L
The inferred atmospheric parameters were used with FT 10 log10
IT dB, (2)
existing scintillation theory [1316] to calculate the
parameters of the LN and GG distributions.
Most previous work on fade statistics focuses on where I0, L is the mean value of the on-axis por-
the probability of fade. Only a few studies include the tion of the irradiance of a Gaussian beam wave. This
expected number of fades per second or the mean fade work involves normalized (on-axis) irradiance data,
time [3 6,9,11], in which predictions made with the I0, L 1.
GG distribution are based on an approximation of
the joint PDF of a GG distributed irradiance and its 1. Lognormal Distribution
time derivative [9,11]. In this paper, a new integral A lognormally distributed normalized irradiance
expression for the expected number of fades per sec- I 1 is described by [13]
ond, based on an exact integral form of the joint PDF,
was developed for the GG distribution. The purpose 1 2
A. Probability Density Function Models and Aperture where ln I2 is the log-irradiance variance, which is
Averaging the parameter of the LN PDF. An analytical expres-
sion for the probability of fade for a lognormally dis-
Several different modulations of two PDFs, or doubly tributed (on-axis) irradiance of a Gaussian beam
stochastic processes, have been proposed as the distri- wave is obtained by inserting the lognormal PDF, Eq.
bution function for the irradiance fluctuations associ- (3), into Eq. (1), which yields [13]
ated with a small (point) aperture receiver. The GG
distribution [1719] is supported by simulation data in
weak, moderate, and strong irradiance fluctuations for 1
point receivers [19]. In the moderate-to-strong turbu- 2 0.23FT
1 2 ln I
lence regime, some previous experimental studies in- Pfa 1 erf , (4)
volving larger, finite-sized, receiving apertures [20,21]
2 2ln I
indicate that receiver-aperture-averaged irradiance is
lognormal for sufficiently large apertures. A recent where erf(x) is the error function. The log-irradiance
study [2] with the same experimental data used in this
variance is related to the scintillation index, I2, by
paper, as well as simulation data, showed that the GG
the relation [13]
distribution provides a good fit to the irradiance fluc-
tuations collected by finite-sized apertures that are
significantly smaller than the coherence radius, 0, in ln I2 lnI2 1, (5)
the moderate-to-strong fluctuation regime. For aper-
tures larger than or equal to the coherence radius, the where ln is the natural logarithm. In weak turbu-
irradiance fluctuations are lognormally distributed. lence ln I2 is approximately equal to I2. In the
moderate-to-strong fluctuation regime it is important
B. Probability of Fade to use the relation in Eq. (5). Recently developed
The probability of fade serves as an estimate of how scintillation theory [1316], valid in all regimes of
likely the detector output is to drop below a pre- turbulence and where the scintillation index depends
1
where x and y are assumed to be gamma distributed nIT I pI,I IT, IdI, (11)
and statistically independent. This modulation re- 2
sults in the GG distribution [13,19]
where pI, I I, I is the joint PDF of the irradiance, I,
2 and its time derivative, I. For any irradiance distri-
pII I2K2I I 0,
I
(7) bution, the joint PDF can be written as
1
exp0.23FT
PrI IT
1 F2; 1, 1; exp0.23FT tIT . (13)
nIT
exp0.23FT
1 1. Lognormal Distribution
1 2
where 1F2a; c1, c2; x is a generalized hypergeometric lnI 2
function. The and parameters are identified as 1 2 ln I
pI,I I, I exp
the inverse of the large-scale scintillation, x2, and Iln I2 2ln I2
small-scale scintillation, y2, respectively [13,19],
1 I2
exp , I 0, (14)
1 1 2bI2 8b2I2
, (9)
x2 expln x2 1
where b is the standard deviation of the time deriv-
1 1 ative of a Gaussian random process. The first part of
, (10)
y2 expln y2 1 the expression is the lognormal PDF, Eq. (3), while
the remainder is the conditional PDF of I. It is a
where ln x2 and ln y2 are the large-scale and small- zero-mean Gaussian distribution with the variance
scale log-irradiance variances. Recently developed depending on the irradiance I. The expected number
scintillation theory [1316], is used to relate the pa- of fades below the threshold for a lognormally distrib-
1 2
2 0.23FT I xy, I xy xy,
2 ln I (19)
nFT 0 exp 2 , (15)
2ln I
where x and y are assumed to be gamma distributed
and normalized x y 1. Consequently, x and x,
where 0 is the quasi-frequency, defined as [13]
as well as y and y, follow the joint PDF of a gamma
random process and its time derivative [13],
0
b
1
ln I 2
Bln I 0
Bln I0
Hz , (16)
px,xx, x
1 32 x x28xb 2
x e e x
, x 0,
8bx2
where Bln I is the covariance function of the log- (20)
irradiance. Bln I is the second derivative of the co-
variance function with respect to . 1 32 y y28yb 2
py,yy, y y e e y
, y 0,
2. Approximation of the Expected Number of Fades 8by2
for a GammaGamma Distributed Irradiance (21)
Previous work involving the joint PDF of a GG dis-
tributed irradiance and its time derivative are based
where bx and by are the standard deviations of the
on an approximation of the conditional PDF for I. The time derivative of Gaussian random processes asso-
functional form of the conditional PDF for I is un- ciated with the large-scale and small-scale irradi-
known, but it reduces to a zero-mean Gaussian PDF ances, respectively. For simplification of the model,
if either or . Consequently, an approx- the approximation bx2 by2 b22 is made. The joint
imation to the joint PDF for a GG distributed nor- PDF of I and I can be formulated in different ways.
malized irradiance and its time derivative is given One approach is to take the 2D inverse Fourier trans-
by [13] form of the two dimensional characteristic function,
pI,I I, I pIIpI II
eiuIvIpI,II, IdIdI
2 I,Iu, v
I2K2I
I
Note that Eq. (17) is exact if either or . 1
x32ex y32eyeiuxy
The expected number of fades below the threshold 4b2
0 0
for a GG distributed normalized irradiance with a
Gaussian approximated conditional distribution for
ex
24b2x
ey
24b2y
its time derivative is obtained by inserting Eq. (17) eivyx eivxydydxdydx.
into Eq. (11), which gives [13]
(22)
20ln I2 1
nFT exp0.23FT 2 The last two integrals in Eq. (22) are solved by com-
pleting the square in the exponent with respect to x
and y, respectively. This yields
K2 exp0.23FT, (18)
1 x
where 0 is the quasi-frequency, defined in Eq. (16). I,I u, v x e y1
0 0
3. New Theory
expy iuxy b2v2x2y y2xdydx.
A. Integral Expression for the Expected Number of Fades (23)
for a GammaGamma Distributed Irradiance
The joint PDF of a GG distributed irradiance and its Taking the inverse Fourier transform of the charac-
time derivative is obtained by modulating the statis- teristics function gives
captured with a sampling rate of 10 kHz. Great cau-
1 x tion was taken to avoid clipping in the received
x e y1ey
irradiance data. Data were taken in 5 min intervals,
0 0 and the background radiation was measured before
1 and after each run. During the data collection, the
eiIvexpb2x2y y2xv2dv weather ranged from full sunshine to mostly sunny,
2
with air temperatures in the range 18 C27 C and
wind speeds between 0 and 7 ms (cross wind speed
1
eiIueixyudu dydx. (24) 05 ms). A more detailed description of the experi-
2 ment is found in Ref. 1.
was subtracted from each data point. The back-
1 x12 ground noise was estimated to be the mean value of
pI,I I, I I32
4b2 0
I x2 the background measurement taken before and after
each 5 min interval of collected data. The normalized
exp x
I
x
2
I2x
4b II x2
dx. (25)
irradiance was obtained by dividing each corrected
data point by the 5 min average of the corrected data.
The probability of fade below the threshold was cal-
culated by counting the number of normalized irra-
The result in Eq. (25) has been confirmed by calcu- diance data points below IT exp0.23 FT and
lating pI,I I, I with alternative methods such as the dividing that number by the total number of data
Jacobian of transformation or using conditional points in the time interval (3 106). The expected
PDFs. number of fades per second was obtained by calculat-
The expected number of fades below the threshold ing the number of times the normalized irradiance
for a GG distributed normalized irradiance is ob- fell below the threshold and dividing that number by
tained by inserting Eq. (25) into Eq. (11). Solving the the number of seconds in the time interval (300). The
integral over I and using the definition of the quasi- mean fade time of the experimental data was deter-
frequency, 0, in Eq. (16) yields mined by dividing the probability of fade at each
threshold level by the corresponding value of the ex-
pected number of fades per second.
ln I0 To compare the fade statistics obtained from the
nIT IT12
x32IT x2
experimental data with theory, the atmospheric con-
ditions along the propagation path are required. Opti-
0
IT cal measurements, in combination with a scintillation
exp x dx. (26) theory for a Gaussian beam wave developed by An-
x
drews et al. [13,16] were used to infer the refractive
nFT is obtained by inserting the relation IT index structure parameter, Cn2, the inner scale of tur-
exp0.23FT into Eq. (26). bulence, l0, and the outer scale of turbulence, L0,
along the propagation path [1]. The scintillation in-
4. Experimental Data dex was measured for the same 5 min of experimental
data used to calculate the fade statistics, at each
A. Experiment receiving aperture (1, 5, and 13 mm diameter). The
The experiment was conducted during February experimental scintillation index from the three aper-
711, 2005, at a laser range located near Adelaide, tures and theoretical scintillation expressions were
South Australia. The terrain was dry grassland. A cw used to create a minimization problem for the atmo-
laser operating at 1550 nm was launched horizon- spheric parameters. It was solved numerically with
tally into free space via a 45 mm diameter collimator, an algorithm called the downhill Simplex method [1].
mounted 2.1 m above ground. The transmitted power Two sets of 5 min intervals of experimental data,
was approximately 22 dBm 150 mW, and the beam collected at 8:47 a.m.8:52 a.m. and 2:16 p.m.2:21
divergence was measured to be 0.46 mrad (full angle). p.m. on 11 February 2005, are presented in this pa-
After propagating 1500 m along a horizontal path per. The inferred atmospheric parameters for the
through the atmosphere, the beam was detected by first data set were Cn2 6.5 1014, l0 6.1 mm, and
Table 1. Atmospheric Parameters Inferred From the Experimental Data and the PDF Parameters Used to Calculate the Lognormal and
GammaGamma Distributions When Compared to the Experimental Data
duce the theoretical curves in Fig. 2 are shown in erture agrees with the LN predictions, while the
Table 1. Different values of the quasi-frequency will same quantity for the 1 and 5 mm apertures shifts
shift the curves vertically, since the theoretical ex- toward the LN results. In this case, the GG distribu-
pressions of the number of fades are directly propor- tion overestimates the number of fades. For the 1 mm
tional to 0. aperture, the experimental number of fades is closest
For the moderate strength of turbulence, it is clear to the predictions made by the approximation of the
from Fig. 2 that the experimental number of fades per GG distribution, which lies between the GG and LN
second is closest to the newly developed GG integral results. The approximation of the joint PDF of a GG
expression for all three receiver aperture sizes. In distributed irradiance and its time derivative in Eq.
this case, the LN distribution underestimates the (18) is a mixture of the LN and GG joint PDFs since
number of fades. For the case of strong turbulence, the conditional PDF of the GG time derivative is
the experimental number of fades for the 13 mm ap- approximated by a zero-mean Gaussian distribution,
which is the functional form of the conditional PDF of used to predict the expected number of fades per
the LN time derivative. second.
The expected number of fades based on the integral Figure 3 shows the mean fade time of the experi-
expression of the GG joint PDF predicts a larger num- mental data (circles), the LN distribution (short
ber of fades than the approximation. The two theo- dashed curve), the approximation of the GG joint
retical curves have different shapes, and the integral PDF (long dashed curve), and the newly developed
expression tracks GG distributed data much closer integral expression for the GG joint PDF (solid curve).
compared with the approximation for moderate tur- The theoretical values of the mean fade time were
bulence strength. In this case the new integral ex- obtained by dividing Eq. (4) by Eq. (15) for the LN
pression makes a notable difference and should be distribution, and Eq. (8) by Eq. (18) or Eq. (26), re-
spectively, for the approximation and integral expres- result of dividing one small number by another small
sion for the GG joint PDF. Different values of the number, which can lead to unstable results. It follows
quasi-frequency than those shown in Table 1 would from Fig. 3 that the reduction in the mean fade time
result in vertical shifts of the theoretical curves. The obtained by increasing FT, is smaller in the strong
experimental data for the moderate case of turbu- turbulence than it is in the moderate case.
lence support the GG predictions for the mean fade The results for the expected number of fades agree
time, while the LN distribution overpredicts this with the results for the probability of fade for both
quantity. For the case of strong turbulence, the GG cases of turbulence strength. The shift from the GG
predictions are favored for the 1 and 5 mm receiver distribution to the LN distribution for the two differ-
apertures. For the 13 mm aperture, a shift toward ent turbulence conditions is in agreement with the
the LN predictions takes place for low values of the results from the PDF study [2] and is believed to be
threshold parameter, FT. The mean fade time is the attributable to the size of the coherence radius rela-