Professional Documents
Culture Documents
Jin Et Al - 2020 - Numerical Simulation of Laser Propagation in Ocean Turbulence With The
Jin Et Al - 2020 - Numerical Simulation of Laser Propagation in Ocean Turbulence With The
Abstract. Our work studies the spatial distribution of ocean optical turbulence through numeri-
cal simulation. We used the ocean optical turbulence phase screen created by the ocean power
spectrum to study the effect of varying the refractive index of water via a laser beam propagating
through ocean optical turbulence. Because the intensity of ocean optical turbulence is much
higher than that of atmospheric turbulence, when simulating the ocean phase screen, traditional
methods based on the fast Fourier transform (FFT) algorithm introduce enormous errors in the
low-frequency band. Some interpolation algorithms commonly used in atmospheric turbulence
simulation can reduce these errors. However, their calculation complexity is always high, so the
computational speed is slow. We combine a nonuniform sampling method based on variations in
the ocean turbulent power spectrum and the nonuniform FFT algorithm to generate phase screens
of ocean optical turbulence. The proposed algorithm can solve the deficiency of the long runtime
of the traditional algorithm with even better accuracy. In addition to the traditional phase struc-
ture function, we measured the far-field distribution of the laser transmitting through water
turbulence caused by temperature contrasts, and the experimental results have outstanding
agreement with the simulated far-field image from our ocean phase screen. To the best of our
knowledge, this is the first work to verify the accuracy of the phase screen by conducting far-field
image experiments. © 2020 Society of Photo-Optical Instrumentation Engineers (SPIE) [DOI: 10.1117/1
.OE.59.10.106109]
1 Introduction
Water is an inhomogeneous medium such that its temperature and density change rapidly and
randomly within a small range, which causes turbulence. When light propagates by water tur-
bulence, there will be beam jittering, spot drifting, intensity fluctuations (scintillation), and beam
expansion of the light, which destroys the coherence and directionality of the beam. In summary,
ocean optical turbulence has significantly restricted the application of lidar, optical communi-
cation, imaging, and sensing technologies in underwater marine environments.
At present, adaptive optics (AO) technology is the best solution for solving problems from
turbulence. However, because a large parameter space makes optimization difficult and time
consuming, a precise simulation is necessary for testing the performance of AO with different
configurations for identifying the most suitable one for system design. Therefore, researchers
have invented the turbulence simulated technology for designing an AO system called turbulence
phase screens. The effect of the turbulence on the beam transmission can be simulated using
turbulence phase screens that perturb the phase of a propagating wavefront. Among various
methods for simulating turbulence phase screens, the “power spectrum inversion” method for
obtaining turbulence phase screens from turbulence power spectra has the advantage of not being
limited by the power spectrum model, so it is widely used to study the spatial characteristics of
turbulence. This method uses the power spectra of turbulence to filter a white noise process based
on fast Fourier transform (FFT) techniques, converting frequency domain information to the
spatial domain.1–4 Generally, traditional atmospheric turbulence phase screens (ATPSs) have
evenly spaced sampling frequencies. The advantage is that the FFT algorithm can speed up the
calculations. However, phase screens from traditional ATPSs are severely deficient in the low-
frequency region and have spatial periodicity.5 By contrast, in nature, numerous kinds of tur-
bulence energy are mainly concentrated in the low-frequency region, and random fluctuations do
not have spatial periodicity. Therefore, the traditional method cannot meet the need for accurate
simulation of real ocean optical turbulence.
To solve the problem of deficiency of the low-frequency component of the phase screen, in
1992, Lane et al.5 tried harmonic compensation, but the results obtained by this method still
lacked the information in the low-frequency region. In 2010, Carbillet and Riccardi6 further
improved the harmonic compensation method. In 2013, Xiang7 separated and interpolated the
low-frequency components from the original frequency spectrum by scale division, generating
a low-resolution phase screen containing only low-frequency components. Then the high-
frequency part generated from the FFT algorithm is superimposed on the interpolated low-
frequency phase screen to obtain a completed one. His test proved that the final simulated results
exactly match the actual atmospheric turbulence. In 2013, Xiang8 improved the accuracy of the
high-frequency part by calculating a two-dimensional phase autocorrelation matrix of the
unsampled part in the high-frequency region through numerical integration. Then an FFT was
applied to the autocorrelation matrix to interpolate the sampling points of the power spectrum,
obtaining a phase screen with a highly accurate high-frequency part. In 2013, Charnotskii9
proposed a spectrum sampling method that uses dense sampling in the low-frequency part and
sparse sampling in the high-frequency part to improve the insufficient low-frequency component
of the traditional phase screen. In 2015, Tao et al.10 used a Shack–Hartmann sensor to study the
spatial and temporal characteristics of turbulence by simulating a rotating atmospheric turbu-
lence phase screen. In 2015, Jia et al.11 proposed a combination of the nonuniform fast
Fourier transform (NUFFT) algorithm and heterogeneous sampling to improve operation speed.
Zhang et al.12 compensated for low-frequency power based on sub-bands division and multi-rate
sampling in 2018. In 2019, Zhang et al.13 combined the classical FFT model with the sparse
spectrum model, developing a hybrid method to generate the turbulence phase screen. For oce-
anic turbulence, Sun14 researched the scintillation effect of a Bessel–Gaussian beam in turbu-
lence using the multiple random phase screen in 2018. Pan15 proposed a modified subharmonic
compensation method for simulating oceanic turbulence.
The above methods have achieved excellent results in the simulation of atmospheric turbu-
lence. However, the intensity of oceanic turbulence is much higher than atmospheric turbulence;
hence, the loss of low-frequency information is much higher such that the algorithms mentioned
above cannot be applied due to the considerable computational burden.
In this paper, we first propose a nonuniform sampling algorithm to accurately analyze the
power spectrum of ocean optical turbulence, including both the high-frequency components
and low-frequency components. Then the NUFFT algorithm is introduced to improve the
computing speed. Finally, we design an experiment to quantitatively verify the accuracy of the
proposed method and demonstrate its advantage in terms of the computing speed. In addition,
we innovatively introduce the far-field image test to prove that the proposed phase screen is able
to accurately simulate real turbulence. The experimental results obtained from measuring the far-
field distribution of the laser transmitting through water turbulence have outstanding agreement
with the simulated far-field image from our ocean phase screen.
spectrum introduced in Ref. 16 for homogeneous and isotropic oceanic water is given by the
following expression:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi−11∕3
EQ-TARGET;temp:intralink-;e001;116;711
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2∕3
Φðkx ; ky Þ ¼ 0.388 × 10−8 ε−0.33 k2x þ k2y 1 þ 2.35 k2x þ k2y η
χ T 2 −AT δ
× ðw e þ e−As δ − 2we−AT δ Þ; (1)
w2
where ε is the rate of dissipation of kinetic energy per unit mass of fluid; x and y present the X
and Y directions, respectively; k is the wavenumber of the beam; χ T is the rate of dissipation of
the mean-squared temperature; η is the Kolmogorov microscale (inner scale); and w defines the
ratio of temperature and salinity contributions to the refractive index spectrum, which in ocean
waters can vary in the interval ½−5;0, with −5 and 0 corresponding to dominantly temperature-
induced and salinity-induced optical turbulence, respectively.16 The other parameters in Eq. (1)
are measured through a large number of experiments:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 4∕3 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 EQ-TARGET;temp:intralink-;e002;116;559
Then the phase spectrum of seawater perpendicular to the beam propagation (z) direction is
calculated:16
where Δz is the thickness of the turbulent layer or the propagation distance of the beam and k is
the wavenumber of the beam. The spatial distribution of the ocean phase screen Φ ðkx ; ky Þ is
obtained by filtering the Gaussian random matrix h ðf x ; f y Þ with the phase spectrum of the ocean
following Eq. (1) above. Then with the phase spectrum in two dimensions, the random phase
distribution of turbulence is expressed as
XX qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ϕðx; yÞ ¼ hðf x ; f y Þ FΦ ðkx ; ky Þ exp½jðf x x þ f y yÞ: (4) EQ-TARGET;temp:intralink-;e004;116;379
fx fy
Next, we transform Eq. (4) with the spatial frequency. We set f x and f y to represent the spatial
frequency in the X direction and Y direction, respectively. If a phase screen of size L is sampled
uniformly and the number of sampling points is N, the minimum sampling interval of the phase
screen is Δx ¼ L∕N x , Δy ¼ L∕N y . Then the sampling frequency interval is f x ¼ m 0 Δf x , f y ¼
n 0 Δf y , and m 0 and n 0 are integer indices. Equation (4) is converted to
NX
x ∕2−1 NX
y ∕2−1
0
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi m m n 0n
0 0 0 0
ϕðm; nÞ ¼ hðm ; n Þ FΦ ðm ; n Þ exp j2π þ : (5)
Nx Ny
EQ-TARGET;temp:intralink-;e005;116;269
m 0 ¼−N x ∕2 n 0 ¼−N y ∕2
Equation (5) is now in a form easily implemented using an FFT, but it has inherent limitations.
Its sampling sequence is
N N N
f n ¼ nΔf n ¼ − þ 1; − þ 2; : : : ; : (6)
2 2 2
EQ-TARGET;temp:intralink-;e006;116;190
The minimum and maximum spatial frequencies of the screens generated by uniform sam-
pling here are f min ¼ Δf ¼ 1∕L and f max ¼ N · Δf∕2 ¼ 1∕2Δx, and there is no information
between ð−Δf∕2; Δf∕2Þ. Uniform sampling does not include the power spectrum in the low-
frequency component, resulting in errors when simulating ocean turbulence.
To obtain a random phase screen that accurately complies with the statistical characteristics
of turbulence, low-frequency compensation of the phase screen is needed.10 The most popular
method is the subharmonic compensation method proposed by Lane et al.5 They compensated
for the low-frequency statistics of the phase screen by interpolating the extra sampling point.
With their subharmonic compensation method, the loss of the low-frequency components of the
random phase screen of the atmospheric turbulence is much improved. However, in contrast, the
intensity of ocean turbulence is higher than that of atmospheric turbulence, and tens of harmon-
ics have to be superimposed. As the order of the superimposed harmonics increases, the required
sampling point increases sharply, resulting in a considerable amount of calculation. Therefore, a
modified algorithm is of great importance to achieving rapid simulation of ocean turbulence.
f xm ¼ f yn ¼ f n ¼ 2 · ðf max Þ · n 0 ∕N
EQ-TARGET;temp:intralink-;e007;116;470 ðn 0 ¼ 1;2; 3; : : : ; NÞ: (7)
The power spectral density varies differently in different frequency bands: it rises sharply in
the low-frequency region and approaches infinity; in the high-frequency region, the changes in
density are very gentle. Therefore, we propose nonuniformly sampling the power spectrum with
0 0
its inverse function in Eq. (8). Here f xm and f yn are the final nonuniform sampling points, and α
is the adjustment coefficient:
0 ¼ f 0 ¼ f 0 ¼ α · ðf Þ−11∕3 :
f xm (8)
EQ-TARGET;temp:intralink-;e008;116;377
yn n n
With the sampled spectrum data, the modified frequency power spectrum after nonuni-
form sampling and the phase spectrum of seawater perpendicular to the beam propagation (z)
direction are
0 0
Φnon-uniform ðf m ; f n Þ ¼ Φðf x ; f y Þδðf x − f xm Þδðf y − f yn Þ
(9)
EQ-TARGET;temp:intralink-;e009;116;307
Finally, the phase screen based on nonuniform sampling is expressed by the following for-
mula transformed from Eq. (2):
X
M X
N qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
EQ-TARGET;temp:intralink-;e010;116;231 ϕ 0 ðm; nÞ ¼ hðf xm ; f yn Þ FΦnon-uniform ðf xm ; f yn Þ exp½j2πðmf xm þ yf yn Þ:
m¼−Mþ1 n¼−Nþ1
(10)
Here M and N are equal to the number of sampling frequency points in the x and y directions,
respectively, and ðx; yÞ is the coordinates.
In Figs. 1 and 2, the spectrum of ocean turbulence is sampled using traditional uniform sam-
pling and our nonuniform sampling method, respectively. Here the number of sampling points is
N ¼ 100, and the length of the phase screen L ¼ 1; the turbulence parameters are set to
ε ¼ 0.001 m2 ∕s3 , χ T ¼ 10−8 K2 ∕s; w ¼ −4, Δz ¼ 30 m, and λ ¼ 628 nm.
The sampling intervals of the nonuniform and uniform sampling methods are both shown in
Fig. 2. The interval of the uniform sampling method is only ½1; 102 ; therefore, it lacks the low-
frequency part below 1 Hz. The interval of the nonuniform sampling method is approximately
Fig. 1 The spectrum used to generate the ocean phase screen of this figure and different
sampling methods. Blue triangles: the uniform sampling method and red circles: the proposed
sampling method.
Fig. 2 Comparison of sampling intervals for different sampling methods. Blue triangles: the
uniform sampling method and red circles: the proposed sampling method.
½10−5 ; 102 , which is a much broader range than that of the uniform sampling method, and obvi-
ously, the information in the low-frequency region is retained.
In summary, our method has a broader sampling range, more low-frequency information, and
a distribution that better fits the characteristics of the ocean turbulence power spectrum than the
conventional uniform sampling method without increasing the number of sampling points.
2.3 Accelerating the Generation of Ocean Phase Screen with the NUFFT
Algorithm
The frequency data obtained from nonuniform sampling cannot be estimated using the traditional
FFT algorithm because the data’s distribution intervals vary and do not satisfy the premise of an
FFT, which requires uniformly distributed input data. Although direct addition can calculate these
nonuniform data, it aggravates the computational efficiency. The NUFFT combines the DFT
(nonuniform sampling) and FFT (interpolated oversampling) algorithms17 for computing uneven
data quickly. For nonuniformly distributed data in the spatial domain, frequency domain, or both,
there are three types of NUFFT methods. Because the frequency data in Eq. (9) are nonuniformly
distributed and the spatial result is uniformly distributed, we use a type 2 NUFFT to speed up our
calculations.
The NUFFT process first needs a window function to perform weighted smoothing on non-
uniform raw data to achieve interpolation. Here we select the Gaussian window function as a
smooth convolution kernel:17
X
∞
X
∞
ðf x − 2lπÞ2 ðf y − 2l 0 πÞ2
gτ ðf x ; f y Þ ¼ exp − þ : (11)
4τx 4τy
EQ-TARGET;temp:intralink-;e011;116;651
l¼−∞ l 0 ¼−∞
Here τx and τy are the widths of the Gaussian convolution kernel function in the x and y direc-
tions, respectively. After confirming the oversampling rate R ¼ Mr ∕M and the spreading
distance Msp , and assuming that the width in the x and y directions is the same, then17
1 π
τx ; τy ¼ Msp : (12)
M 2 RðR − 0.5Þ
EQ-TARGET;temp:intralink-;e012;116;565
With the convolution kernel window function, the interpolation for nonuniform data is
obtained from the nonuniform sampled frequency spectrum data:
M X
X N qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2πm 0 2πn 0 2πm 0 2πn 0
fτ ; ¼ hðf xi ; f yj Þ FΦ ðf xi ; f yj Þ × gτ − f xi ; − f yj : (13)
Mr Nr Mr Nr
EQ-TARGET;temp:intralink-;e013;116;496
i¼0 j¼0
By performing an inverse FFT on the above formula, the uniformly distributed spatial data are
obtained:
1 X M X N
2πm 2πn 2πm 2πn
ϕτ ðm; nÞ ¼ fτ ; exp j2π þ : (14)
Mr N r m¼0 n¼0 Mr N r Mr Nr
EQ-TARGET;temp:intralink-;e014;116;421
p ^
Once Φτ ðm; nÞ is known, the Fourier transform of gτ is 2τ e ð−k2τÞ ; Φðm; nÞ, which is the final
phase distribution of ocean turbulence, is obtained by the following equation:17
rffiffiffi
π ðm2 þn2 Þτ
ϕðm; nÞ ¼ e ϕτ ðm; nÞ: (15)
τ
EQ-TARGET;temp:intralink-;e015;116;344
Fig. 3 The 128 × 128 phase screens generated from different methods. (a) Uniform sampling
without compensation; (b) uniform sampling with 12th-order compensation; and (c) nonuniform
sampling with the NUFFT algorithm.
Δz ¼ 10 m, and λ ¼ 628 nm. These ocean turbulence phase screens generated by different
methods are shown separately below.
Because the spectral energy near 0 Hz is not appropriately sampled, the phase screen in
Fig. 3(a) based on uniformly sampled data without extra harmonic sampling points does not
contain the low-frequency component of the ocean power spectrum. Therefore, what is missing
in this phase screen is the tilt and other features that come from turbulent low-frequency infor-
mation. There are currently a few methods for generating a uniformly sampled phase screen.
Herman18 and Johansson19 incorporated subharmonic information by generating a low-
frequency phase screen that is many times larger than the high-frequency screen. The expres-
sion of the low-frequency phase screen is written as
Np
X X
1 X
1 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ϕLF ðm; nÞ ¼
EQ-TARGET;temp:intralink-;e016;116;232 hðf xm 0 f yn Þ FΦ ðf xm 0 f yn Þ exp½j2π · 3−p ðmf xm þ yf yn Þ: (16)
p¼1 m 0 ¼−1 n 0 ¼−1
Here p is the subharmonic order. Compared with the phase screen in Fig. 3(a) with a sample
interval Δf p and minimum sampled frequency f min, the sampling interval in the low-frequency
0
phase screen is Δf p ¼ Δf∕3p, and the minimum sampled frequency is reduced to f min ¼
f min ∕3p. The results from the subharmonic algorithm are presented in Fig. 3(b), which com-
bines the high-frequency part and the low-frequency part. Although Johansson’s method suc-
cessfully compensated for the low-frequency information, harmonic compensation is a process
of resampling, and by adding one compensation level, the computational complexity needs to
increase to Oð3pÞ.
Our proposed nonuniformly sampled phase screen is shown in Fig. 3(c). It not only has many
phase fluctuations that come from high-frequency components but also contains the tilt and other
features that represent the low-frequency feature of the ocean power spectrum. In addition to
comparing the image results, to identify the unique attributes of our algorithm, we chose some
tests that have been used in atmospheric phase screens and designed an underwater experiment.
The first test is the phase structure function. We evaluate these algorithms by how well they
reproduce the desired phase structure function:20
Here ρ and r are the polar coordinates; unlike the atmospheric phase screen, for ocean turbulence
generated by varying temperature and salinity, its phase structure function should be16
5∕3
1 − Δz2 ð1.84w2 − 40.341w þ 2077Þ r
Dϕ ð~rÞ ¼ 2 ; (18)
3.063 × 10−7 k2 ð1.116w2 − 2.235w þ 1.119Þ r0
EQ-TARGET;temp:intralink-;e018;116;619
where r represents the separation from the center point to a varying point and r0 is the coherence
length of ocean turbulence:16
−3∕5
−7 2 −1∕3 χT 2
r0 ¼ 3.603 × 10 k Δzε ð1.116w − 2.235w þ 1.119Þ ; (19)
2w2
EQ-TARGET;temp:intralink-;e019;116;549
where k is the wavenumber of the beam and Δz is the beam propagation distance. The other
parameters were introduced in Sec. 2.
In the structure function test, we performed statistical analysis on phase screen samples (1000
for each method) generated by nonuniform sampling and uniform sampling with 4-, 8-, and 16-
level harmonic compensation19 to reproduce the structure function of ocean turbulence. The
simulated results are shown in Fig. 5 and are compared with the theoretical phase structure func-
tion of ocean turbulence according to Eq. (17).
We define the relative error Re ðrÞ to analyze the accuracy of the phase structure function from
the phase screen as in the following equation:
Dsim ðrÞ
Re ðrÞ ¼ − 1: (20)
Dtheory ðrÞ
EQ-TARGET;temp:intralink-;e020;116;397
Here Dsim ðrÞ is the simulated phase structure function value from the phase screen, and
Dtheory ðrÞ is the value of the theoretical phase structure function. The relative error curve between
the theoretical phase structure function and that calculated from the phase screen produced by
various methods is shown in Fig. 5. A computer equipped with a 3.07-GHz Intel i7 processor and
16 GB memory was used in the experiments, and MATLAB was used as the software platform.
To evaluate the calculation speed of these algorithms, the runtimes of various algorithms are
recorded below:
Figures 4 and 5 show that the simulated structure function generated by uniform sampling is
significantly smaller than the theoretical ocean phase structure function; therefore, harmonic
Fig. 4 Comparison of the theoretical structure function of ocean turbulence and the derived func-
tion from various methods (1000 screens per method).
Fig. 5 Relative error and time taken to simulate the atmospheric turbulence phase screen.
compensation is required. However, the higher the sampling frequency is, the higher the cal-
culation complexity and computing cost are. With 16-level harmonic compensation, the structure
function simulated from the uniform phase screen fits well with the theoretical function, but the
runtime of this process is nearly 12 s.
In contrast, the proposed nonuniform sampling algorithm provides a phase screen with suf-
ficient low-frequency information, and as Fig. 5 shows, the errors in both the high-frequency and
low-frequency sections are small. On the other hand, our method is not perfect, even though the
absolute error is acceptable. Its first drawback is that accuracy is unstable and fluctuates. Second,
the experimental results in Fig. 5 show that our method has relatively high errors in the high-
frequency band because, in our opinion, our nonuniform sampling algorithm sparsely samples
the high-frequency component of the power spectrum. In the NUFFT process, this part requires
the interpolation of more speculative points to map the nonuniformly distributed discrete data to
a uniform grid, which introduces more errors.
The tests of the phase function above demonstrate that, although our approach produces
slightly more error than the traditional harmonic compensation method, considering the accel-
eration ratio (3000 times) of the computing speed, these other inaccuracies are trade-offs that we
consider worthwhile to satisfy the practical requirement. In general, our method reaches accept-
able accuracy with fewer sample points, thus accelerating the speed of producing accurate phase
screens.
Fig. 6 Device for measuring the wave front of a beam in water: (a) photograph of experiment
instruments and (b) the block diagram of our experiment instruments
Fig. 7 (a) System block diagram and temperature settings; (b) wavefront lattice received by the
Shack–Hartmann sensor; and (c) far-field image of the laser
Our phase screen method can simulate far-field images of lasers transmitting through ocean
turbulence. First, several different parameters need to be set: ε is the kinetic dissipation rate per
unit of seawater; χ is the mean-squared temperature dissipation rate of seawater; w is the ratio of
the temperature-induced and salinity-induced ocean turbulence; and Δz is the length that the
laser has traveled through the turbulence. In our experiment, Δz ¼ 0.8 m and w ¼ −5 for the
completely temperature-induced turbulence. However, the measurement of the kinetic dissipa-
tion rate and temperature dissipation rate requires extremely complicated instruments. To sim-
plify this process, we use the maximum likelihood method to estimate the possible values of the
two parameters: χ ¼ 10−6 and ε ¼ 10−6 . Based on the above data, the simulated phase distri-
bution of a laser beam passing through water turbulence is obtained and is denoted by Φ. Then
the light intensity distribution is also determined:
E ¼ A · eiϕ :
EQ-TARGET;temp:intralink-;e021;116;86 (21)
The far-field distribution E 0 of a laser is the angular distribution of the light intensity as it
approaches infinity. It is obtained by measuring the distribution of light intensity on the back
focal plane. Because the distribution of light at this location is a Fourier variation in the ampli-
tude on the diffraction plane, E 0 is derived as in the following formula, where λ is the wavelength
and f is the focal length of the lens:
i
E0 ¼ − F½Eðx; yÞ: (22)
λf
EQ-TARGET;temp:intralink-;e022;116;675
After repeating this calculation 1000 times and taking the average, we obtained the simulated far-
field image as shown in Fig. 8.
We use the spectrograms test to quantitatively describe the fit of the simulated far-field image
to the experimental data. As the spatial frequencies increase, the corresponding energy values
of the simulated and experimental far-field images are plotted in Fig. 9. As shown in Fig. 9,
the simulated results achieve a good fit for all of the spatial frequencies. It is worth mentioning
that energy in the 0- to 100-Hz low-frequency part has extremely high fitness, proving that our
algorithm accurately presents the energy in the low-frequency part. To the best of our knowledge,
this is the first test to prove experimentally the high precision of ocean turbulence simulation
methods.
Fig. 8 (a) Simulated far-field image from the phase screen generated by the proposed method
and (b) far-field image measured from the experiment.
Fig. 9 Comparison of the spatial frequency spectra from the phase-screen-simulated and
experimental far-field images.
4 Conclusion
The effect of ocean turbulence on the phase of light waves can be simulated with a random phase
screen. High-fidelity phase screen generation is crucial for AO performance modeling, but the
generation speed is limited by the existing algorithms used, and these simulation methods lack
experimental validation for accuracy. We have developed a new method of ocean phase screen
generation that is well suited to the problem of fast and accurate generation of large-scale ocean
turbulence for marine scientific research. We nonuniformly sampled the ocean power spectrum
and then implemented the NUFFT algorithm to reconstruct the precise phase of ocean turbulence
from the frequency and phase information.
The structure function test showed that logarithmic sampling of the turbulence spectrum
leads to accurate phase screens for ocean turbulence simulation. Moreover, compared with the
previous method, our method achieved a significant improvement in the calculation speed and
consumed fewer computer resources. It is appropriate for rapidly simulating large-scale, high-
resolution ocean turbulence phase screens. Additionally, we implemented an experiment to
obtain the far field image of the laser beam through water turbulence to demonstrate the accuracy
of the proposed approach. The excellent fitness between the experimental and simulated far-field
images demonstrates that the method developed here is prescriptively accurate.
References
1. A. J. Fleck, J. R. Morris, and M. D. Feit, “Time-dependent propagation of high energy laser
beams through the atmosphere,” Appl. Phys. 10, 129–160 (1976).
2. D. L. Knepp, “Multiple phase-screen calculation of the temporal behavior of stochastic
waves,” Proc. IEEE 71, 722–737 (1983).
3. J. D. Schmidt, Numerical Simulation of Optical Wave Propagation, pp. 149–184, SPIE,
Washington, DC (2010).
4. D. Voelz, Computational Fourier Optics, pp. 178–182, SPIE, Washington, DC (2011).
5. R. G. Lane, A. Glindemann, and J. C. Dainty, “Simulation of a Kolmogorov phase screen,”
Waves Random Media 2(3), 209–224 (1992).
6. M. Carbillet and A. Riccardi, “Numerical modeling of atmospherically perturbed phase
screens: new solutions for classical fast Fourier transform and Zernike methods,” Appl.
Opt. 49(31), G47–G52 (2010).
7. J. Xiang, “Accurate compensation of the low-frequency components for the FFT-based tur-
bulent phase screen,” Opt. Express 20(1), 681–687 (2012).
8. J. Xiang, “Fast and accurate simulation of the turbulent phase screen using fast Fourier
transform,” Opt. Eng. 53(1), 016110 (2014).
9. M. Charnotskii, “Sparse spectrum model for a turbulent phase,” J. Opt. Soc. Am. A 30(3),
479 (2013).
10. T. Tao et al., “Measurement of temporal and spatial characteristics of atmospheric
turbulence simulated by a rolling phase screen,” Prog. Laser Optoelectron. 52(8), 080101
(2015).
11. P. Jia et al., “Real-time generation of atmospheric turbulence phase screen with non-uniform
fast Fourier transform,” Mon. Not. R. Astron. Soc. 450, 38–44 (2015).
12. D. Zhang et al., “Atmospheric turbulence phase screen modeling method based on sub-
bands division and multirate sampling,” Optik 163, 72–80 (2018).
13. Z. Dong-Xiao et al., “Generation of turbulence phase screen based on gravitational search
algorithm,” Acta Phys. Sin. 68, 134205 (2019).
14. R. Sun et al., “Multiple random phase-screen simulation of scintillation effect of Bessel–
Gaussian beam in ocean turbulence,” in 12th Int. Symp. Antennas, Propagation and EM
Theory, Hangzhou, China (2018).
15. S. Pan et al., “An effective way for simulating oceanic turbulence channel on the beam
carrying orbital angular momentum,” Sci. Rep. 9, 14009 (2019).
16. L. Lu, X. Ji, and Y. Baykal, “Wave structure function and spatial coherence radius of
plane and spherical waves propagating through oceanic turbulence,” Opt. Express 22(22),
27112–27122 (2014).
17. L. Greengard and J. Lee, “Accelerating the non-uniform fast Fourier transform,” SIAM Rev.
46(3), 443–454 (2004).
18. B. J. Herman and L. A. Strugala, “Method for inclusion of low-frequency contributions in
numerical representation of atmospheric turbulence,” Proc. SPIE 1221, 183–192 (1990).
19. E. M. Johansson and D. T. Gavel, “Simulation of stellar speckle imaging,” Proc. SPIE 2200,
372–383 (1994).
20. J. W. Giles and I. N. Bankman, “Underwater optical communications systems. Part 2: basic
design consideration,” in IEEE Military Commun. Conf., Vol. 3, p. 1700 (2005).
Ruiyan Jin received his master’s degree from Brown University, USA, in 2016. He is now an
assistant engineer at the Institute of Optics and Electronics, Chinese Academy of Sciences,
Chengdu, China. His research interests include image signal processing and neural networks.
Shuai Wang received his PhD from the Institute of Optics and Electronics, Chinese Academy of
Sciences, Chengdu, China, where he currently works as an associate professor. His research
interests include signal process and applied optics.
Ping Yang received his PhD from the Institute of Optics and Electronics, Chinese Academy of
Sciences, Chengdu, China, where he is currently working as a professor. His research interests
include image signal processing, neural networks, and applied optics.