A Novel Model of Influence Function: Calibration of a Continuous Membrane Deformable Mirror

ACEEE Int. J. on Control System and Instrumentation, Vol. 03, No. 02, March 2012

A Novel Model of Influence Function: Calibration of a
Continuous Membrane Deformable Mirror
M. B. Roopashree1, Akondi Vyas1,2, and B. Raghavendra Prasad2
1
Indian Institute of Astrophysics, Bangalore, India
Email: brp@iiap.res.in
2
Indian Institute of Science, Bangalore, India

Abstract—M easurement and modeling of the influence measurements and the weighting associated with the basis
function plays a vital role in assessing the performance of the functions used to fit the wavefront shape on the DM, to
continuous MEMS deformable mirror (DM) for adaptive optics account for the actuator coupling [4].
applications. The influence function is represented in terms Obtaining a fit function for the measured influence
of Zernike polynomials and shown that the dominant modes
function for all the actuators individually can be useful in
for representation of central actuators of the DM are different
from those for the edge actuators. In this paper, a novel and accurately modeling the DM. Most researchers use a Gaussian
effective method of modeling the influence function for all model,
the DM actuators is proposed using the 2D sinc-squared  
IF ( x, y )  a  b exp  ( wx x 2  wy y 2 ) (1)
function.
Index Terms—deformable mirror, influence function, adaptive
which is not a true representation of the influence function.
optics The outer portion of the influence function is different from a
Gaussian as was pointed out by Wenhan Jiang, et. al. [5].
I. INTRODUCTION With azimuthal and radial modification to the Gaussian
function, the influence function can be modeled more
Deformable mirrors (DMs) are adaptive devices capable accurately [6]. But, it should be noted that the edge actuators
of correcting the dynamic effects on optical wavefronts due are poorly modeled by the earlier suggested models. J. H.
to random processes, in the likes of atmospheric turbulence. Lee, et. al [7] used the 1D sinc function to model the 1D
These are often used in adaptive optics systems and are influence function. In this paper, it is shown that the 2D sinc-
commonly controlled by the application of forces on actuators squared function and Zernike moments can be used to
that hold the mirror surface. Two kinds of DMs are available effectively model the influence function. The measurements
– segmented and continuous. Unlike the discrete actuators were made using the Shack Hartmann Wavefront Sensor
holding individual mirror segments in the segmented DM (SHWS) for all the actuators. The SHWS measures the local
case, the later uses a continuous membrane. Minimal edge slopes of the incident wavefront (in terms of the shift in the
diffraction in the case of a continuous DM makes it more focal spots of the microlenses from ideal positions) and the
desirable of the two. Since in most applications, the shape of the wavefront is calculated using the least square
wavefronts to be corrected by the DM are often much techniques.
smoother than the resolution limit of the device, it is In our study, the Multi-DM, manufactured by Boston
advantageous to use the continuous DM with precise Micromachine Corporation (BMC) was used. The following
calibration [1-3]. The effect of individual actuators is desired section discusses the method used for the measurement of
to be localized, but, in the case of a continuous DM, the the influence function. Evaluation of the inter-actuator
application of a certain force on one of the actuators results coupling is presented in section III. Section IV details the
in a finite displacement (that cannot be neglected) at the modeling of the influence function.
neighboring actuator location. This error due to non-localized
actuation can be largely compensated by determining the II. INFLUENCE FUNCTION MEASUREMENT
influence function, which is a measure of the response of the
DM surface when a known voltage (called poke voltage) is A. Experimental details
applied to a single DM actuator (a constant bias voltage is The optical arrangement for the measurement of the DM
applied to all the other actuators). The motive of the present influence function is shown in Fig. 1. A 15mW Melles Griot
work is to construct a mathematical model for the measured He-Ne laser was used as the source of light. The laser beam
influence function. was filtered using the spatial filter (S.F) and then collimated
Two other metrics used for the assessment of the DM with a doublet lens (CL). The beam was allowed to reflect
performance include, the inter-actuator coupling coefficient from the DM surface. The beam was then filtered using the 4f
and the interaction matrix. The inter-actuator coupling lens geometry (L1, L2 are equi-focal lenses, f=12.5cm). The
coefficient is defined as the ratio of the displacement of the assembled SHWS was made-up of a contiguous microlens
DM surface at the poked actuator to that obtained at its array (from Flexible Optical BV, APO-Q-P200-F40 at 633nm,
neighboring actuator. It is often practically convenient to with a focal length of f=40mm and a pitch of p=200µm) and a
employ the interaction matrix, that relates the wavefront sensor compact, high-resolution progressive monochrome CCD
© 2012 ACEEE 10
DOI: 01.IJCSI.03.02.49


camera (Pulnix TM-1325CL). The camera was placed at the B. Influence function calculation from SHWS slopes
focal plane of the microlens array (a distance f=4cm away). It The influence function was calculated from the SHWS
is important to note here that the performance of the SHWS measurements, assuming the slope sensing geometry of
is highly sensitive to placing the CCD at the focal plane of Southwell [10] and applying the least square technique. The
the microlenses [8]. measured influence functions for the edge, penultimate and
central actuators are shown in Fig. 3. It can be observed that
the influence functions for each type of actuator (central,
edge and penultimate) are different from one other. This is
attributed to the presence of the edge and penultimate
actuators near the fixed periphery. The systematic error that
can be observed on the background in Figs. 3a, 3b and 3c
can be remarkably reduced by taking the slope values,
measured when 0V is applied to all the actuators, as reference.
The background-error-free influence function measurements
Figure 1. Optical arrangement for the measurement of the DM are shown in Figs. 3d, 3e and 3f.
influence function
In view of the fact that the clear aperture on the DM is
4.9×4.9 mm2 and the pitch of the microlens array is 200µm,
25×25 microlenses were used to incorporate the entire active
region of the DM, for sensing. Also, the pitch of the CCD
pixels is 6.45µm, which implies that a single mircolens
corresponds to 31 pixels on the CCD, given the pitch of the
microlens array to be 200µm. For the reason that, the light
levels and frame rates are not the points of concern, the high
readout noise (50 e- per frame per pixel) is not an issue in our
case [9]. In contrary, the high resolution of the camera
provides high-quality, accurate wavefront sensing
capabilities. The maximum voltage that was applied to the
Figure 3. Measured influence function: A poke voltage of 30V is
DM actuators was limited to 40V to avoid SHWS cross-talk.
applied to actuators: (a, d) 7, edge actuator (b, e) 19, penultimate
The 140 actuators of the DM are numbered as shown in Fig. actuator (c, f) 67, center actuator, with zeros bias voltage. The
2 (when quoted, they are written in bold throughout the surfaces, complete wavefront reconstructions of the DM surface,
paper). In order to differentiate central actuators from the shown above are each normalized from zero to unity
ones at the edges, they are shaded in blue. The corners (1, It is to be remembered that there can be multiple errors
12, 133, 144), although numbered, are positions of zero involved while sensing the displacement of the DM surface.
activity and no actuator is present here (shaded black). The Apart from the experimental errors due to beam alignment
actuator positions shaded in green are called the penultimate and control, the only other error arises from the inaccuracies
actuators here. The behavior of these actuators will be soon in the wavefront sensing algorithm. The experimental errors
shown to be somewhat different from center and edge can be largely minimized by adopting cautious beam control
actuators. steps. The wavefront sensing algorithm is limited by the
wavefront sampling geometry, centroiding algorithm used to
detect the position of the SHWS spots and the method used
for reconstructing the DM surface from slope measurements.

III. EVALUATION OF THE INTER-ACTUATOR COUPLING
The inter-actuator coupling for all the central actuators
was evaluated. A histogram of the inter-actuator coupling
coefficient (computed for the nearest neighbor) at different
voltages for central actuators is shown in Fig. 4. It can be
noted here that with increasing poke voltage, the inter-
actuator coupling coefficient increases. The inter-actuator
coupling coefficient for the next nearest neighbors is very
low in the case of central actuators (see Fig. 5). The measured
Figure 2. Actuator Map: Blue region is the central portion and the
influence function shown in Fig. 3 hints a possible asymmetry
blackened actuators are the corner actuators, the uncolored region in the inter-actuator coupling which is pointed out more
shows the edge actuators and green shaded region depicts convincingly in Fig. 6 for the case of central actuators. Clearly,
penultimate actuators the actuator coupling is stronger in ‘x’ direction.

© 2012 ACEEE 11
DOI: 01.IJCSI.03.02.49


IV. MODELING THE MEASURED INFLUENCE FUNCTION
A. Fitting using Zernike polynomials
The measured influence function can also be exactly fitted
using the Zernike polynomials, Zj(x,y) via the calculation of
Zernike moments, aj’s:

IF ( x, y )   a j Z j ( x, y ) (4)
j 1

Since it is not practical to calculate infinite moments, the
influence function is approximated using ‘M’ number of finite
Zernike moments. The error involved in representing the
influence function using a finite number of Zernike moments
can be made sufficiently low by choosing a large enough
value for ‘M’. Influence function for actuator 66 represented
using M=75 Zernike moments is shown in Fig. 7. The residual
Figure 4. Inter-actuator coupling for the nearest neighbor at error (see Fig. 7c) is evaluated by subtracting the measured
different poke voltage. X-axis shows the inter-actuator coupling influence function (Fig. 7a) from the fitted model (Fig. 7b) for
and Y-axis shows the number of occurrences, analysis on central the influence function.
actuators alone

Figure 7. Fit result of actuator 66 with 75 Zernike polynomials
with application of a poke voltage, Vp=20V, zero bias (a) Measured
influence function (b) Fit function (c) Residual error
Zernike polynomials are defined on the unit disk and the
measured influence function is spread over a square grid.
This shape incompatibility compelled us to calculate the root
mean squared error (RMSE), that determines the fitting
accuracy, over one half of the square grid, close to the center
alone, excluding the circular boundary where the error is
Figure 5. Inter-actuator coupling coefficient for the next nearest exceedingly large. The value, log10(RMSE) reduces almost
neighbor at different poke voltage. X-axis shows the inter-actuator
coupling and Y-axis shows the number of occurrences, analysis on linearly with the number of computed Zernike moments up to
central actuators alone j ~ 40 as can be seen in Fig. 8. Beyond j > 40, the reduction
continues, but at a smaller rate. This is because the dominant
aberrations in the influence function are the lower order
Zernikes.
A plot of the Zernike moments for actuators 55, 67 and
77 is shown in Fig. 9. Here, it must be observed that Zernike
moments with index, j = 4, 12 and 24 dominate over other
Zernike moments for the central actuators. This is because
the measured influence functions for central actuators are
circularly symmetric, and so are the Zernike polynomials, j =
4, 12 and 24 (see Fig. 12). In the higher spatial frequency
range, j = 40, 60 appear to be more different from 0, than the
remaining higher order moments. The major drawback in
representing the influence function with Zernike polynomials
is that the wavefront sensing is performed on a square grid
and the representation is done on a unit disk, as is the
property of Zernike polynomials. This becomes even more
Figure 6. Histogram showing the asymmetry in the inter-actuator
coupling along x (marked left, right) and y (marked top, bottom)
significant while trying to represent an edge actuator.
directions

© 2012 ACEEE 12
DOI: 01.IJCSI.03.02.49


Figure 8. RMSE reduces with increasing number of Zernike
moments used for fitting: case of 20V applied to actuator 66
(central actuator), zero bias voltage
Figure. 12. Zernike polynomials that play vital role in IF
representa tion

B. Sinc-squared fit
From the observation of the line profile taken along the
X-axis of the measured influence function (IF) for actuator
79 as shown in Fig. 13, it can be concluded that the influence
function can be modeled as a sinc-squared function. This
novel influence function model can be very effectively used
to precisely represent the response of individual actuators.

Figure 9. Zernike coefficients on fitting IF for actuators 55, 67, 79
with a poke voltage of 25V, at zero bias
The edge actuator can be represented with very low residual
error using the Zernike moments as shown in Fig. 10. The
dominating moments in the case of edge actuators are j = 1, 5
and 6 (see Fig. 11, 12).

Figure 13. Line profile (1D view) of the measured influence
function
In the 2D case, a most general sinc-squared model of the
influence function model can be expressed,
Figure 10. Fit result of edge actuator, 7 with Zernike moments: case 2
of poke voltage of 25V, at zero bias. The residual error is nearly   x   y 
zero in the region where the fitting is done sinc  c x   sinc  c y 
s  s 
  x   y 
IF model ( x, y )  a0  a1  
  xy   (2)
 
  sinc s  c xy   d 
  xy  
or by neglecting the cross term, it can become,
2
  x   y  
IF model
( x, y )  a0  a1 sinc   c x   sinc   c y   d 
s  s  (3)

  x   y  

where, a0, a1, sx, cx, sy, cy, sxy, cxy and d are the fitting parameters.
Looking at the Eqs. (2) and (3), the assumed model for the
influence function certainly looks complicated with the need
to estimate multiple parameters, nine and seven in number
Figure 11. Zernike coefficients on fitting IF for actuators 6, 7, 8 respectively. The expression in Eq. (3) can be further simplified
with a poke voltage of 25V, at zero bias
by considering cx and cy to be identically equal to zero, in
those cases where this approximation to the influence
function remains valid. The trust-region reflective Newton
method (large-scale) was used for the optimization of the
fitting parameters. A central actuator, 78 was fitted using this
© 2012 ACEEE 13
DOI: 01.IJCSI.03.02.49


method and it was found that the residual error is very low CONCLUSIONS
(see Fig. 14). R-square, that measures the success of the fit
In this paper, a novel model for the measured influence
via the calculation of the correlation between the measured
function of a continuous DM is presented. The sinc-squared
influence function and the fitted model, is used as a metric
function can be used to closely model the influence function.
that determines the fit accuracy and in this case, it is 0.9901.
It is indicated that the behavior of the edge actuators and
The motivation behind favoring sinc-squared instead of sinc
penultimate actuators is not the same as that of the central
function is that with the former, the fitting root mean square
actuator which is made clear with the domination of different
error (RMSE) is lesser by a factor of 5, than that of the later. It
Zernike modes in each of the cases. The inter-actuator
was also noted that, in comparison with the Gaussian fitting,
coupling coefficient increases as the poke voltage increases.
[see Eq. (1)] the sinc-squared fitting [see Eq. (2)] improves
The most affected actuator is the nearest neighbor and the
the R-square value by a factor of 1.8. In the case of central
next nearest neighbor is affected a little too. There is
actuators, the cross term in Eq. (2) can further improve the
asymmetry in the inter-actuator coupling coefficient in
fitting accuracy, but the increase in fit accuracy is too small
different directions. Actuator modeling is better with sinc-
(R-square is greater by 0.02) and does not significantly affect
squared function due to the requirement of circular aperture
our modeling and hence may be ignored and Eq. (3) can be
for representation in terms of Zernike polynomials. The
used for simplicity.
current analysis will gain more significance when the DM is
The values of ‘x’ and ‘y’ were chosen to be positive
used in an optical layout for wavefront correction
integers alone, while using the sinc-squared model of
accommodating the modeled influence function.
influence function, in the case of edge actuators (see Fig.
15). It was observed that retaining the cross term in the case
REFERENCES
of edge actuators and penultimate actuators leads to
erroneous fitting and hence Eq. (3) was used for edge actuator [1] T. G. Bifano, J. Perreault, M. R. Krishnamoorthy, M. N.
modeling. Horenstein, “Microelectromechanical deformable mirrors,” IEEE
Journal of Selected Topics in Quantum Electronics, vol.5, no.1, pp.
83-89, Jan/Feb 1999.
[2] Dani Guzmán, Francisco Javier de Cos Juez, Fernando Sánchez
Lasheras, Richard Myers, and Laura Young, “Deformable mirror
model for open-loop adaptive optics using multivariate adaptive
regression splines,” Opt. Express, 18, 6492-6505 (2010).
[3] Michael Shaw, Simon Hall, Steven Knox, Richard Stevens,
and Carl Paterson, “Characterization of deformable mirrors for
Figure 14. Fit result of central actuator 78 with sinc functions with spherical aberration correction in optical sectioning microscopy, “
application of 20V, at zero bias Opt. Express, 18, 6900-6913 (2010).
[4] R. G. Lane and M. Tallon, “Wave-front reconstruction using a
Shack-Hartmann sensor,” Appl. Opt., 31, pp. 6902-6908, 1992.
[5] Wenhan Jiang, Ning Ling, Xuejun Rao and Fan Shi, “Fitting
capability of deformable mirror”, Proc. SPIE 1542, 130, 1991.
[6] Linhai Huang, Changhui Rao, and Wenhan Jiang, “Modified
Gaussian influence function of deformable mirror actuators,” Opt.
Express, 16, pp. 108-114, 2008.
[7] J. H. Lee, T. K. Uhm, S. K. Youn, “First-Order Analysis of
Figure 15. Fit result of edge actuator 7 with Eq. (3): case of poke Thin-Plate Deformable Mirrors,” Journal of the Korean Physical
voltage of 25V, at zero bias. Here, cx , c y take non-zero values Society, Vol. 44, No. 6, pp. 1412-1416, 2004.
[8] M. B. Roopashree, A. Vyas, and B. R. Prasad, “Automated
It was found that for the central actuators, the median of the
ROI selection and calibration of a microlens array using a MEMS
R-square value for fitting the sin-squared function in Eq. (2) CDM,” in Adaptive Optics: Methods, Analysis and Applications,
is just above 0.98. Also, for the central actuators, the medians OSA Technical Digest (CD) (Optical Society of America, 2011),
of a0, cx, cy, cxy are 0.0388, 0.0145, 0.0276 and -0.0049 and paper ATuA5.
hence may be approximated to zero. This observation allows [9] M. B. Roopashree, A. Vyas, and B. R. Prasad, “Experimental
us to reduce Eq. (2) into a 5-parameter optimization problem. evaluation of centroiding algorithms at different light intensity and
From our observations, the most probable model of the DM noise levels,” AIP Conf. Proc., to be published.
using the sinc-squared function as defined in Eq. (2) is: [10] W. H. Southwell, “Wave-front estimation from wave-front
slope measurements,” J. Opt. Soc. Am., 70, pp. 998-1006, 1980.
2
  x   y 
sinc 0.74   sinc 0.92 
   
IF model ( x, y )  0.63
  xy   (4)
  sinc   0.27 
  0.65  

© 2012 ACEEE 14
DOI: 01.IJCSI.03.02.49

A Novel Model of Influence Function: Calibration of a Continuous Membrane Deformable Mirror

More Related Content

What's hot (18)

Viewers also liked (7)

Similar to A Novel Model of Influence Function: Calibration of a Continuous Membrane Deformable Mirror (20)

More from IDES Editor (20)

Recently uploaded (20)

A Novel Model of Influence Function: Calibration of a Continuous Membrane Deformable Mirror