Effect of Rain on Free Space Optics

C. Capsoni*, R. Nebuloni

Dipartimento di Elettronica e Informazione, Politecnico di Milano, Piazza L. da Vinci 32,

Abstract
Two years of experimental observations gathered by the 319-m optical link installed at the campus of
Politecnico di Milano, Italy, are discussed. Ancillary measurements of rain rate intensity collected by a tipping
bucket rain gauge and of visibility collected by a visibility sensor are also used. The visibility sensor and the
rain gauge are located nearby one of the two link terminals. A data reduction procedure has been carried out on
the time series of received laser power to separate the effects due to precipitation from those due to other
contributions, which are by non way negligible. Different drop size distributions (DSD) are taken into account
during the simulations of path attenuation being the effect of DSD much more evident at optical frequencies that
at microwaves.It is shown that rain attenuation can reduce system availability of links with small system margin
in areas where fog is absent.

1. Introduction
There is a growing interest in the telecommunication community towards the use of Free Space Optics
(FSO) for short and very short path applications requiring the availability of very large bandwidth. The
advantages of FSO over classical microwave links and fiber systems are low equipment cost, and flexible,
licence free, installation. FSO components are largely derived from the well-established fiber optic technology:
accordingly, the frequency bands used are basically the first and third optical windows. The most popular
applications of FSO are interconnection between buildings belonging to the same company/organization; links
between base stations in a dense broad band access network; backup of the telecommunication network during
emergency situations; last mile connections and so on.
Unfortunately, laser beam propagation in the atmosphere is severely affected by adverse weather which
limits the path length for a given link availability. Despite dense fog is the primary challenging factor for FSO
[1], propagation impairments due to atmospheric turbulence [2] and rain are not negligible. This paper is
focused on the attenuation caused by rain, that is frequently experienced in the spring-summer period in
temperate regions.

2. Rain attenuation
Rain is the most frequent type of atmospheric precipitation in temperate climate. It causes various
degradations at frequencies above 10 GHz where the wavelength becomes comparable to the size of raindrops.
At microwave frequencies, the effects of rain have been extensively studied and the microphysical aspects of
main concern are nowadays well-established [3]. In particular, to estimate the specific attenuation [dB/km]
(caused by rain of a given intensity R [mm/h]) the size distribution of raindrops (DSD), the terminal velocity v
[m/s] and the scattering cross section of each particle must be known. The raindrop radius r [mm] ranges from
tenths of mm to about 4 mm; larger drops are hydro-dynamically unstable and tend to break. The terminal
velocity (in stagnant air) of falling drops used in the computation of R has been here assumed:

v(r ) = 9.65 10.3e 1.2 r

[m/s]

(1)

Rain is constituted by a population of particles of different size. A generally accepted analytic form of
the DSD is the gamma distribution:
ELEVENTH URSI COMMISSION F OPEN SYMPOSIUM ON
RADIO WAVE PROPAGATION AND REMOTE SENSING
PUC-Rio, Rio de Janeiro, Brazil, 30 Oct / 02 Nov 2007
PR5.1 1

n ( r ) = N 0 r e r
4

N 0 = 6 10 e

[mm-1m-3]

(2)

[m-3mm-1- ]

(3)

( 3.2 ln 5 ) ln 5

= 0.2 R /(33.31N 0 5 +1 (4.67 + ))

1 /( 4.67 + )

[mm-1]

(4)

where is the shape parameter, which approximately ranges between -2 and 3 and is the gamma function.
Accordingly, R is given by:
+

R = 4.8 r 3 v(r )n(r )dr

[mm/h]

(5)

Note that is related to N0 by Eq.(3) so that Eq.(2) is, in fact, a two parameter function. If =0, (eq.2) takes the
form of the standard Marshall-Palmer DSD, a well established average distribution worldwide accepted for its
good statistical validity. Nevertheless, if the possible variations of the fade levels from event to event have to be
accounted for, the variability of the DSD must be considered.
The usual equation:

= 4.34 n(r ) sca (r )dr

[dB/km]

(6)

has been used to evaluate the attenuation experienced by an optical wave in the presence of rain. Table 1 reports
the coefficients a and b of the classical power law = aRb that relates the specific attenuation to rain intensity.
In Figures 1a) through c) we show theoretical curves of attenuation as a function of wavelength at three
precipitation rates (1 mm/h, 10 mm/h and 50 mm/h) using different DSD models. Wavelengths range from the
visible window (fractions of m) to the microwave region up to 33 mm (9 GHz). For sake of simplicity,
calculations were carried out assuming spherical raindrops despite the spheroidal shape would be a better
approximation. However here we are interested in addressing the dependence of attenuation on wavelength, rain
rate, and DSD rather than on polarization effects. Throughout the optical region (calculations at 0.3, 0.55,
0.785, 1.55, 3.7 and 10.6 m), the wavelength is much smaller than the size of particles, hence the
electromagnetic modelling is greatly simplified and the asymptotic formulas for large particles can be applied.
In the millimetre wave region, Mie theory should be applied (alternatively an approximation provided by the
anomalous diffraction theory could be used [4]), while Rayleigh scattering is responsible for the monotonic
decrease in attenuation at microwaves.
The optical attenuation is nearly independent of the wavelength moving from the visible region (0.3
m) to the far-infrared (10.6 m) but it exhibits significant variations with the shape of the DSD at low and high
rain rates. At 50 mm/h, for instance, attenuation may change from less than 12 dB/km to about 21 dB/km, with a
percent difference in excess of 70% (this difference is much higher than that at microwave frequencies). By
passing, consider that in our area, 50 mm/h rain intensity is exceeded, on the average, one hour per year
(exceedance probability 10-4), hence for high-availability applications rain should not be neglected.

3. Measurement Set-up
The experimental set up is constituted by the commercial optical link Terescope 3000 manufactured by
Optical Access (now MRV), formed by two identical transceivers able to transmit both data (up to 155Mbps)
and a single carrier at 0.785 m. Each transmitter uses three identical and independent semiconductor laser
diodes (nominal output power of 10 mW) assembled in triangular shape and the beam divergence is 2.5 mrad.
The receiver has a field of view of 2.8 mrad, a receiving area of 0.021 mm2 and is sampled every 1 s. The link is
installed within the campus Leonardo of the Politecnico di Milano in Milan (Italy) at about 20 m above ground,
and its length is 319 m. At one side there is a meteorological station (which provides 1-min time series of
temperature, relative humidity, pressure solar radiation and rain intensity), an ultrasonic anemometer and a
visibility sensor. The set up has been collecting data almost continuously since 2003. Here we present the
analyses carried out on two years of data, collected from April 2004 to March 2006. Row time series of laser
attenuation and visibility, collected at 1 sample per second, were subsequently averaged over 1 minute.

ELEVENTH URSI COMMISSION F OPEN SYMPOSIUM ON

4. Results
In order to analyze the impact of rain one needs to identify and separate the intervening contributions. In
some cases, in fact, laser attenuation before and after rain is negligible. In other cases, rain attenuation adds to
other contributions as shown in Figure 2b) where a constant 5 dB loss is measured even in the absence of rain.
To this aim, the following data reduction procedure was envisaged:
a) Two attenuation levels A1 and A2 are calculated by averaging laser attenuation over a convenient time
interval before and after rain occurrence.
b) The start time of each rain event t1 is found when laser attenuation exceeds a threshold A1+. In
order to avoid fluctuations of the attenuation curve above and below the threshold a two-threshold
method is used. A similar procedure is adopted to find the end time t2.
c) A straight line is drawn between the two points (t1, A1) and (t2; A2) and rain attenuation is found by
subtracting the quantity ((A2-A1)/(t2- t1))*( tk- t1)+ A1 from the laser attenuation at each sample tk where
rain is detected.
At this stage of the study we are not able to give a definite explanation of these attenuation contributions
not caused by rain, but they are not at all negligible and will require a specific investigation.
Figure 2 shows the application of the above data reduction procedure in the case of a shower occurred on 2
Oct 2005. In Figure 2b) are drawn the curves of laser attenuation, the contribution not due rain (labelled as
bias) and the estimated attenuation derived from the measured rain rate (Figure 2a) using a Marshall-Palmer
DSD model. Finally Figures 2c) and 2d) compare the attenuation due to rain with the estimate using time and
rain rate as x-axis, respectively. The agreement between data and estimation is satisfactory.
Repeating the above procedure over a number of events where the contribution of rain can be identified
with good accuracy, we get the scatterplot of Figure 3. To reduce the dispersion of data, to a great extent due to
the attenuation resolution of the laser receiver (we have to remind that we are using a commercial equipment),
the rain rate axis has been divided into a number of bins and the attenuation samples corresponding to rain rate
falling in the same bins have been averaged. The dotted straight line represents a best-fit of data in the
scatterplot. It turns out that the DSD which best fits data is more likely to be the Gamma DSD with =-2. We
point out that not all the rain events allowed to apply the data reduction procedure, hence, strictly speaking, we
cannot state that our conclusion has a statistical validity, but it sound quite realistic. Moreover, the predicted
attenuation is computed by assuming a constant rain rate, equal to the measured one, along the whole
propagation path (and this is not strictly true, of course). The above feature can be explained in the following
way. It has been shown that many small raindrops of 50 to 500 m diameter exist in common rainfall [5]. In
fact, at microwaves they are much smaller than the wavelength, hence Rayleigh scattering rules apply and the
effects on path attenuation would be negligible. On the contrary, at optical frequencies the contribution of such
particles wouldnt be negligible if their number concentration were high enough. Such small particles are to a
certain extent included in the gamma models with < 0 which are indeed the ones which more closely match
data, especially at low rain rates where the effect of small particles prevails..
Figure 4 shows the cumulative distribution function (CDF) of path attenuation normalized to a 1 km path in
the presence of rain. The curves were obtained by turning the CDF of rain rate at the measurement site into CDF
of path attenuation through the best fit coefficients in Table 1. The CDF corresponding to the coefficients
derived by the best fit of data (fig. 3) has been plotted too. The CDFs are conditioned to the probability of rain
occurrence which is about 4-5% of time. For instance, a 10 dB/km attenuation would be exceeded in slightly
more than 10% of the rainy time periods. However it should be pointed out that the above values are lower
bounds for attenuation when rain occurs in the optical region, as other contributions may significantly
overwhelm the one due to raindrops: for instance during autumn and winter in temperate regions, rain is often
accompanied by a significant reduction in visibility due to non precipitating water particles.

5. Conclusions
We have here presented the results of a 2 years experiment carried out with a FSO installed in the Milan
(Italy) area. In particular the paper is focused on the contribution to attenuation due to rain. The study points out
clearly that the role of the DSD is much more evident at optical frequencies than at microwaves and, indirectly,
evidence of the presence of an important amount of very small drops in the precipitation.

References
[1]

Andrews L.C., and R. L. Phillips, Laser beam propagation through random media, SPIE Optical
Engineering Press, 1998.
ELEVENTH URSI COMMISSION F OPEN SYMPOSIUM ON
RADIO WAVE PROPAGATION AND REMOTE SENSING
PUC-Rio, Rio de Janeiro, Brazil, 30 Oct / 02 Nov 2007
PR5.1 3

[2]

Deirmendjian D., Far-infrared and submillimeter wave attenuation by clouds and rain, J. Appl.
Meteorol., vol. 14, pp. 1584-1593, 1975.
[3]
COST Action 255 Final Report, Chapter 3.2: Precipitation, Clouds and other related non-refractive
effects, Editor J.P.V. Poiares Baptista, ESA Publication Division, The Netherlands, ISSN: 0379-6566, ISBN:
92-9092-608-2255
[4]
Evans B.T.N., and G.R. Fournier, "Simple approximation to extinction efficiency valid over all size
parameters", Appl. Opt., vol. 29, pp. 4666-4670, 1990
[5]
Ugai S., K. Kato, N. Nishijima, T. Kan, K. Tazaki, Fine structure of rainfall, Ann. Telecommunic.,
32, pp. 422-429, 1977

ELEVENTH URSI COMMISSION F OPEN SYMPOSIUM ON

(a) Precipitation rate: 1 mm/h

Attenuation (dB/km)

10

10

10

-1

10

-2

10
-1
10

Gamma (=-2)
Marshall-Palmer
Gamma (=3)
0

10

10

10
Wavelength (m)

10

10

10

(b) Precipitation rate: 10 mm/h

Attenuation (dB/km)

10

10

10

-1

10

-2

10
-1
10

Gamma (=-2)
Marshall-Palmer
Gamma (=3)
0

10

10

10
Wavelength (m)

10

10

10

(c) Precipitation rate: 50 mm/h

Attenuation (dB/km)

10

10

10

-1

10

-2

10
-1
10

Gamma (=-2)
Marshall-Palmer
Gamma (=3)
0

10

10

10
Wavelength (m)

10

10

10

Figure 1. Calculated attenuation due to rain as a function of wavelength at three precipitation rates. The
curves refer to three different DSDs, namely the Marshall-Palmer DSD, and two Gamma DSDs with
characteristic parameter =-2 and =3, respectively. Included are the visible range (0.3and 0.55 m), the
IR windows (0.785, 1.550, 3.7 and 10.6 m), the millimetre waves (1 to 10 mm, that is 30-300 GHz) and
the microwaves down to 33 mm (9 GHz).
ELEVENTH URSI COMMISSION F OPEN SYMPOSIUM ON
RADIO WAVE PROPAGATION AND REMOTE SENSING
PUC-Rio, Rio de Janeiro, Brazil, 30 Oct / 02 Nov 2007
PR5.1 5

(a)
Rain rate (mm/h)

80
70
60
50
40
30
20
10
0
17:45

18:05

18:25

18:45
Time

19:05

19:25

19:45

(b)
12
Laser
MP
Bias

Attenuation (dB)

10
8
6
4
2
0
17:45

18:05

18:25

18:45
Time

19:05

19:25

19:45

Rain attenuation (dB)

(c)
8
Laser
MP

6
4
2
0
17:45

18:05

18:25

18:45
Time

19:05

19:25

19:45

(d)
2

Attenuation (dB/km)

10

Data
MP
1

10

10
0
10

10
Rain rate (mm/h)

10

Figure 2. Shower detected on 2 Oct 2005. Rain attenuation is found by subtracting from the laser Table 1.
Coefficients a and b for the optical attenuation due to rain for different DSDs. The specific attenuation
expressed in dB/km can be approximated by the following power-law rule: a Rb.
ELEVENTH URSI COMMISSION F OPEN SYMPOSIUM ON
RADIO WAVE PROPAGATION AND REMOTE SENSING
PUC-Rio, Rio de Janeiro, Brazil, 30 Oct / 02 Nov 2007
PR5.1 6

Table 1. Coefficients a and b for the optical attenuation due to rain for different DSDs. The specific
attenuation expressed in dB/km can be approximated by the following power-law rule: a Rb.

Marshall/Palmer
Gamma (=-2)
Gamma (=-1)
Gamma (=0)
Gamma (=1)
Gamma (=2)
Gamma (=3)

a
1.45
2.83
1.66
1.31
1.15
1.06
1.00

B
0.64
0.37
0.54
0.64
0.71
0.75
0.78

Attenuation (dB/km)

10

Data
Best fit
MP
Gamma (=-2)
Gamma (=3)
1

10

10
0
10

10
Rain rate (mm/h)

10

Figure 3. Measured rain attenuation against rain rate (averaged scatterplot). The dotes straight line is the
best-fit of measurements. Three other straight lines are plotted which represent calculated attenuation
from the different DSD models (see Section 2)

Best fit
MP

10

Gamma (=-2)
Gamma (=3)

-1

CDF

10

-2

10

-3

10

10

15
20
Attenuation (dB/km)

25

30

35

Figure 4 Conditioned CDF of rain attenuation along a 1 km path as estimated through measured rain
attenuation (dotted curve) and three different DSD models.
ELEVENTH URSI COMMISSION F OPEN SYMPOSIUM ON
RADIO WAVE PROPAGATION AND REMOTE SENSING
PUC-Rio, Rio de Janeiro, Brazil, 30 Oct / 02 Nov 2007
PR5.1 7