Research Article Applied Optics 1

Virtual pyramid wavefront sensor for phase

unwrapping
V YAS A KONDI1,* , B RIAN VOHNSEN2 , AND S USANA M ARCOS1
1 VisualOptics and Biophotonics Laboratory, Instituto de Óptica “Daza de Valdés,” Consejo Superior de Investigaciones Científicas, C/Serrano 121, 28006
Madrid, Spain
2 Advanced Optical Imaging Group, School of Physics, University College Dublin, Belfield, Dublin 4, Ireland
* Corresponding author: [email protected]

Compiled September 15, 2017

Noise affects wavefront reconstruction from wrapped phase data. A novel method of phase unwrap-
ping is proposed with the help of a virtual pyramid wavefront sensor. The method was tested on noisy
wrapped phase images obtained experimentally with a digital phase-shifting point diffraction interfer-
ometer. The virtuality of the pyramid wavefront sensor allows easy tuning of pyramid apex angle and
modulation amplitude. It is shown that an optimal modulation amplitude obtained by monitoring the
Strehl ratio helps in achieving better accuracy. Through simulation studies and iterative estimation, it is
shown that the virtual pyramid wavefront sensor is robust to random noise. © 2017 Optical Society of America
OCIS codes: (010.1080) Active or adaptive optics; (010.7350) Wave-front sensing; (120.5050) Phase measurement; (280.4788)
Optical sensing and sensors; (080.1010) Aberrations (global); (110.4280) Noise in imaging systems; (120.3180) Interferometry.
http://dx.doi.org/10.1364/ao.XX.XXXXXX

1. INTRODUCTION errors due to centroiding arising from random noise and has
a cumulative effect on the unwrapped phase. In addition, the
Phase unwrapping is an important step in several optical inter-
wrapped phase is sub-divided into a finite number of subaper-
ferometry, adaptive optics and imaging applications [1–3]. The
tures and the slope sampling is limited by the resolution of
effects of noise become crucial when the number of pixels per
the phase map. Ambiguities due to jumps in phase close to
interference fringe is not sufficiently large and the signal to noise
the borders of the subapertures can cause further inaccuracies.
ratio is low. In such situations, conventional phase unwrapping
Aforementioned shortcomings place a lower limit on the signal-
methods are not consistent [4]. Most phase unwrapping meth-
to-noise ratio (SNR) for which diffraction-limited performance
ods are application dependent and demand an optimization of
may be achieved [14]. In this letter, a novel method of phase
multiple parameters. The sources of noise are inherent diffrac-
unwrapping is proposed based on the principles of a pyramid
tion, light source fluctuations, aberrations due to the sample and
wavefront sensor (PWS) [15] that significantly lowers this limit.
the optical system. Noise can be overcome by smartly avoiding
The PWS performs better averaging of random noise with a
noisy pixels in a branch cut phase unwrapping algorithm [1], al-
larger pupil sampling and better resolution. Here, the wrapped
though this fails if the noise is high [5–11]. Filtering is often used
phase is assumed to be located in the back focal plane of a lens,
to reduce noise and such an operation allows smoothening near
which is situated at a focal distance away from a simulated pyra-
the phase jumps and elimination of potential useful information,
midal prism with four facets. Since the PWS is not physically
which could adversely affect phase unwrapping [12].
present, this method of phase unwrapping is called the virtual
Recently, a virtual Hartmann-Shack (HS) method was pro-
PWS. Modulation plays an important role in controlling the
posed for phase unwrapping [4]. The wrapped phase was as-
dynamic range and sensitivity of the PWS [16] and its role in
sumed to be incident on an array of simulated microlenses. The
phase unwrapping is studied. The method is tested on noisy
locations of the simulated HS focal spots were estimated us-
interferograms recorded experimentally with a CCD camera in a
ing the intensity weighted centroiding algorithm [13] and the
digital phase-shifting (PS) point diffraction interferometer (PDI)
unwrapped phase was recovered from the calculated local wave-
[4, 17] by using a spatial light modulator to avoid mechanically
front slopes. It was shown that the accuracy of phase unwrap-
moving components. The light used was a 632.8 nm He-Ne laser.
ping primarily depends on sampling of the wrapped phase and
Aberrations are introduced with the help of a MEMS deformable
diffraction-limited wavefront sensing can be achieved using an
mirror (Boston Micromachines Corporation™) in closed-loop
iterative estimation procedure in the presence of noise. How-
with a commercial HS wavefront sensor.
ever, the performance of the virtual HS is limited by localized
 Research Article Applied Optics 2

2. METHOD
Let us imagine that the wrapped phase, φw ( x, y) (experimen-
tally obtained in a PS-PDI when aberrations were induced with
a deformable mirror) defined in the interval [-π π] within a cir-
cular pupil, P( x, y) be located at the back focal plane of a lens
with a focal length of 1 m (physically not present) that focusses
light onto a pyramidal prism (physically not present) with its
phase, T ( X, Y ) [16, 18] as illustrated in Fig. 1. This pyramidal
phase would divide the incident light into four distinct pupils
and using a fast Fourier transform (FFT) method [19], the pupil
plane intensity, I pyr can be evaluated as shown below:
w 2
I pyr ( x, y) = FFT ( FFT ( P( x, y).eiφ ( x,y) ).T ( x, y)) (1)

A few examples of the sources of aberrations used when record-

ing the wrapped interferogram are imperfections in the optical
elements, a patient’s eye [20], a specimen under an optical micro-
scope [21] and atmospheric turbulence in a telescope [22]. From
a linear combination of the intensities in the four pupils of the
virtual PWS (Ij , 1≤ j ≤ 4) in I pyr ( x, y), the local ‘x’ and ‘y’ wave-
front slopes, Sx ( x, y) and Sy ( x, y) can be evaluated [16]. The un-
wrapped phase, φ( x, y) is reconstructed from the estimated slope
values using the slope geometry of Southwell [13, 14, 23]. The
Fig. 2. Experiments: (a) wavefronts measured with a commer-
Zernike polynomials are used to decompose the reconstructed
cial HS wavefront sensor; (b) wrapped phase (radians: -π to
wavefront φ( x, y) using a least square fitting technique to elimi-
π) obtained in a PS PDI; (c) wavefront reconstructed with the
nate the high spatial frequency components and artifacts arising
virtual PWS and (d) residual wavefront error.
from noise.

deformable mirror and the HS wavefront sensor. The wrapped

phase images are calculated using Eq. (2) from the interfero-
grams recorded by a digital PS PDI and are shown in Fig. 2(b).
From the wrapped phase, the intensity at the pupil plane, I pyr
of a pyramid wavefront sensor is evaluated using Eq. (1) and 25
measurements in a circular modulation such that the tip of the
pyramid oscillates in a circular path around the focal point of
the focussed beam. The radius of the circular path is the mod-
Fig. 1. Illustration of the proposed virtual pyramid wavefront
ulation amplitude. Then, the local wavefront slopes, Sx and Sy
sensor. Here, the lens ‘L’ and the pyramidal phase T ( x, y) are
are estimated. Finally, the wavefronts are reconstructed using
physically not present. The wrapped phase is obtained in a
the singular value decomposition technique [13]. The recon-
digital PS-PDI [17] and I pyr is evaluated using Eq. 1.
structed wavefronts are decomposed using the first four orders
of Zernike polynomials and are shown in Fig. 2(c). It may be
To compare and test the capability of the proposed method noted that the wavefronts reconstructed with the virtual PWS
in the presence of noise, simulations are performed. The PS PDI match well with the induced aberrations as validated by the low
interferograms (I1 , I2 and I3 corresponding to phase shifts of residual wavefront error shown in Fig. 2(d), which is calculated
0, π/2 and π respectively) and hence the wrapped phase are by subtracting the decomposed wavefronts in Fig. 2(c) from the
calculated as follows: induced aberrations in Fig. 2(a). Here, a modulation amplitude
2I2 ( x, y) − I3 ( x, y) − I1 ( x, y) that maximizes the estimated Strehl ratio for each aberration is
 
φw ( x, y) = tan−1 . (2) used. This optimal modulation depends on the nature of the
I3 ( x, y) − I1 ( x, y)
aberrations, their magnitude and SNR. Although 1529 × 1529
Here, the interferograms (I1−3 ) in a digital PS PDI are obtained ‘x’ and ‘y’ local slope values were calculated from the wrapped
as described earlier [4, 17]. phase, for reconstruction of the wavefront, the slope matrices
Root-mean-square (RMS) of the residual wavefront is used were resized to 31 × 31 by using bilinear interpolation. This
to evaluate the quality of reconstruction [4]. Alternately, the sampling was chosen because a higher sampling needs more
Strehl ratio of the residual wavefront is evaluated by assuming computations and increasing the slope sampling beyond 31 ×
that the pupil plane has aberrations equal to the difference of 31 did not increase the Strehl ratio any further. On average,
the induced aberration and the reconstructed wavefront. If the the Strehl ratio and RMS were found to be 0.28 and 0.14 µm
estimated wavefront is different from the induced wavefront, respectively for the residuals in Fig. 2(d), indicating a big leap in
the Strehl ratio would be lower than unity. performance over virtual HS [4] that resulted in a mean Strehl
ratio and mean RMS error of 0.11 and 0.20 µm for the same
3. RESULTS wrapped phase images.
Figure 2(a) shows the wavefronts - defocus, astigmatism, coma The estimated Strehl ratio and RMS error as a function of
and secondary astigmatism - generated in closed-loop using the modulation amplitude are shown in Fig. 3. Simulations were
 Research Article Applied Optics 3

To study the effects of noise through simulations, indepen-

dently generated white Gaussian noise was added to each of
the interferograms (I1−3 ) before calculating the wrapped phase
(see Eq. (2)). As the SNR increases, the discrepancies in the
detected slopes reduce resulting in an increase in the Strehl ratio
of the residual wavefront map as illustrated by the simulation
results in Fig. 5(a) while sensing the aberrations introduced
by the deformable mirror (see Fig. 2(a)). With an increase in
the magnitude of aberrations or with the inclusion of higher-
order aberrations, the minimum SNR needed for convergence
increases. This is attributed to an increase in the number of
fringes per pixel. However, for a fixed modulation amplitude
Fig. 3. Effect of modulation amplitude on wavefront sensing (optimal value predicted at SNR = ∞) and following an iterative
accuracy in the experiments and theoretical predictions at SNR approach while estimating Strehl ratio [4, 16], it is possible to
= ∞ in terms of (a) Strehl ratio and (b) RMS error. improve performance as illustrated in Fig. 5(b). This is done by
assuming that the residual wavefront obtained in the first loop
of the phase unwrapping process is the new wavefront for the
performed at SNR = ∞ and clearly, the accuracy is higher under second iteration. This new wavefront is used to calculate the
noise-free conditions. In addition, the experimental curves are interferograms and the wrapped phase for the second iteration
broader than the theoretical counterparts. The differences arise and so on. This iterative procedure can be implemented in an op-
due to inaccuracies in the generation of the aberrations, mis- tical system by monitoring the point spread function with a CCD
alignment errors in the digital PS PDI and the presence of noise camera. It can be seen in Fig. 5(b) that for the case of defocus,
in the interferograms. The tilt due to PS (as seen in Fig. 2(b)) a decrease in the SNR from 10 dB to 5 dB requires an increase
prevents convergence with increasing modulation as predicted in the minimum number of iterations by three to go beyond the
earlier [16]. diffraction limit. Since most applications including visual optics
Figure 4 shows the results of simulations using the virtual
PWS. Here, the wavefronts measured by the HS wavefront sen-
sor were used as the starting point in the simulations (see Fig.
4(a)) and no effects of noise were included while estimating the
wrapped phase shown in Fig. 4(b) from the calculated interfer-
ograms [4, 17]. The wavefronts reconstructed with the virtual
PWS from the wrapped phase in Fig. 4(b) are shown in Fig. 4(c).
The mean RMS error shown in Fig. 4(d) is 0.10 µm and the best
accuracy was achieved near a modulation of 0.5 mm (see Fig. 3).

Fig. 5. Simulations: (a) Accuracy improves with increasing

signal while sensing simulated aberrations (Fig. 4(a)) and
results of (b) iterative evaluation. Here, random noise (single
trial) is added to the calculated interferograms.

involve a combination of the low-order aberrations that play a

prominent role, the proposed method was tested with randomly
simulated wavefronts containing up to three orders of Zernike
polynomials excluding the piston and the two tilt terms. On
average, the peak-to-valley of the randomly simulated aberra-
tions is 2.3 µm and the chosen modulation amplitude is 0.5 mm
(sensitivity improves with smaller modulation and the lower
limit is given by the number of iterations for convergence). The
modulation amplitude is progressively reduced with increasing
iteration number to avoid over-estimation. The performance of
the virtual PWS is compared with virtual HS [4] and Fourier [24]
phase unwrapping methods in the presence of noise. In the case
of the virtual HS, reconstruction was performed with 20 × 20
subapertures and 32 × 32 pixels were used for centroiding the
Fig. 4. Simulations: (a) wavefronts measured with a commer- virtual HS focal spots. An identical slope sampling of 20 × 20
cial HS wavefront sensor; (b) simulated wrapped phase (radi- was used in the case of the virtual PWS for comparison. This
ans: -π to π) in a PS PDI; (c) wavefront reconstructed with the sampling was chosen to compare against the optimal sampling
virtual PWS and (d) residual wavefront error. in a virtual HS when sensing aberrations with a digital PS PDI
 Research Article Applied Optics 4

[4]. At SNR = 10 dB, the Fourier phase unwrapping algorithm diameter is a larger matrix. This makes the virtual HS method
fails [4] and the virtual PWS is evidently superior and saturates slow and computationally challenging for low SNR. The virtual
at a higher Strehl ratio value in comparison with the virtual HS PWS is relatively robust and the 4 pupil intensity images enable
for cases of 10 dB and 5 dB SNR (Fig. 6), illustrating the potential a direct evaluation of global wavefront slopes and does not need
of the virtual PWS at very low SNR. The error bars for each itera- local wavefront slope estimation that may involve additional
tion shown in Fig. 6 indicate the standard deviation of the Strehl errors as is the case with the virtual HS. The size of the pupils
ratio values of ten independently generated random wavefronts. can be adjusted to the number of slope measurements needed
Likewise, in the case of high-order aberrations and their random for phase unwrapping. And, by controlling the apex angle of
linear combinations, the virtual PWS consistently converged to the pyramidal prism, the pupils can be placed right next to one
higher Strehl ratios in the presence of random noise. another in a 2 × 2 grid. For instance, to compute 31 × 31 slope
values, an intensity matrix, I pyr with 62 × 62 pixels is sufficient
to estimate the wavefronts. In comparison, a virtual HS uses
∼ 20 × 20 pixels per subaperture and to obtain 31 × 31 slope
values, it requires to compute an intensity matrix of dimension
620 × 620 pixels, ten times larger than the matrix needed in a
virtual PWS. In the absence of noise, the virtual HS and virtual
PWS gave similar results with minor differences in the residual
wavefronts.
Modulation amplitude has an important role in controlling
the dynamic range and sensitivity of a pyramid wavefront sensor
[26]. It was shown earlier that the optimal modulation ampli-
tude increases in the presence of noise [27]. The same can be
noted in Fig. 3. In addition, a decrease in the magnitude of the
Fig. 6. Comparison of phase unwrapping methods at (a) aberrations leads to a decrease in the optimal modulation ampli-
10 dB and (b) 5 dB while sensing ten randomly simulated tude [16] and hence the modulation amplitude is premeditatedly
wavefronts. Corresponding sample wrapped phase maps reduced with increasing iterations as the residual decreases. In
are shown within the plots. practical situations, optimal modulation can be obtained by min-
imizing the difference between the wrapped phase image and
The method was also tested on a randomly simulated atmo- the rewrapped image obtained from the reconstructed wave-
spheric turbulence phase screen for a 1 m class telescope on a 125 front. The accuracy can be improved further by increasing the
x 125 grid [16] as shown in Fig. 7(a). Its high spatial frequency number of facets in a pyramid wavefront sensor [27]. In ad-
components result in a challenging wrapped phase shown in dition, increasing the number of phase-shifts in a PS PDI, the
Fig. 7(b). The reconstructed wavefront (Fig. 7(c)) obtained by measurement noise can be reduced. The conclusions derived
applying the virtual PWS resulted in a residual (Fig. 7(d)) with a with Strehl ratio as an evaluation metric in Figs. 5 and 6 were
Strehl ratio of 0.88 and RMS error of 0.05 µm after four iterations not affected when RMS error is used.
when 61 × 61 slope sampling is used. Here, Zernike decompo-
The virtuality of the PWS that is proposed here is unaffected
sition is not applied to avoid eliminating vital high-frequency
by the practical limitations of a physical PWS including the need
information.
for moving parts to achieve modulation, the precise alignment
of the tip of the pyramidal prism at the focus of a lens and
manufacturing defects, all of which can adversely affect recovery
of the aberrations. There exist non-moving solutions for closed-
loop operation [28]. However, they need several iterations to
achieve diffraction-limited point spread function [16].
In conclusion, a novel method of phase unwrapping called
the virtual PWS has been proposed. Although the virtual PWS is
tested on wrapped phase images in a PS PDI, the efficacy of the
Fig. 7. (a) Simulated turbulence phase screen (b) Wrapped method is not limited to this application. It can be easily adapted
phase corresponding to (a) in the interval [-π π]; (c) recon- to any complex phase unwrapping application including digital
structed with the virtual PWS (d) residual wavefront error. holographic microscopy, sample motion-detection in spectral do-
main optical coherence tomography and other medical imaging
modalities such as magnetic resonance elastography.

4. DISCUSSION
Funding. The research leading to these results has received
Focal spot centroiding influences the reconstruction accuracy of funding from Consejería de Educación, Juventud y Deporte
HS wavefront sensors [25]. The centroiding errors arising from of Comunidad de Madrid and the People Programme (Marie
noise in the wrapped images lead to inaccuracies in the virtual Curie Actions) of the European Union’s Seventh Framework Pro-
HS method [4] and the estimated Strehl ratio drops with increas- gramme (FP7/2007-2013) under REA grant agreement n◦ 291820
ing noise [14]. Furthermore, the total number of subapertures to VA; Science Foundation Ireland (grants: 07/SK/B1239a
needs to be increased to compensate for a decrease in SNR and and 08/IN.1/B2053); University College Dublin (seed funding:
in order to retain the same number of pixels per subaperture SF665) to BV; ERC Grant Agreement [ERC-2011-AdG 294099]
(and not compromise centroiding accuracy), the effective pupil and Spanish Government Grant FIS2014-56643-R to SM.
 Research Article Applied Optics 5

REFERENCES 25. V. Akondi, R. M. Basavaraju and R. P. Budihal, “Centroid detection

by Gaussian pattern matching in adaptive optics,” Int. J. Comput. Appl.
1. R. M. Goldstein, H. A. Zebker and C. L. Werner, “Satellite radar interfer- 26(1), 30–35 (2010).
ometry: Two-dimensional phase unwrapping,” Radio Sci. 23(4), 713–720 26. A. Burvall, E. Daly, S. Chamot, and C. Dainty, “Linearity of the pyramid
(1988). wavefront sensor,” Opt. Express 14(25), 11925–11934 (2006).
2. D. L. Fried, “Adaptive optics wave function reconstruction and phase 27. V. Akondi, S. Castillo and B. Vohnsen, “Multi-faceted digital pyramid
unwrapping when branch points are present,” Opt. Commun. 15, 43–72 wavefront sensor,” Opt. Commun. 323, 77–86 (2014).
(2001). 28. R. Ragazzoni, E. Diolaiti, and E. Vernet, “A pyramid wavefront sensor
3. B. J. Vakoc, S. H. Yun, J. F. de Boer, G. J. Tearney and B. E. Bouma, with no dynamic modulation,” Opt. Commun. 208, 51–60 (2002).
“Phase-resolved optical frequency domain imaging," Opt. Express 13(14),
5483–5493 (2005).
4. V. Akondi, C. Falldorf, S. Marcos and B. Vohnsen, “Phase unwrapping
with a virtual Hartmann-Shack wavefront sensor,” Opt. express 23(20),
25425–25439 (2015).
5. M. A. Navarro, J. C. Estrada, M. Servin, J. A. Quiroga and Javier Vargas,
“Fast two-dimensional simultaneous phase unwrapping and low-pass
filtering,” Opt. Express 20(3), 2556–2561 (2012).
6. S. Heshmat, S. Tomioka and S. Nishiyama, “Performance evaluation
of phase unwrapping algorithms for noisy phase measurements,” Int. J.
Optomechatronics, 8(4), 260–274 (2014).
7. M. A. Herráez, D. R. Burton, M. J. Lalor, and D. B. Clegg, “Robust,
simple, and fast algorithm for phase unwrapping," Appl. Opt. 35(29),
5847–5852 (1996).
8. J. M. Huntley, “Noise-immune phase unwrapping algorithm,” Appl. Opt.
28(16), 3268–3270 (1989).
9. J. M. Huntley and H. Saldner, “Temporal phase unwrapping algorithm
for automated interferogram analysis,” Appl. Opt. 32(17), 3047–3052
(1993).
10. J. A. Quiroga and E. Bernabeu, “Phase-unwrapping algorithm for noisy
phase-map processing," Appl. Opt. 33(29), 6725–6731 (1994).
11. R. Cusack, J. M. Huntley, and H. T. Goldrein, “Improved noise-immune
phase-unwrapping algorithm," Appl. Opt. 34(5), 781–789 (1995).
12. J-S. Lee, K. P. Papathanassiou, T. L. Ainsworth, M. R. Grunes, and
A. Reigber, “A new technique for noise filtering of SAR interferometric
phase images,” IEEE Trans. Geosci. Remote Sens. 36(5), 1456–1465
(1998).
13. R. M. Basavaraju, V. Akondi, S. J. Weddell and R. P. Budihal, “Myopic
aberrations: Simulation based comparison of curvature and Hartmann
Shack wavefront sensors,” Opt. Commun. 312, 23–30 (2014).
14. V. Akondi and B. Vohnsen, “Myopic aberrations: impact of centroiding
noise in Hartmann Shack wavefront sensing,” Ophthalmic Physiol. Opt.
33(4), 434–443 (2013).
15. R. Ragazzoni, “Pupil plane wavefront sensing with an oscillating prism,”
J. Mod. Opt. 43(2), 289–293 (1996).
16. V. Akondi, S. Castillo and B. Vohnsen, “Digital pyramid wavefront sensor
with tunable modulation," Opt. Express 21(15), 18261–18272 (2013).
17. V. Akondi, A. R. Jewel and B. Vohnsen, “Digital phase-shifting point
diffraction interferometer,” Opt. Letts. 39(6), 1641–1644 (2014).
18. V. Korkiakoski, C. Vèrinaud, M. Le Louarn, and R. Conan, “Comparison
between a model-based and a conventional pyramid sensor reconstruc-
tor,” Appl. Opt. 46, 6176-6184 (2007).
19. C. Vèrinaud, “On the nature of the measurements provided by a pyramid
wave-front sensor,” Optics Communications 233, 27–38 (2004).
20. V. Akondi, A. R. Jewel, and B. Vohnsen, “Closed-loop adaptive optics
using a spatial light modulator for sensing and compensating of optical
aberrations in ophthalmic applications,” J. Biomed. Opt. 19(9), 096014
(2014).
21. T. J. Gould, D. Burke, J. Bewersdorf, and M. J. Booth, “Adaptive optics
enables 3D STED microscopy in aberrating specimens,” Opt. Express
20(19), 20998–21009 (2012).
22. B. L. Ellerbroek, “First-order performance evaluation of adaptive-optics
systems for atmospheric-turbulence compensation in extended-field-
of-view astronomical telescopes,” J. Opt. Soc. Am. A 11(2), 783-805
(1994).
23. W. H. Southwell, “Wave-front estimation from wave-front slope mea-
surements," J. Opt. Soc. Am. 70(8), 998–1006 (1980).
24. M. D. Pritt and J. S. Shipman, “Least-squares two-dimensional phase
unwrapping using FFT’s,” IEEE Trans. Geosci. Remote Sens 32(3), 706–
708 (1994).