HHS Public Access

Author manuscript
J Mod Opt. Author manuscript; available in PMC 2017 January 01.
Author Manuscript

Published in final edited form as:

J Mod Opt. 2016 ; 63(1): 80–84. doi:10.1080/09500340.2015.1073804.

Optimization of focusing through scattering media using the

continuous sequential algorithm
J.V. Thompson*, B.H. Hokr, and V.V. Yakovlev
Texas A&M University, College Station, TX, USA

Abstract
Author Manuscript

The ability to control the propagation of light through scattering media is essential for atmospheric
optics, astronomy, biomedical imaging and remote sensing. The optimization of focusing light
through a scattering medium is of particular interest for the case of highly scattering materials.
Optical wavefront beam-shaping plays a critical role in optimizing such a propagation; however,
an enormous field of adjustable parameters makes the overall task complicated. Here, we propose
and experimentally evaluate several variations on the standard continuous sequential algorithm
that hold a promise of revealing new, faster and more efficient optimization algorithms for
selecting an optical wavefront to focus light through a scattering medium. We demonstrate that the
order in which pixels are chosen in the continuous sequential algorithm can lead to a 2-fold
decrease in the number of iterations required to reach a given enhancement.
Author Manuscript

Keywords
beam phase shaping; wavefront shaping; optimization algorithm; scattering; continuous sequential
algorithm

1. Introduction
As light propagates through a scattering material, interference and scattering effects cause
the beam to be scattered in all directions. This reduces the amount of light directly
transmitted through the material. This is illustrated in Fig. 1, where a laser focus without a
scattering material in the beam is compared to the same laser beam with a scattering material
in the beam. This scattering of light appears to cause the information of the laser beam to be
lost. However, research has discovered that through the use of adaptive optics and wavefront
shaping, the scattering effects caused by a material may be suppressed, and the information
Author Manuscript

recovered. This is demonstrated in Fig. 1(c).

These techniques that utilize wavefront shaping to control the way that light scatters in a
scattering medium have been the subject of much research recently. Many possible
applications for these techniques have been demonstrated. One of the more interesting
applications is the use of wavefront correction to image through[1–5] or around[2, 6]
scattering material. Other applications include sub-diffraction and subwavelength

*
Corresponding author. [email protected].
 Thompson et al. Page 2

focusing[7, 8], Compression of ultrashort pulses[9], creating a dynamic waveplate[10],

Author Manuscript

spectral filtering[11], deep tissue chemical imaging[12, 13], and random Raman lasing[14,
15]. It has also been shown that wavefront correction can substantially enhance the total
transmission of light through a scattering media[16]. In this case of the enhancement of total
transmission, the optimization is not local. On the other hand, the optimization of focusing
through scattering media is local and linear, thus larger enhancements are realized more
easily.

By modulating the phase of spatial components of a beam of laser light, the amount of light
transmitted by a scattering medium can be enhanced[8, 17]. The modulation of the phase of
the beam is typically done by a two dimensional liquid crystal spatial light modulator. A
pattern is displayed on the spatial light modulator determined by an optimization algorithm
that requires feedback from some sort of detector. Different algorithms have been developed
in order to find an algorithm that both converges well and fast. The stepwise sequential and
Author Manuscript

continuous sequential algorithms[18] converge well, but the convergence begins slowly.
Another method is to directly measure the transmission matrix of the scattering material and
then correct for it with phase modulation[19–22]. Other algorithms, such as the partitioning
algorithm[18] and genetic algorithms[23] converge more quickly, but may or may not
converge as well as the sequential algorithms, depending on the sample used.

One major obstacle to realizing the goal of real time imaging in tissue is the large number of
iterations that each of the above algorithms requires. A typical spatial light modulator has on
the order of 4 × 105 8-bit pixels. This means there are 108 total independent phase masks
possible. At a refresh rate of 10 Hz, it is not possible for a typical spatial light modulator to
search every possible phase mask in a reasonable amount of time. A decent algorithm can
reduce the number of phase masks required to search down to 103 or 104. This means a full
Author Manuscript

optimization may still take anywhere from hundreds of seconds to hours. In order to
optimize for scattering in living tissue, this time needs to be less than the decorrelation time
of the sample which is on the order of one millisecond[24–26] to seconds[27]. Thus, the
need for an algorithm that can reach a high level of optimization with a very low number of
iterations is vital to the realization of the previously mentioned applications.

In this paper we investigate three geometrical variations of the standard continuous

sequential algorithm[18]. This analysis provides some insights into new possible
enhancement algorithms. Additionally, by selecting a different geometry of regions of the
beam to modulate, the more sensitive regions may be modulated sooner, resulting in a faster
optimization.

2. Experimental Setup
Author Manuscript

Our experimental setup is shown in Fig. 2. The power and polarization of a 532 nm diode
pumped solid state laser are selected by a half-wave plate and polarizing beam cube. The
polarization is set to match the phase only modulation orientation of a two dimensional
liquid crystal spatial light modulator (Hamamatsu X10468-07). This spatial light modulator
has an array of 600 × 792 8-bit pixels, and refreshes at a frequency of 10 Hz. The transverse
mode of the laser is also spatially filtered and the beam expanded before the spatial light

J Mod Opt. Author manuscript; available in PMC 2017 January 01.

 Thompson et al. Page 3

modulator to fill more of the 2-D pixel array. After modulation, the light is passed through a
Author Manuscript

telescope with an iris at the focus to remove higher order diffraction modes caused by the
spatial light modulator.

After the telescope, the modulated light is directed with a mirror and focused into a ∼100
μm thick sample of Teflon. Teflon is chosen as a sample because it has a high scattering
cross section and a long persistence time. The power of the laser light focused onto the
sample is 0.8 mW. On the other side of the sample, a 20× objective with 0.40 numerical
aperture is used to collect the transmitted light and image it onto a CCD detector(Mightex
CGE-B013-U). A square region of 100 pixels near the center of the CCD is selected, and the
individual intensity counts are summed to provide a single value to use as feedback for the
optimization algorithm.

3. CSA Algorithm variations

Author Manuscript

The standard continuous sequential algorithm(CSA)[18] works by selecting a square block

of pixels on the spatial light modulator. A series of phase shifts are applied to the selected
region of pixels, and measurements made for each phase shift. The optimum phase shift is
then applied to the spatial light modulator, and a new block of pixels are selected. This
process iterates in a linear fashion rastering over the 2 dimensional pixel array as shown in
Fig. 3(a), and can be repeated for additional enhancement.

We investigate the effect of three variations in the pixel selection method on the standard
continuous sequential algorithm. These variations are depicted in Fig. 3(b-d). The first two
are very similar to the original CSA, however instead of raster scanning through the 2d pixel
array, the modulated pixel groups are selected in a spiral pattern starting at the edges (b) or
at the middle (c) of the array.
Author Manuscript

The last geometry (d) selects pixel groups to modulate in a ring geometry. The algorithm
cycles through nine rings with different origins, then the radius of the ring is decreased, and
the cycle repeats until the ring becomes a spot at each origin. When the rings are large
enough to overlap, the phase shift of the new ring is added to the phase shift of the old ring
rather than replacing it. The phase shift is wrapped back to zero when the new value exceeds
4π radians. Thus, the structure of the phase mask is not overwritten by new iterations.
Additionally, before selecting the rings, the pixel array is grouped into “macropixels” of 36
pixels (6×6) to avoid large power loss due to diffraction.

For the measurements presented here, 12 sizes of rings (or 108 iterations total) are used. For
the raster and spiral geometries, 110 pixel blocks are used. Thus, the four algorithms have a
Author Manuscript

comparable number of iterations. For all geometries, eight measurements spanning 2π

radians phase shift are made for each iteration.

4. Results
The enhancement as a function of iteration number for the four pixel selection geometries
are shown in Fig. 4. The enhancement is calculated as the intensity of light in the region of
interest on the CCD divided by the initial intensity of light in the same region. As can be

J Mod Opt. Author manuscript; available in PMC 2017 January 01.

 Thompson et al. Page 4

seen in the figure, the enhancements at the end of the algorithm cycle are comparable.
Author Manuscript

However, the speed at which they reach that level depends on how quickly the algorithm
optimizes the center pixels in the 2d pixel array. This is intuitive because the laser beam
contains more power in that region.

A simple simulation was done to verify the results of the spiral pixel selection geometries.
The intensity of the light detected by the CCD is simulated via the equation[18]

where only one output channel (m = 1) is measured, tmn is the transmission matrix of the
Author Manuscript

scattering material, An is the 2 dimensional amplitude distribution of the laser (Gaussian), φn

is the phase shift applied by the spatial light modulator, and N is the number of pixel groups
(110). tmn is approximated using a complex random Gaussian distribution[18]. While in
reality the elements of tmn are correlated[16], this approximation of uncorrelated elements
still provides insight into the improvement of optimization algorithms for the case of
focusing.

The simulation results are shown in Fig. 5 for zero measurement noise (a) and for noise on
the order of 0.1I0 (b). In both cases, the spiral out geometry enhances the focus more quickly
than the other geometries. This effect comes from the An coefficients, which are chosen to
mimic the Gaussian profile of the laser beam. Since the center of the beam contains more
laser power, modulation of those pixels first returns larger enhancement earlier on in the
optimization process. For the zero noise case, the algorithms converge to the same
Author Manuscript

enhancement after a full cycle. However, when noise is introduced, the spiral out geometry
does slightly better. This demonstrates the benefit of enhancing the system above the noise
earlier on rather than later in the algorithm cycle.

5. Conclusion
In conclusion, we have proposed and experimentally evaluated three variations of the
continuous sequential algorithm for wavefront optimization. These variations provide
insight into possible ways to improve the optimization algorithms of wavefront shaping.
Mainly, by first selecting the region where laser light is brightest on the spatial light
modulator, the algorithm will have a greater enhancement early on in the optimization
process. Additionally, this can be even more beneficial if there is noise in the system since
Author Manuscript

enhancing the signal above the noise early on will improve the sensitivity of the later, less
sensitive modulations.

Furthermore, we have presented a pixel selection geometry that takes into consideration the
cylindrical nature of the laser beam, but still provides for a symmetry breaking to relate to
the random nature of the scattering material. This new geometry performs similar to the
standard continuous sequential algorithm, but future modifications to the technique may

J Mod Opt. Author manuscript; available in PMC 2017 January 01.

 Thompson et al. Page 5

prove fruitful. In the very least, it gives new ideas of how faster optimization may be
Author Manuscript

achieved.

The technique of wavefront shaping to control scattered light is a very powerful tool with
many diverse applications. More applications will continue to become readily available as
further investigation of optimization algorithms improve the speed of enhancement.

Acknowledgments
This research was is part supported by the National Science Foundation (PHY-1241032 (INSPIRE CREATIV),
CBET award #1250363, DBI awards #1455671 and 1532188, and ECCS award #1509268), the National Institute of
Environmental Health Sciences of the National Institutes of Health under grant number P30ES023512, the
Kazakhstan Ministry of Education and Science (grant #5385/GF4), the Robert A. Welch Foundation (Award
A-1261), and the United States-Israel Binational Science Foundation (Grant No. 2012357). JVT is supported by the
Herman F. Heep and Minnie Belle Heep Texas A&M University Endowed Fund held/administered by the Texas
A&M Foundation. This work was partially supported by the start-up funds available through Texas A&M
Author Manuscript

University. BHH would like to acknowledge a graduate fellowship from the Department of Defense Science,
Mathematics and Research for Transformation (SMART) fellowship program.

References
1. Bertolotti J, van Putten EG, Blum C, Lagendijk A, Vos WL, et al. Nature. 2012; 491(7423):232–4.
[PubMed: 23135468]
2. Katz O, Heidmann P, Fink M, et al. Nat Photonics. 2014; 8(10):784–790. Aug.
3. Lai P, Wang L, Tay JW, et al. Nat Photonics. 2015; 9(2):126–132. [PubMed: 25914725]
4. Judkewitz B, Wang Y, Horstmeyer R. Nat Photonics. 2013; 7(4):300–305. [PubMed: 23814605]
5. Xu X, Liu H, Wang LV. Nat Photonics. 2011; 5(3):154–157. Jan. [PubMed: 21532925]
6. Katz O, Small E, Silberberg Y. Nat Photonics. 2012; 6(8):549–553. Jul.
7. Vellekoop IM, Lagendijk A, Mosk A. Nat Photonics. 2010; 4(5):320–322. Feb.
8. Mosk AP, Lagendijk A, Lerosey G, et al. Nat Photonics. 2012; 6(5):283–292. May.
Author Manuscript

9. Katz O, Small E, Bromberg Y, et al. Nat Photonics. 2011; 5(6):372–377. May.

10. Park JH, Park C, Yu H, Cho YH, et al. Opt Express. 2012; 20(15):17010–17016.
11. Park JH, Park C, Yu H, Cho YH, et al. Opt Lett. 2012; 37(15):3261–3263. [PubMed: 22859152]
12. Arora R, Petrov GI, Yakovlev VV, et al. Proc Natl Acad Sci. 2012; 109(4):1151–1153. [PubMed:
22215594]
13. Arora R, Petrov GI, Yakovlev VV, et al. Anal Chem. 2014; 86(3):1445–1451. [PubMed:
24380569]
14. Hokr BH, Bixler JN, Noojin GD, Thomas RJ, Rockwell BA, Yakovlev VV, et al. Proc Natl Acad
Sci. 2014; 111(34):12320–12324. [PubMed: 25114231]
15. Hokr BH, Bixler JN, Cone MT, Mason JD, Beier HT, Noojin GD, Petrov GI, Golovan LA,
Thomas RJ, Rockwell BA, et al. Nat Commun. 2014; 5:4356. [PubMed: 25014073]
16. Popoff SM, Goetschy A, Liew SF, Stone aD, et al. Phys Rev Lett. 2014; 112(13):133903.
[PubMed: 24745422]
17. Vellekoop IM, Mosk AP. Opt Lett. 2007; 32(16):2309–2311. [PubMed: 17700768]
Author Manuscript

18. Vellekoop IM, Mosk AP. Opt Commun. 2008; 281(11):3071–3080. Jun.
19. Chaigne T, Katz O, Boccara aC, Fink M, Bossy E, et al. Nat Photonics. 2014; 8(1):58–64. Nov.
20. Popoff SM, Lerosey G, Fink M, Boccara aC, et al. New J Phys. 2011; 13(12):123021.
21. Popoff S, Lerosey G, Fink M, Boccara AC, et al. Nat Commun. 2010; 1:81. [PubMed: 20865799]
22. Yoon J, Lee K, Park J, et al. arXiv Prepr arXiv 1502 04761. 2015
23. Conkey DB, Brown AN, Caravaca-Aguirre AM, et al. Opt Express. 2012; 20(5):4840–4849. Feb.
[PubMed: 22418290]
24. Wang YM, Judkewitz B, DiMarzio Ca, et al. Nat Commun. 2012; 3:928. [PubMed: 22735456]

J Mod Opt. Author manuscript; available in PMC 2017 January 01.

 Thompson et al. Page 6

25. Draijer M, Hondebrink E, Van Leeuwen T, et al. Lasers Med Sci. 2009; 24(4):639–651. [PubMed:
19050826]
Author Manuscript

26. Hajjarian Z, Xi J, Jaffer Fa, Tearney GJ, et al. J Biomed Opt. 2011; 16(2):026005. [PubMed:
21361689]
27. Cui M, McDowell EJ, Yang C. Opt Express. 2010; 18(1):25–30. [PubMed: 20173817]
Author Manuscript
Author Manuscript
Author Manuscript

J Mod Opt. Author manuscript; available in PMC 2017 January 01.

 Thompson et al. Page 7
Author Manuscript

Figure 1.
Laser spot with (b) and without (a) scatterer in the beam. This illustrates what happens when
light propagates through a scattering material. However, wavefront shaping offers a method
to counter these effects (c). (a) was attenuated with 4.4 OD and (b) and (c) were attenuated
Author Manuscript

with 1 OD neutral density filters.

Author Manuscript
Author Manuscript

J Mod Opt. Author manuscript; available in PMC 2017 January 01.

 Thompson et al. Page 8
Author Manuscript

Figure 2.
Experimental Setup. Laser power and polarization from a 532 nm laser are selected by a
half-wave plate (WP) and polarizing beam splitter (PBS). The laser light is then spatially
Author Manuscript

filtered and expanded. Then the light is modulated by a spatial light modulator (SLM).
Higher order diffraction modes from the SLM are removed by an iris inside another
telescope. The modulated light is focused onto a sample (S) of Teflon. Feedback to the
optimization algorithm is supplied by a CCD camera on the backside of the sample. BS,
beam splitter; M, mirror; L, lens; Obj, objective.
Author Manuscript
Author Manuscript

J Mod Opt. Author manuscript; available in PMC 2017 January 01.

 Thompson et al. Page 9
Author Manuscript

Figure 3.
Geometry variations of the pixel selection order of the continuous sequential algorithm. The
grayscale illustrates the amount of phase shift applied to the spatial light modulator (SLM),
and the arrows show in which direction the pixel selection proceeds. The typical geometry is
to raster scan (a) through the pixel groups on the SLM. Other methods examined are to start
on the edge and spiral inwards (b), or start in the center and spiral outwards (c). Another
pixel selection geometry is constructed of rings of decreasing radius centered at nine
Author Manuscript

different origins (d). When the rings overlap, the phase shifts applied by each ring are added.
Author Manuscript
Author Manuscript

J Mod Opt. Author manuscript; available in PMC 2017 January 01.

 Thompson et al. Page 10
Author Manuscript
Author Manuscript

Figure 4.
Optimization results for four different pixel selection geometries for one complete loop of
the continuous sequential algorithm.
Author Manuscript
Author Manuscript

J Mod Opt. Author manuscript; available in PMC 2017 January 01.

 Thompson et al. Page 11
Author Manuscript

Figure 5.
Simulation results for the raster and spiral pixel selection geometries for one complete loop
of the continuous sequential algorithm. Results for zero noise (a) and noise of 10% of the
Author Manuscript

initial intensity (b) are shown. Note that for the zero noise scenario, the three geometries
result in an identical enhancement.
Author Manuscript
Author Manuscript

J Mod Opt. Author manuscript; available in PMC 2017 January 01.