Sei sulla pagina 1di 8

See

discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/281902385

Estimation of turbulence strength, anisotropy,


outer scale and spectral slope from an LED array

CONFERENCE PAPER · SEPTEMBER 2015


DOI: 10.1117/12.2191287

READS

5 AUTHORS, INCLUDING:

Christian Eisele Rui Barros


Fraunhofer Institute for Optronics, System T… Fraunhofer Institute for Optronics, System T…
11 PUBLICATIONS 69 CITATIONS 12 PUBLICATIONS 15 CITATIONS

SEE PROFILE SEE PROFILE

Available from: Rui Barros


Retrieved on: 05 October 2015
Invited Paper

Estimation of turbulence strength, anisotropy, outer scale and spectral


slope from an LED array
Szymon Gladysz, Max Segel, Christian Eisele, Rui Barros, Erik Sucher

Fraunhofer Institute of Optronics, System Technologies and Image Exploitation, Gutleuthausstr. 1,


76275 Ettlingen, Germany

ABSTRACT

A simple grid of 10×10 white-light LEDs allows for simultaneous measurement of several characteristics of atmospheric
turbulence. With this device, an imaging sensor and the model of tilt anisoplanatism one can determine turbulence
strength, anisotropy, outer scale and spectral slope of turbulence. We describe the theory and present preliminary results
obtained over a 270-m path.

Keywords: Turbulence, outer scale, tilt anisoplanatism, non-Kolmogorov turbulence

1. INTRODUCTION

Decorrelation of tip/tilt information is of fundamental importance in imaging and laser propagation through the
atmosphere without a cooperative beacon on the target. The simplest adaptive optics system corrects beam/image wander
through a fast-steering tip/tilt mirror. When the target or scientific object is bright enough one can use it as reference for
the tip/tilt sensor (usually a centroid detector). On the other hand, when the object does not provide a cooperative beacon
a neighboring light source must be used or created. Subsequently, a question arises: how close does the beacon have to
be to the object of interest in order to provide a reliable tip/tilt reference. To answer this question, the value of
differential tilt (or tip) for the separation between the target and the reference is needed. It is defined as differential
variance:

2
〈(𝜃(𝑟) − 𝜃(𝑟 + 𝑑)) 〉 (1)

where θ is the angle of arrival (tip/tilt), d is the separation between the two sources and . denotes ensemble average.
Beyond the practical reason given above, differential tilt is often measured/studied in experimental investigations of
optical turbulence. This is because measurement of differential image motion is not affected by any of the effects which
plague direct motion measurement: transmitter’s and receiver’s vibrations, drifts, thermal effects on optics, etc. This
realization has led to proliferation of differential image motion monitors (DIMMs) in astronomical site testing1-2. With a
single light source and two apertures one obtains two estimates of the path-integrated strength of optical turbulence,
usually quantified using the Fried’s parameter r01. Another approach is to use the Shack-Hartmann wavefront sensor
whereby one obtains a high signal-to-noise ratio estimate of r0 from many baselines corresponding to lenslet
separations3. Here, we propose an alternative: a single aperture observing an array of equidistant sources. Just like in the
case of the Shack-Hartmann sensor we benefit from a multitude of baselines over which r0 can be computed, thereby
increasing signal-to-noise ratio on the estimate. Absolute image motion scales as D-1/6, where D is diameter of the
telescope or pitch of the lenslet array. This means that with our geometry apparent image motion will be less than in the
case of the Shack-Hartmann sensor behind the same telescope. To illustrate this: a typical wavefront sensor with lenslets
of pitch D/10 will see almost 1.5 more motion, in rms terms, than the imaging camera mounted in the focal plane of the

Laser Communication and Propagation through the Atmosphere and Oceans IV, edited by Alexander M. J. van Eijk,
Christopher C. Davis, Stephen M. Hammel, Proc. of SPIE Vol. 9614, 961402
© 2015 SPIE · CCC code: 0277-786X/15/$18 · doi: 10.1117/12.2191287

Proc. of SPIE Vol. 9614 961402-1

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 09/23/2015 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


same telescope of diameter D. This disadvantage of the proposed geometry is more than compensated by the fact that the
two components of rms noise on the centroids, connected to photon and readout processes, will decrease by a factor of
100 in the new setup because the camera would see 100 times more photons than a lenslet2. This means that the multiple-
sources-single-aperture approach is more sensitive to turbulence than the single-source-many-apertures method.
Additionally, sampling of the tilt structure function (Equation (1)) can be much better controlled with an array of
transmitters because one has complete freedom where to position the sources and in which geometry. One can even
design an electronic system allowing for switching between arbitrary patterns/geometries (like we did). Another
advantage of observing with a full aperture is reduction of image scintillation due to aperture averaging. Finally, it is
worth noting that fabrication of an LED array is significantly cheaper than manufacturing of a lenslet array.
In this paper we describe the theory pertaining to tilt anisoplanatism measured over a horizontal path. An approach to
characterize optical turbulence beyond path-integrated refractive index structure constant 𝐶𝑛2 is elucidated. Finally, a
two-day experiment in Baldersheim, France, on the premises of the French-German Research Institute of Saint-Louis
(ISL) is described and a sample of results is shown.

2. THEORY

In this paper tilt is determined from the gradient (“G-tilt”) as opposed to the Zernike-tilt (“Z-tilt”). The second case,
which is more important to the design of adaptive optics systems, was solved by Sasiela4. We follow his approach to
solve for the G-tilt and the spherical-wave-propagation scenario. This is because centroid measurements are traditionally
believed to be closer to G-tilt than to Z-tilt. In our experiments we cannot assume plane-wave propagation because
propagation path is short.
2.1 Tilt anisoplanatism
The spatial-frequency filter functions to find gradient-tilt angle variance in x and y directions are:

𝐹𝑥 (𝜿) 4 2 2 cos2 (𝜑)


[ ]=( ) 𝐽1 (𝜅𝐷/2) [ 2 ]
𝐹𝑦 (𝜿) 𝑘0 𝐷 sin (𝜑)
(2)

where κ is the 2-D spatial frequency (of magnitude κ and angle φ), k0 is the wavenumber, D is diameter of the imaging
sensor, and J1(.) is Bessel function.
The expressions for tilt anisoplanatism in the direction parallel (𝜎∥2 ) and perpendicular (𝜎⊥2 ) to the sources’ separation are
then:

𝜎∥2 𝐿
2 (𝑧)
cos2 (𝜑) −11/3 4 2 2 𝑧𝜅𝐷
[ ] = 0.2073 ∫ 𝑑𝑧 𝐶𝑛 ∫ 𝑑𝜿 [ ]𝜅 ( ) 𝐽1 ( ) 2{1 − cos[𝜅𝑑cos(𝜑)]}
𝜎⊥2 0 sin2 (𝜑) 𝐷 2𝐿
(3)

where we have explicitly included the spherical-wave propagation factor z/L, with L being the total propagation distance.
The separation of the sources is given by the variable d. Kolmogorov turbulence model was assumed.
Using Euler's formula together with the following identity

𝜋
±𝑖𝑥cos𝜑 2𝜈
1 1 2 𝜈
∫ 𝑑𝜑 𝑒 sin 𝜑 = Γ (𝜈 + ) Γ ( ) ( ) 𝐽𝜈 (𝑥)
0 2 2 𝑥
(4)

Proc. of SPIE Vol. 9614 961402-2

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 09/23/2015 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


one can arrive at the result:
𝜎∥2 41.7 2 𝐿 𝐼𝑇 − 𝐼1
[ 2 ] = 2
𝐶𝑛 ∫ 𝑑𝑧 [ ]
𝜎⊥ 𝐷 0 𝐼1
(5)
with

𝑧𝜅𝐷 1 𝐽1 (𝜅𝑑)
𝐼1 = ∫ 𝑑𝜅 𝜅 −8/3 𝐽12 ( )( − )
0 2𝐿 2 𝜅𝑑


𝑧𝜅𝐷
𝐼𝑇 = ∫ 𝑑𝜅 𝜅 −8/3 𝐽12 ( ) (1 − 𝐽0 (𝜅𝑑))
0 2𝐿

where constant 𝐶𝑛2 was assumed along the propagation path.


2.2 Measurement of path-integrated turbulence strength
The reader should note that in Equation (5) the turbulence strength 𝐶𝑛2 is simply a factor scaling the result when one
assumes constant value along the path. This opens the possibility to estimate 𝐶𝑛2 by model-fitting. Given the
experimental parameters D and L, the double integral in Equation (5) is solved numerically, multiplied by all factors in
front of the integral except 𝐶𝑛2 and stored as a template for fitting. Estimation of turbulence strength consists then in
multiplying the template by a value of 𝐶𝑛2 in the range 10-16 – 10-11 m-2/3 and checking which of these multipliers yields
the smallest discrepancy, in the least-squares sense, from the measured tilt structure function (see Figure 2). In the fitting
process it should be taken into account that the estimation noise on the measured tilt structure function increases linearly
with the baseline d (for a quadratic LED array). This is because one has less and less pairs of LEDs to be used for
computation of the structure function with increasing d.
We note here that with an array of transmitters one obtains as many as four values of 𝐶𝑛2 in the end (from parallel and
perpendicular variances and from x and y directions). These can be averaged to produce one final value or treated
separately, as we do, to study turbulent anisotropy.
2.3 Measurement of outer scale
The discussion of possible measurements from an LED array was started with 𝐶𝑛2 because traditionally this is the metric
of most interest. However, it is recommended to estimate outer scale, L0, of the von Kármán model first (see Section
2.6). This is because the LED array offers a very simple way to measure it independently from 𝐶𝑛2 . First, re-write
Equation (3) for the von Kármán model, assuming constant 𝐶𝑛2 along the path:
𝜎∥2 6.6336 𝐶𝑛2 𝐿 cos 2 (𝜑) 𝑧𝜅𝐷
[ 2 ] = ∫ 𝑑𝑧 ∫ 𝑑𝜿 [ ] (𝜅 2 + 𝜅02 )−11/6 𝐽12 ( ) {1 − cos[𝜅𝑑cos(𝜑)]}
𝜎⊥ 𝐷 2
0 sin (𝜑)
2 2𝐿
(6)
2𝜋
where 𝜅0 = . Note that parallel and perpendicular variances have different functional forms (see also Figure 2; parallel
𝐿0
variances are always greater than perpendicular variances) and that 𝐶𝑛2 is outside the integral. The ratio of 𝜎∥2 and 𝜎⊥2
cancels out the 𝐶𝑛2 dependence and has only one unknown, that is L0. It can be safely assumed that the influence of inner
scale on image motion is minimal.
The three-dimensional integral in Equation (6) can be reduced to a double integral the same way as in Equations (3)-(5).
A number of templates, in our case 100 in the range of outer scales between 10 cm and 10 m, can then be computed and
stored. The ratios of measured differential variances in the directions parallel and perpendicular to the sources’
separation are then compared to the templates and the template minimizing the least-squares error yields the L0 estimate.
Finally, this outer scale value can be plugged back into Equation (6), or rather its two-dimensional version, and four
values of 𝐶𝑛2 can be recomputed using the more realistic von Kármán model. The effect will be a slight increase in the
measured value of 𝐶𝑛2 in comparison with the measurement with the Kolmogorov model (Equation (3)). The effect of
finite outer scale is dramatic on the absolute image motion but only mild on the differential motion 4.

Proc. of SPIE Vol. 9614 961402-3

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 09/23/2015 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


2.4 Measurement of power law of optical turbulence
In Equations (3) and (6) the Kolmogorov exponent of the power spectrum of the refractive index fluctuations, -11/3, was
used. On the other hand, one can also experiment with different power laws. Non-Kolmogorov turbulence is currently a
very active topic of research5-9. To measure the possibly non-Kolmogorov exponent, α, we follow the same approach as
in the case of outer scale estimation, namely we divide out the 𝐶𝑛2 dependence. This approach has been taken before in
the context of Shack-Hartmann wavefront sensing6. Equation (3) adapted to account for non-Kolmogorov turbulence
takes the form:

𝜎∥2 6.6336 𝐶̃𝑛2 𝐴(𝛼) 𝐿 cos2 (𝜑) −𝛼 2 𝑧𝜅𝐷


[ ] = ∫ 𝑑𝑧 ∫ 𝑑𝜿 [ ] 𝜅 𝐽1 ( ) {1 − cos[𝜅𝑑cos(𝜑)]}
𝜎⊥2 𝐷2 0 sin2 (𝜑) 2𝐿
(7)

where we replaced the refractive index structure constant 𝐶𝑛2 with its generalized counterpart 𝐶̃𝑛2 which has units m3-α.
Additionally, a normalizing constant 𝐴(𝛼) has been included but the details of its calculation are irrelevant as it gets
canceled out together with 𝐶̃𝑛2 when taking the ratio of 𝜎∥2 and 𝜎⊥2 .
The estimation of α proceeds like in Section 2.3: Equation (7) is reduced to a double integral the same way as in
Equations (3)-(5). Hundred templates of this function for α between 3 and 4 are then computed. The ratios 𝜎∥2 /𝜎⊥2 are
compared to the templates and the template minimizing the least-squares error yields estimate of α.
There is a serious caveat with this approach: we have assumed an infinite outer scale even though when carrying out the
computations described in Section 2.3 most probably finite, and in our case on the order of a few m – see Section 3,
outer scale had been already found. Unfortunately, non-Kolmogorov-von-Kármán model has two parameters and so
computation of 𝜎∥2 /𝜎⊥2 leaves us with one equation and two unknowns for every baseline. The problem is exacerbated in
that the effects of finite L0 and non-Kolmogorov exponent α on the ratio 𝜎∥2 /𝜎⊥2 are similar. The solution is to use non-
linear optimization scheme (e.g. Levenberg-Marquardt technique) or to iterate between fixed L0 and varying α and vice
versa, until convergence.
2.5 Measurement of turbulent anisotropy
The LED array allows for first-order observation of anisotropy. Spatial rescaling of turbulence spectrum in one direction
and anisotropy at different scales10 are phenomena which are too subtle to be seen in our experiments. On the other hand,
the setup yields one estimate of 𝐶𝑛2 in x and one estimate in y, and similarly for L0. We will see in Section 3 that
turbulence can, at least in terms of its average strength, be anisotropic.
2.6 Guidelines for data processing
When observing close to the ground, as we do, it is important to follow a certain succession of procedures. As mentioned
before, we recommend estimating L0 first. If there are circumstances implicating the existence of non-Kolmogorov
turbulence we recommend estimating L0 and α jointly. The value of L0 (and possibly α) can then be used to estimate
𝐶𝑛2 or its generalized counterpart 𝐶̃𝑛2 from Equation (5). The values of turbulence strength and outer scale corresponding
to horizontal and vertical directions will then hint at possible anisotropy in turbulence.

Proc. of SPIE Vol. 9614 961402-4

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 09/23/2015 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


3. EXPERIMENT
Over two days, 28th and 29th of July 2015, we performed experiments on the proving ground of French-German Research
Institute of Saint-Louis in Baldersheim, France. The experiments took place in the afternoon of 28.7 and morning of
29.7. Both days were quite hot, with temperatures near 30°C around midday.
Our LED array consists of 10×10 white-light LEDs separated by 5 cm (Figure 1, left). It was positioned 270 m from the
receiver (Figure 1, second picture from the left). Our receiver was a 145-mm diameter lens coupled with the 4.2MP
CMOS sensor (RAPTOR Osprey, 12-bit, 5.5μm pixels, TE-cooled) running at 100 Hz in subarray mode. The main
experimental difficulty consisted of choosing the optimal exposure time or alternatively, the LED current. One wants to
exploit the full 12-bit dynamic range but this proves to be difficult when turbulence strength changes: when 𝐶𝑛2
decreases more light is concentrated in the centers of the spots (Strehl ratio increases). This can lead to saturation of the
images. On the other hand, saturation or signal fade can happen with stronger turbulence because of increased
scintillation and exponential speckle statistics. After trial-and-error an exposure time of 1 ms was chosen. Thousand
frames per sequence were saved, every two minutes. The two rightmost panels of Figure 1 show sample frames from
sequences corresponding to stronger and weaker turbulence experienced during the field trial. Evidence of optic drift
probably due to temperature change is also visible.

Figure 1. From left to right: LED array, sensors used in the experiments (telescope trained on the LED array is the leftmost
device; next to it is the PSF/MTF measurement setup and the rightmost instrument is the receiver of the BLS 900
scintillometer), LED array seen during a period of stronger turbulence and during weaker turbulence.

As reference for the results from the LED array, a scintillometer (BLS 900 from Scintec) and a PSF/MTF-based
measurement setup were also employed during the trial. Both were looking at transmitters located at the distance of 477
m. The distance was chosen in order to provide sufficient scintillation signal for BLS 900. The latter setup was
developed at Fraunhofer IOSB as a portable means of estimating turbulence strength from Fried’s “short-exposure”
modulation transfer function (MTF)11. In brief: each speckle frame from a fast readout camera is Fourier-transformed on-
the-fly, and subsequently only the magnitude of the resulting quantity is kept and averaged over time. Finally, least-
squares fitting of the theoretical MTF model is employed to find Fried’s parameter r0 and consequently 𝐶𝑛2.

- BLS900 29July2015 (477m


3.0x10" 1f -12
- PSF 29 July 2015 (477m)
- LED Array 29 July 2015 (271m)
2.5x1011

2.0x1011

1.5x1011

1.0x10"

5.0x1012 parallel variance (measurement)


perpendicular variance (measurement)
- parallel variance (theory)
0.0
- perpendicular variance (theory)
1 E -15
10 20 30 40 50 10:00 11:00 12:00 13:00 14 00
LED separation in cm Time (29/07/2015)

Figure 2. Left: Illustration of the agreement between measured and predicted differential variances, a typical result of our
field trial. Right: Comparison of refractive index structure constant measurements with the three approaches discussed in the
paper.

Proc. of SPIE Vol. 9614 961402-5

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 09/23/2015 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


3.1 Comparison of path-averaged 𝑪𝟐𝒏 estimates
For each sequence, positions of LEDs were identified and their centroids computed using the thresholding approach2.
Subsequently, for each investigated baseline between 5 and 45 cm, the relevant pairs of centroids were subtracted from
each other and the difference squared, as in Equation (1). Averaging was then performed over all relevant squared
differences for each baseline. Pure Kolmogorov model, Equation 3, proved to be an excellent fit to all measurements.
This is demonstrated in Figure 2 (left) which shows typical result from the field trial in Baldersheim. Preliminary
estimates of 𝐶𝑛2 obtained directly from such fits, disregarding finite outer scale, are in very good agreement with the
scintillometer and the PSF/MTF-based measurements. This is shown in Figure 2 (right).
3.2 Outer scale estimates
As proposed in Section 2.3, we estimate L0 from the ratio 𝜎∥2 /𝜎⊥2 . We show here only a sample of results while the rest of
the data awaits processing. As can be seen in the left panel of Figure 3, most of the values of L0 in the chosen time
interval of 2.5 hours are below 3 m, and mean L0 equals 1.3 m which is absolutely within range of values predicted and
measured for the 1 m height by different authors.
3.3 Anisotropy
The right panel of Figure 3 shows vertical 𝐶𝑛2 vs. horizontal 𝐶𝑛2, again for the chosen period of 2.5 hours on the 29 th of
July 2015. It is clear that the former was higher than the latter. This effect must be related to convection processes which
are not negligible at the height of 1 m.

lE 13
horizontal outer scale
vertical outer scale
4

' , 1E-14

o
10 04 10 33 11 02 11 31 12 00 12 28 1E-14 lE 13
local time horizontal C2 in m -213

Figure 3. Left: Outer scale estimates found over a period of 2.5 hours. Right: comparison of vertical and horizontal
refractive index structure constants found in the same time interval.

4. CONCLUSIONS

We have shown the premise of differential motion as easy-to-implement, robust to vibrations, wavelength-independent
method of measuring optical turbulence. Specifically, an LED array gives an opportunity to measure differential motion
over many baselines (noise averaging) and is in principle less affected by noise than the Shack-Hartmann method.
Additionally, the ratio of parallel and perpendicular tilt anisoplanatism is, in principle, a very rich source of information
about turbulence. One can obtain estimates of outer scale, non-Kolmogorov spectral exponents and first-order
anisotropy information. The data analysis of the experiment in Baldersheim is ongoing and we plan to obtain complete
statistics of outer scale and spectral exponents soon.

Proc. of SPIE Vol. 9614 961402-6

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 09/23/2015 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


ACKNOWLEDGMENTS

This research is part of the project ATLIMIS (Atmospheric Limitations of Military Systems, No.
E/UR1M/9A265/AF170), commissioned and sponsored by the WTD91 (Technical Centre of Weapons and Ammunition)
of the German Armed Forces. The authors would like to thank ISL staff for organizing an experiment without which this
paper would not have been possible: Bernd Fischer, Dejan Simicic, Philippe Chaillet, Stephane Schertzer and Emmanuel
Bacher.

REFERENCES

[1] M. Sarazin and F. Roddier, "The ESO differential image motion monitor," Astron. Astrophys. , 227, 294-300
(1990).
[2] A. Tokovinin, "From differential image motion to seeing," Publ. Astron. Soc. Pac. 114, 1156-1166 (2002).
[3] M. Schoeck, D. Le Mignant, G. Chanan, P. L. Wizinowich and M. A. van Dam, "Atmospheric characterization with
the Keck adaptive optics system I: open-loop data," Applied Optics 42 , 3705-3720 (2003).
[4] R. J. Sasiela, Electromagnetic wave propagation in turbulence. Evaluation and application of Mellin transforms,
2nd ed. (SPIE Publications, 2007).
[5] B. E. Stribling, B. M. Welsh and M. C. Roggemann, "Optical propagation in non-Kolmogorov atmospheric
turbulence," Proc. SPIE 2471, 181-196 (1995).
[6] T.W. Nicholls, G. D. Boreman, and J. C. Dainty, "Use of a Shack–Hartmann wave-front sensor to measure
deviations from a Kolmogorov phase spectrum," Opt. Lett. 20, 2460-2462 (1995).
[7] N. S. Kopeika, A. Zilberman and E. Golbraikh "Generalized atmospheric turbulence: implications regarding
imaging and communications", Proc. SPIE 7588, 758808 (2010).
[8] I. Toselli, L. C. Andrews, R. L. Phillips, and V. Ferrero, "Freespace optical system performance for laser beam
propagation through non-Kolmogorov turbulence," Opt. Eng. 47, 026003(2008).
[9] S. Gladysz, K. Stein, E. Sucher, D. Sprung, "Measuring non-Kolmogorov turbulence," Proc. SPIE 8890, 889013
(2013).
[10] I. Toselli, "Introducing the concept of anisotropy at different scales for modeling optical turbulence," J. Opt. Soc.
Am. A 31, 1868-1875 (2014).
[11] D. L. Fried, "Optical resolution through a randomly inhomogeneous medium for very long and very short
exposures," J. Opt. Soc. Am. 56, 1372-1379 (1966).

Proc. of SPIE Vol. 9614 961402-7

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 09/23/2015 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

Potrebbero piacerti anche