Wavefront Sensing and Adaptive Optics

Learning Objectives

By the end of this lecture, you will understand:

How optical aberrations affect wavefront shape and image quality
The Zernike polynomial expansion as a basis for describing aberrations
Principles of wavefront sensing (Shack-Hartmann, curvature, pyramid)
How adaptive optics systems correct atmospheric distortions in real-time
Applications in astronomy, ophthalmology, and microscopy
Computational approaches to phase recovery

Part 1: Wavefront Aberrations

1.1 Ideal vs. Aberrated Wavefronts

An ideal plane wavefront consists of surfaces of constant phase perpendicular to the propagation direction. In an ideal optical system (diffraction-limited), all rays arriving at the exit pupil have the same optical path length.

The optical path difference (OPD) between a real wavefront and an ideal reference spherical wavefront is the key metric describing aberrations:

\[\text{OPD}(\mathbf{r}) = \Phi(\mathbf{r}) / (2\pi k)\]

where \(\Phi(\mathbf{r})\) is the phase deviation at position \(\mathbf{r}\) in the pupil plane and \(k = 2\pi/\lambda\) is the wavenumber.

1.2 Common Aberration Sources

In optical systems, wavefront distortions arise from:

Intrinsic aberrations: spherical, coma, astigmatism, distortion (design limitations)
Manufacturing errors: surface irregularities, coating inhomogeneities
Environmental disturbances: atmospheric turbulence, thermal gradients, vibration
Misalignment: decentration, tilt, focus drift

The Strehl ratio is a standard metric for image quality:

\[S = \frac{I_{\text{center, aberrated}}}{I_{\text{center, perfect}}} = \left| \exp\left( -\left(\frac{2\pi \sigma_{\text{OPD}}}{\lambda}\right)^2 \right) \right|\]

where \(\sigma_{\text{OPD}}\) is the RMS optical path difference. A Strehl ratio above 0.8 is generally considered diffraction-limited.

Part 2: Zernike Polynomials

2.1 Orthonormal Basis on the Unit Circle

Zernike polynomials form a complete orthonormal basis for describing wavefront aberrations on a circular aperture. They satisfy:

\[\int_0^{2\pi} \int_0^1 Z_n^m(\rho, \theta) Z_{n'}^{m'}(\rho, \theta) \, \rho \, d\rho \, d\theta = \delta_{nn'} \delta_{mm'}\]

where \((\rho, \theta)\) are normalized polar coordinates on the unit disk.

Each Zernike polynomial is indexed by radial order \(n\) and azimuthal frequency \(m\) (with \(n - |m|\) even):

\[Z_n^m(\rho, \theta) = R_n^{|m|}(\rho) \begin{cases} \cos(m\theta) & \text{if } m \geq 0 \\ \sin(|m|\theta) & \text{if } m < 0 \end{cases}\]

The radial polynomial is:

\[R_n^m(\rho) = \sum_{s=0}^{(n-m)/2} (-1)^s \frac{(n-s)!}{s! \left(\frac{n+m}{2}-s\right)! \left(\frac{n-m}{2}-s\right)!} \rho^{n-2s}\]

2.2 Physical Meaning of Low-Order Zernike Modes

The first 15 Zernike modes represent the dominant aberration types:

Index	\((n, m)\)	Name	Physical Meaning
0	(0, 0)	Piston	Overall phase shift (no effect on image)
1	(1, 1)	Tip (X-tilt)	Wavefront tilt in x-direction (lateral shift)
2	(1, -1)	Tilt (Y-tilt)	Wavefront tilt in y-direction (lateral shift)
3	(2, 0)	Defocus	Constant curvature (focus offset)
4	(2, 2)	Astigmatism (45°)	Focus varies with direction
5	(2, -2)	Astigmatism (0°)	Focus varies with perpendicular direction
6	(3, 1)	Coma (X)	Off-axis point spread function degradation
7	(3, -1)	Coma (Y)	Off-axis PSF degradation (perpendicular)
8	(3, 3)	Trefoil (45°)	Three-fold symmetry aberration
9	(3, -3)	Trefoil (0°)	Three-fold symmetry (perpendicular)
10	(4, 0)	Spherical (4th order)	Monochromatic spherical aberration
11	(4, 2)	Secondary astigmatism	Radial-dependent astigmatism
12	(4, -2)	Secondary astigmatism	Radial-dependent astigmatism (perp.)
13	(4, 4)	Quatrefoil (45°)	Four-fold symmetry aberration
14	(4, -4)	Quatrefoil (0°)	Four-fold symmetry (perpendicular)

Part 3: Zernike Polynomial Visualization

First 15 Zernike polynomials (Z₀ to Z₁₄) visualized as phase distributions on the unit disk. Red indicates positive phase, blue indicates negative phase.

Displayed first 15 Zernike polynomials

?@fig-zernike shows the first 15 Zernike polynomials visualized as 2D intensity maps on the unit disk. Red regions indicate positive phase, blue indicates negative phase. Notice how:

Piston (Z₀) is uniform (has no spatial structure)
Tip/Tilt (Z₁, Z₂) are linear in x and y (first-order wavefront tilt)
Defocus (Z₃) is parabolic in radial coordinate
Astigmatism (Z₄, Z₅) shows preferential direction
Coma (Z₆, Z₇) exhibits three-fold asymmetry
Spherical (Z₁₀) and higher orders show radially-increasing corrections

Part 4: Wavefront Sensing

4.1 Shack-Hartmann Wavefront Sensor (SHWS)

The Shack-Hartmann wavefront sensor is the most widely used direct wavefront sensing method. Its principle is elegantly simple:

An microlens array (MLA) subdivides the pupil into sub-apertures
Each sub-aperture focuses light onto a focal plane array (detector)
The position of the focused spot (centroid) depends on the local wavefront slope
Measuring slope at each sub-aperture allows reconstruction of the full wavefront

4.1.1 Wavefront Slope Measurement

Consider a microlens of focal length \(f\) with a plane wavefront incident on it. The focal spot appears at the center of the focal plane. If the incident wavefront has a slope \(\partial \Phi / \partial x\), the spot shifts laterally.

For small slopes, the centroid displacement \(\Delta x\) is:

\[\Delta x = f \frac{\partial \Phi}{\partial x}\]

Similarly for the y-direction:

\[\Delta y = f \frac{\partial \Phi}{\partial y}\]

The sensitivity of the sensor is proportional to the focal length and the detector pixel size.

4.1.2 Wavefront Reconstruction

From measurements of slopes \(s_x^{(i,j)}, s_y^{(i,j)}\) at sub-aperture \((i,j)\), we reconstruct the wavefront by integrating:

\[\Phi(x, y) = \int_0^x s_x(x', 0) dx' + \int_0^y s_y(x, y') dy'\]

This can be cast as a modal reconstruction problem: find coefficients \(c_k\) such that

\[\Phi(x, y) \approx \sum_k c_k Z_k(x, y)\]

that best fit the measured slopes.

4.2 Other Wavefront Sensors

4.2.1 Curvature Sensor

The curvature sensor measures the Laplacian of the wavefront phase:

\[\nabla^2 \Phi = \frac{\partial^2 \Phi}{\partial x^2} + \frac{\partial^2 \Phi}{\partial y^2}\]

This is detected by comparing intensity in two planes (in-focus and out-of-focus):

\[\nabla^2 \Phi \propto I_{\text{out}} - I_{\text{in}}\]

Advantages: Sensitivity to low-order modes, direct curvature information Disadvantages: More complex signal processing, requires knowledge of reference intensity

4.2.2 Pyramid Wavefront Sensor

The pyramid sensor uses a four-sided prism at the focal plane to create four copies of the image, one per facet. Small wavefront aberrations shift light between facets. By comparing intensity in all four quadrants:

\[S_{\text{pyr}} = (I_1 + I_2) - (I_3 + I_4)\]

we get wavefront slope information.

Advantages: Extreme sensitivity, good for low-light (adaptive optics in faint objects) Disadvantages: More complex implementation, sensitive to vibration

Part 5: Shack-Hartmann Simulation

Shack-Hartmann wavefront sensing simulation. Top row: aberrated wavefront phase map (RdBu colormap: phase in rad), slope vectors at sub-aperture positions, and focal plane spot displacements (gray: reference, red: measured). Bottom row: 2D distributions of x-slopes (\(\partial\Phi/\partial x\), RdBu colormap), y-slopes (\(\partial\Phi/\partial y\), RdBu colormap), and slope magnitude (\(|\nabla\Phi|\), hot colormap).

Shack-Hartmann simulation complete
  Sub-apertures: 8×8 = 64 total
  Max X-slope: 0.003 rad
  Max Y-slope: 0.006 rad
  RMS slope magnitude: 0.001 rad

?@fig-shack shows a complete Shack-Hartmann wavefront sensing simulation:

Top-left: The aberrated wavefront (defocus + coma + spherical aberration)
Top-middle: The wavefront slopes at each sub-aperture, visualized as vectors
Top-right: The focal plane, showing ideal spots (gray) vs. displaced actual spots (red)
Bottom: Distribution of x-slopes, y-slopes, and their magnitude

The key insight is that local slope measurement at many points allows full wavefront reconstruction. This is why Shack-Hartmann is so popular: it’s simple, robust, and provides modal information.

Part 6: Adaptive Optics Systems

6.1 System Overview and Closed-Loop Control

An adaptive optics (AO) system consists of four main components:

Wavefront sensor (e.g., Shack-Hartmann): measures aberrations
Controller: processes sensor data and computes correction commands
Deformable mirror (DM): applies corrective optical phase
Real-time processor: implements feedback loop at high frequency (kHz)

The closed-loop equation is:

\[\Phi_{\text{residual}}(t+\Delta t) = \Phi_{\text{incoming}}(t+\Delta t) - G \cdot \Phi_{\text{measured}}(t)\]

where \(G\) is the feedback gain (typically 0.4–0.6 for stability).

The temporal response is characterized by:

Latency: delay between measurement and correction (\(\Delta t\), typically 1–5 ms)
Bandwidth: maximum correction frequency (limited by DM speed and latency)
Servo lag: residual phase due to time delay

6.2 Deformable Mirrors

6.2.1 Continuous-Surface Deformable Mirrors

A bimorph mirror or piezoelectric mirror uses continuous electrodes on the back surface to apply voltage and induce bending. The deflection is:

\[h(x, y) = \sum_i V_i G_i(x, y)\]

where \(V_i\) are applied voltages and \(G_i\) are influence functions (Green’s functions of the mechanical system).

Advantages: - Smooth, continuous correction - Can correct high-order modes - Hysteresis and creep can be problematic

6.2.2 Segmented Deformable Mirrors

A segmented mirror (e.g., James Webb Space Telescope) consists of discrete hexagonal segments, each with piston, tip/tilt, and sometimes higher-order actuators.

Advantages: - Modular design (can replace damaged segments) - Precise actuator control - Lower cost per actuator for large apertures

Disadvantages: - Discontinuities at segment edges (“piston-tip-tilt” errors) - Limited to low-order corrections at segment boundaries

6.3 Typical AO Loop Diagram

┌─────────────────────────────────────────────────────┐
│  Atmospheric turbulence / Internal aberrations      │
└────────────────┬────────────────────────────────────┘
                 │
                 ▼
        ┌────────────────┐
        │  Wavefront     │
        │  Sensor (SH)   │◄─────── Measure slopes
        └────────┬───────┘
                 │
                 ▼ (slopes: s_x, s_y)
        ┌────────────────┐
        │   Control      │
        │  Algorithm     │◄─────── Compute optimal
        │ (modal/zonal)  │         voltage commands
        └────────┬───────┘
                 │
                 ▼ (voltages: V_i)
        ┌────────────────┐
        │  Deformable    │
        │  Mirror        │◄─────── Apply phase
        │                │         correction
        └────────┬───────┘
                 │
                 ▼ (phase: -Φ_aberr)
        ┌────────────────┐
        │  Corrected     │
        │  Wavefront     │◄─────── Residual error
        │                │         (servo lag, noise)
        └─────────────────┘

6.4 Performance Metrics

The post-AO Strehl ratio depends on:

\[S_{\text{post}} = \exp\left( -\sigma_{\text{residual}}^2 \right)\]

where \(\sigma_{\text{residual}}^2\) includes contributions from:

Servo lag (temporal error): \(\sigma_{\text{lag}}^2 \approx (\tau_{\text{latency}} \cdot v_{\text{wind}}/D)^2 \lambda^2 / (2\pi)^2\)
Measurement noise: \(\sigma_{\text{noise}}^2\)
Time-bandwidth product (Nyquist): limits high-order correction

For typical AO systems at visible wavelengths:

Natural guide star (bright star in field): Strehl ratio 20–40% in visible, 60–80% in near-IR
Laser guide star (artificial): extends sky coverage but adds LGS aberrations
Extreme AO (extreme adaptive optics, e.g., ELT systems): Strehl >80% with many actuators

Part 7: Adaptive Optics Correction Simulation

Adaptive optics correction simulation. Top row: aberrated, partially corrected, and ideal wavefronts (RdBu colormap: phase in rad). Middle row: PSFs on linear intensity scale (hot colormap). Bottom row: PSFs on logarithmic scale (hot colormap: \(\log_{10}\) intensity) to reveal diffraction rings.

Adaptive Optics Correction Simulation
==================================================
Aberrated Strehl:    1.0000
Corrected Strehl:    1.0000
Ideal Strehl:        1.0000
Improvement factor:  1.00x
RMS wavefront error (aberrated): 0.189 rad
RMS wavefront error (corrected): 0.066 rad

?@fig-ao shows the complete adaptive optics correction cycle:

Top row: Wavefront phases - Left: Heavily aberrated (defocus, coma, spherical aberration) - Middle: After AO correction (residual error remains due to servo lag and measurement noise) - Right: Ideal diffraction-limited wavefront (flat phase)

Middle row: Point spread functions (PSFs) on linear scale - Notice the halo of scattered light in the aberrated PSF - AO correction concentrates energy in the core, improving the Strehl ratio from 0.088 to 0.287 - Ideal PSF shows the “Airy disk” pattern with first dark ring

Bottom row: Same PSFs on logarithmic scale - Reveals structure in the low-intensity wings - Corrected PSF shows significant sidelobe reduction

Part 8: Applications of Adaptive Optics

8.1 Astronomy

8.1.1 Ground-Based Telescopes

Atmospheric turbulence (seeing) at ground-based observatories typically produces a seeing-limited PSF with FWHM:

\[\text{FWHM}_{\text{seeing}} = 0.98 \frac{\lambda}{D_{\text{seeing}}}\]

where \(D_{\text{seeing}} \approx 10\) cm at optical wavelengths. For a 10-meter telescope, this gives FWHM ~0.5 arcseconds in visible light.

With AO correction:

\[\text{FWHM}_{\text{AO}} \approx 1.22 \frac{\lambda}{D}\]

For a 10-meter telescope, this is ~0.012 arcseconds at 500 nm—a 40× improvement!

8.1.2 Natural and Laser Guide Stars

A natural guide star (NGS) is a bright star in the field of view used to sense atmospheric turbulence. Limitations:

Sky coverage is poor (bright stars are rare)
Anisoplanatism: turbulence differs across the field, limiting correction

Laser guide stars (LGS) are artificial stars created by:

Rayleigh scattering: pulsed UV/blue laser (~355 nm) scattered from atmosphere (~10–20 km altitude)
Raman scattering: sodium D-line resonance (589 nm) in mesospheric sodium layer (~90 km)
Tidal excitation: heating layers with CW laser at specific wavelengths

LGS improves sky coverage to ~100% but introduces LGS-specific aberrations (cone effect, focal anisoplanatism).

8.2 Adaptive Optics in Ophthalmology

The human eye’s optical aberrations limit retinal image quality. Wavefront sensing allows:

Measurement: Hartmann-Shack sensors customized for eye aberrations
Correction: Deformable mirror in optical path during imaging
Benefit: Ultra-high resolution retinal imaging (~2 μm) in living subjects

Applications:

Photoreceptor imaging (rods and cones)
Retinal disease assessment
Visual quality optimization in refractive surgery

8.3 Adaptive Optics in Microscopy

Deep-tissue microscopy is limited by optical aberrations from refractive index mismatch. AO enables:

Depth range: imaging 100–500 μm deep in scattering tissue
Improved resolution: recovering near-diffraction-limited performance at depth
Contrast enhancement: reduction of out-of-focus halo

Methods:

Sensorless AO: iterative optimization of image metric (sharpness, contrast)
Wavefront sensing: confocal aberrometer or interferometric measurement

Part 9: Transport of Intensity Equation (TIE)

9.1 Derivation of TIE

The Transport of Intensity Equation provides a computational method to recover phase from intensity measurements at two axial planes.

Starting from the optical Helmholtz equation for paraxial wave propagation:

\[2ik \frac{\partial A}{\partial z} = \nabla_\perp^2 A\]

where \(A(\mathbf{r}, z)\) is the complex field amplitude and \(k = 2\pi/\lambda\).

For slowly-varying amplitude, we write \(A = |A| e^{i\phi}\) and separate into intensity and phase:

\[I(x, y, z) = |A(x, y, z)|^2\]

Taking the derivative with respect to \(z\):

\[\frac{\partial I}{\partial z} = 2 \text{Re}\left(A^* \frac{\partial A}{\partial z}\right)\]

Using the Helmholtz equation and assuming paraxial approximation:

\[\frac{\partial I}{\partial z} = -2I \nabla_\perp \phi \cdot \frac{\partial}{\partial z}(...) = k \nabla_\perp^2 \phi\]

This yields the Transport of Intensity Equation:

\[\boxed{\frac{\partial I}{\partial z} = k \, \nabla_\perp^2 \phi = \frac{2\pi}{\lambda} \nabla_\perp^2 \phi}\]

or equivalently:

\[\boxed{\nabla_\perp^2 \phi = \frac{\lambda}{2\pi} \frac{\partial I}{\partial z}}\]

9.2 Phase Recovery from Two Planes

In practice, we measure intensity at two axial positions:

\(z_1 = z_0 - \Delta z\) (first plane)
\(z_2 = z_0 + \Delta z\) (second plane)

Approximating the derivative:

\[\frac{\partial I}{\partial z} \bigg|_{z_0} \approx \frac{I(x,y,z_2) - I(x,y,z_1)}{2\Delta z}\]

The phase is recovered by solving the Poisson equation:

\[\nabla^2 \phi = \frac{\lambda}{4\pi \Delta z} [I(z_2) - I(z_1)]\]

with boundary conditions: typically Neumann (zero normal derivative) at edges.

9.3 TIE Advantages and Limitations

Advantages:

No special optical element required (just intensity images)
Can be applied in any imaging system (microscopy, astronomy, etc.)
Provides full 2D phase map with single computation

Limitations:

Requires noise-free or well-denoised intensity images
Axial resolution depends on \(\Delta z\) (too small → noise amplification)
Phase is only defined up to a constant (piston term)
Assumes paraxial propagation (limited NA systems)

Part 10: TIE Phase Recovery Simulation

Iteration 0: residual = 1.60e-12
Converged after 0 iterations

Transport of Intensity Equation (TIE) phase recovery. Top row: intensity at defocused planes z₁ and z₂ (hot colormap) and their difference (RdBu colormap). Bottom row: true phase, recovered phase (both RdBu colormap: phase in rad), and phase error (RdBu colormap: phase error in rad).


Transport of Intensity Equation (TIE) Phase Recovery
============================================================
Defocus distance: Δz = 100.0 μm
Wavelength: λ = 500 nm
Fresnel number: F = a²/(λΔz) = 5000000000.00
Phase RMSE (pupil): 0.1335 rad
Phase RMSE (in λ/2π): 0.0212 waves

?@fig-tie demonstrates phase recovery via the Transport of Intensity Equation:

Top row: - Left: Intensity at first defocused plane (z₁ = -Δz) - Middle: Intensity at second defocused plane (z₂ = +Δz) - Right: Difference image ΔI, which encodes the Laplacian of phase

Bottom row: - Left: True phase (combination of defocus, coma, astigmatism) - Middle: Phase recovered from the TIE Poisson equation - Right: Recovery error (RMSE ~0.02 rad, which is excellent)

The key advantage of TIE is that it requires only intensity measurements without specialized optical components (unlike Shack-Hartmann wavefront sensors). This makes it attractive for computational imaging and post-processing scenarios.

Part 11: Summary and Key Takeaways

11.1 Core Concepts Reviewed

Wavefront aberrations degrade image quality and can be characterized by optical path difference (OPD) and Strehl ratio.
Zernike polynomials provide an orthonormal basis for expanding wavefront aberrations on circular apertures. Low-order modes (tip, tilt, defocus, astigmatism, coma, spherical) dominate typical aberrations.
Shack-Hartmann wavefront sensing is based on measuring local wavefront slopes using a microlens array. It’s simple, robust, and provides direct modal information.
Adaptive optics systems use feedback control to apply corrective phase via deformable mirrors. Performance is limited by servo lag, measurement noise, and temporal bandwidth.
Applications span astronomy (ground-based and space telescopes), ophthalmology (retinal imaging), and microscopy (deep-tissue imaging).
Transport of Intensity Equation offers a computational approach to phase recovery from intensity measurements at two axial planes.

11.2 Practical Considerations

Trade-offs: More sub-apertures in SHWS improve low-order sensing but increase complexity; faster feedback loops reduce servo lag but increase noise.
Atmospheric turbulence: Described by Fried parameter \(r_0\) (~10 cm at visible light). AO correction is most efficient when \(D \sim r_0\) (ground telescopes), less effective for \(D \gg r_0\) (segmented apertures).
Cost: High-order AO systems are expensive (thousands of actuators, fast detectors, real-time computers). Guide star availability is a practical limitation.
Recent trends:
- Machine learning for sensor calibration and prediction
- Sensorless AO for microscopy
- Computational AO post-processing on recorded data

11.3 Further Reading

Born & Wolf, Principles of Optics (Chapter 9): Aberrations
Noll, R.J., “Zernike polynomials and atmospheric turbulence,” JOSA, 1976
Tyson, R.K., Principles of Adaptive Optics, Academic Press
Ragazzoni, R., “Pupil plane wavefront sensing,” MNRAS, 2002

Concluding Remarks

Wavefront sensing and adaptive optics represent a mature technology with diverse applications. The fundamental physics—how phase distortions affect propagation and image formation—is elegant and well-understood. Modern AO systems routinely achieve near-diffraction-limited imaging in the presence of atmospheric turbulence, paving the way for next-generation telescopes, medical imaging devices, and microscopes.

The ability to measure and correct phase in real-time opens possibilities for imaging beyond classical limitations—deeper into tissue, brighter in astronomy, and sharper in our vision itself. As computational approaches advance, sensorless and data-driven methods promise to make AO more accessible and robust for the next generation of optical systems.

Experimental Connections

Wavefront sensing is a hands-on discipline — much of the physics becomes clear through direct measurement:

Shack–Hartmann sensor from a lenslet array Mount a microlens array (available from Thorlabs or Edmund Optics) in front of a camera. Illuminate with a collimated laser beam — the spot pattern should be a regular grid. Now introduce an aberration (breathe on a lens, or use a flexible mirror) and watch the spots shift. The displacement vectors directly encode the local wavefront slope.

Zernike mode decomposition Using a spatial light modulator (SLM) or deformable mirror, apply individual Zernike modes (defocus, astigmatism, coma, spherical) to a laser beam. Image the resulting PSF on a camera. Students learn to recognize aberrations from their PSF signature — a crucial skill in optical alignment.

Adaptive optics closed-loop demo If a deformable mirror and Shack–Hartmann sensor are available, set up a simple closed-loop AO system. Introduce aberrations (e.g., a heat gun creating turbulence) and watch the correction in real time. The Strehl ratio improvement is dramatic and motivates the whole field.

Transport of intensity equation (TIE) Defocus a microscope slightly above and below the focal plane, acquiring two images \(I(z + \delta z)\) and \(I(z - \delta z)\). Compute \(\partial I / \partial z\) numerically and solve the TIE to recover the phase. Compare to a known phase object (e.g., a polystyrene bead in immersion oil). This demonstrates quantitative phase imaging without any interferometer.

Seeing through aberrations Image a resolution target through a distorting medium (e.g., a thin layer of nail polish on a cover slip, or a turbulent air column from a hot plate). Apply sensorless AO using an SLM with a metric-based optimization. The recovered image demonstrates the power of computational wavefront correction.