def set_phase_map(self, z_coef, rho_zern=1., PV_rescale=True, PV_goal=50):
"""
Generates a phase map based on Zernike polynomials
This usually represents an NCPA map
Uses the fast methods implemented in zern_core.py
:param z_coef: coefficient of the Zernike series expansion
:param PV_goal: desired Peak-To-Valley [nm] of the phase map
"""
# Construct the base phase map
self.zern_model = zern.ZernikeNaive(mask=self.pupil_mask)
phase_map = self.zern_model(coef=z_coef,
rho=self.rho_m / rho_zern,
theta=self.theta_m,
normalize_noll=False,
mode='Jacobi',
print_option='Silent')
if PV_rescale:
# Compute the current PV and rescale the coefficients
current_pv = np.max(phase_map[np.nonzero(phase_map)]) - np.min(
phase_map[np.nonzero(phase_map)])
# Remember that at this point the phase map is a 1D array (a masked 2D)
phase_map = zern.rescale_phase_map(phase_map, peak=PV_goal / 2)
self.ncpa_coef = self.zern_model.coef * (PV_goal / current_pv
) # Save the coefficients
self.phase_map = zern.invert_mask(phase_map, self.pupil_mask)
else:
self.phase_map = zern.invert_mask(phase_map, self.pupil_mask)
self.ncpa_coef = self.zern_model.coef
def callback_function(self, coef):
"""
Callback to print intermediate results at each iteration
"""
cost_now = self.cost(coef)
self.cost_array.append(cost_now)
coef_copy = coef
coef_copy[0] = 0 #remove piston
print('\nAt iteration %d :' % self.counter)
r = self.rho_m * (self.rho_zern / self.rho_aper)
nominal_map = self.zern_model(coef=coef_copy,
rho=r,
theta=self.theta_m,
normalize_noll=False,
mode='Jacobi',
print_option='Silent')
nominal_map = zern.invert_mask(nominal_map, self.pupil_mask)
try:
p0 = self.true_phase
PV = compute_PV_2maps(phase_ref=self.true_phase,
phase_guess=nominal_map)
self.PV_array.append(PV)
RMS = compute_rms((self.true_phase - nominal_map)[self.pupil_mask])
self.RMS_array.append(RMS)
print('Merit Function: %.3E' % cost_now)
print('PV : %.3f' % PV)
print('RMS: %.3f' % RMS)
except AttributeError:
pass
# cost_at_iter = self.cost(coef)
# print 'Merit function = %e' %cost_at_iter
self.guesses[:, :, self.counter] = nominal_map
self.counter += 1
def rand_zern(randker):
N = 48
N_zern = 10
rho_max = 0.9
eps_rho = 1.4
randgen = RandomState(randker)
extents = [-1, 1, -1, 1]
# Construct the coordinates
x = np.linspace(-rho_max, rho_max, N)
rho_spacing = x[1] - x[0]
xx, yy = np.meshgrid(x, x)
rho = np.sqrt(xx ** 2 + yy ** 2)
theta = np.arctan2(xx, yy)
aperture_mask = rho <= eps_rho * rho_max
rho, theta = rho[aperture_mask], theta[aperture_mask]
rho_max = np.max(rho)
extends = [-rho_max, rho_max, -rho_max, rho_max]
# Compute the Zernike series
coef = randgen.normal(size=N_zern)
z = zern.ZernikeNaive(mask=aperture_mask)
phase_map = z(coef=coef, rho=rho, theta=theta, normalize_noll=False, mode='Jacobi', print_option='Silent')
phase_map = zern.rescale_phase_map(phase_map, peak=1)
phase_2d = zern.invert_mask(phase_map, aperture_mask)
return phase_2d
def set_true_phase(self, ncpa_coef):
r = self.rho_m * (self.rho_zern / self.rho_aper)
nominal_map = self.zern_model(coef=ncpa_coef,
rho=r,
theta=self.theta_m,
normalize_noll=False,
mode='Jacobi',
print_option='Silent')
self.true_phase = zern.invert_mask(nominal_map, self.pupil_mask)
def evaluate_phase(self, zern_coef):
_phase = self.zern_model(coef=zern_coef,
rho=self.rho,
theta=self.theta,
normalize_noll=False,
mode='Jacobi',
print_option='Silent')
phase = zern.invert_mask(_phase, self.aper_mask)
return phase
def evaluate_phase(self, zern_coef):
r = self.rho_m * (self.rho_zern / self.rho_aper)
nominal_map = self.zern_model(coef=zern_coef,
rho=r,
theta=self.theta_m,
normalize_noll=False,
mode='Jacobi',
print_option='Silent')
nominal_map = zern.invert_mask(nominal_map, self.pupil_mask)
# plt.figure()
# plt.imshow(nominal_map)
# plt.colorbar()
return nominal_map
def callback_function(self, coef):
"""
Callback to print intermediate results at each iteration
"""
cost_now = self.cost(coef)
self.cost_array.append(cost_now)
grad = self.grad_analytic(coef)
coef_copy = coef
print('\nAt iteration %d :' % self.counter)
print('Merit Function: %.3E' % cost_now)
print('Grad Norm: %e' % (np.linalg.norm(grad)))
nominal_map = self.zern_model(coef=coef_copy,
rho=self.rho,
theta=self.theta,
normalize_noll=False,
mode='Jacobi',
print_option='Silent')
nominal_map = zern.invert_mask(nominal_map, self.aper_mask)
self.guesses[:, :, self.counter] = nominal_map
self.counter += 1
def set_phase_diversity(self, n=2, m=0, rho_zern=1., ratio=10):
"""
Creates the Phase Diversity map which will be used to 'defocus'
the images. Although the common thing is to use a pure defocus term (n=2, m=0)
any Zernike polynomial is possible
The Phase Diversity map is rescaled according to a desired PV which is
'ratio' times the PV of the NCPA
"""
diversity = self.zern_model.Z_nm(n=n,
m=m,
rho=self.rho_m / rho_zern,
theta=self.theta_m,
normalize_noll=False,
mode='Jacobi')
pv_diversity = np.max(diversity) - np.min(diversity)
pv_phase = np.max(self.phase_map) - np.min(self.phase_map)
phase_diversity = (ratio * pv_phase) * (diversity / pv_diversity)
# Remember that at this point the phase diversity is a 1D array (a masked 2D)
self.phase_diversity = zern.invert_mask(phase_diversity,
self.pupil_mask)
def cost(self, zern_coef):
norm_pix = 1. / (self.N_pix)
r = self.rho_m * (self.rho_zern / self.rho_aper)
nominal_map = self.zern_model(coef=zern_coef,
rho=r,
theta=self.theta_m,
normalize_noll=False,
mode='Jacobi',
print_option='Silent')
nominal_map = zern.invert_mask(nominal_map, self.pupil_mask)
# Nominal Image
pupil_nominal = complex_function(self.pupil_mask, nominal_map,
self.wave)
propagated_nominal = norm_pix * fftshift(fft2(pupil_nominal))
image_nominal = self.t_exp * (np.abs(propagated_nominal))**2
J_nominal = (self.image_nominal - image_nominal)**2
# + Phase Diversity
pupil_plus_defocus = complex_function(
self.pupil_mask, (nominal_map + self.phase_diversity), self.wave)
propagated_plus_defocus = norm_pix * fftshift(fft2(pupil_plus_defocus))
image_plus_defocus = self.t_exp * (np.abs(propagated_plus_defocus))**2
J_plus = (self.image_plus_defocus - image_plus_defocus)**2
if self.N_images == 3:
# - Phase Diversity
pupil_minus_defocus = complex_function(
self.pupil_mask, (nominal_map - self.phase_diversity),
self.wave)
propagated_minus_defocus = norm_pix * fftshift(
fft2(pupil_minus_defocus))
image_minus_defocus = self.t_exp * (
np.abs(propagated_minus_defocus))**2
J_minus = (self.image_minus_defocus - image_minus_defocus)**2
return np.sum(J_nominal + J_plus + J_minus) / (self.t_exp**2)
else:
return np.sum(J_nominal + J_plus) / (self.t_exp**2)
rho, theta = rho[aperture_mask], theta[aperture_mask]
rho_max = np.max(rho)
extends = [-rho_max, rho_max, -rho_max, rho_max]
# Compute the Zernike series
coef = randgen.normal(size=N_zern)
z = zern.ZernikeNaive(mask=aperture_mask)
phase_map = z(coef=coef,
rho=rho,
theta=theta,
normalize_noll=False,
mode='Jacobi',
print_option=None)
# phase_map = zern.rescale_phase_map(phase_map, peak=1)
phase_2d = zern.invert_mask(phase_map, aperture_mask)
# Introduce some noise in the map
noised_phase_map = phase_map + 0.5 * np.random.normal(size=phase_map.shape[0])
noised_2d = zern.invert_mask(noised_phase_map, aperture_mask)
plt.figure()
plt.imshow(phase_2d, extent=extends, cmap='jet')
plt.title("Zernike Series (%d polynomials)" % N_zern)
plt.xlabel('x')
plt.ylabel('y')
plt.colorbar()
plt.figure()
plt.imshow(noised_2d, extent=extends, cmap='jet')
plt.title("Zernike Series with Noise")
def reshape_model_matrix(matrix, mask):
# Reshape the Zernike model matrix from flattened to square
H = [zern.invert_mask(matrix[:, i], mask) for i in range(matrix.shape[-1])]
H = np.array(H)
H_new = np.moveaxis(H, 0, -1)
return H_new
def grad_analytic(self, zern_coef):
"""
Analytic gradient for speed purposes
"""
grad_start = timer()
N_zern = zern_coef.shape[0]
wave_factor = 2 * np.pi / self.wave * 1j
norm_pix = 1. / (self.N_pix)
r = self.rho_m * (self.rho_zern / self.rho_aper)
nominal_map = self.zern_model(coef=zern_coef,
rho=r,
theta=self.theta_m,
normalize_noll=False,
mode='Jacobi',
print_option='Silent')
nominal_map = zern.invert_mask(nominal_map, self.pupil_mask)
# Nominal Image
pupil_nominal = complex_function(self.pupil_mask, nominal_map,
self.wave)
propagated_nominal = norm_pix * fftshift(fft2(pupil_nominal))
image_nominal = self.t_exp * (np.abs(propagated_nominal))**2
# + Phase Diversity
pupil_plus_defocus = complex_function(
self.pupil_mask, (nominal_map + self.phase_diversity), self.wave)
propagated_plus_defocus = norm_pix * fftshift(fft2(pupil_plus_defocus))
image_plus_defocus = self.t_exp * (np.abs(propagated_plus_defocus))**2
# - Phase Diversity
pupil_minus_defocus = complex_function(
self.pupil_mask, (nominal_map - self.phase_diversity), self.wave)
propagated_minus_defocus = norm_pix * fftshift(
fft2(pupil_minus_defocus))
image_minus_defocus = self.t_exp * (
np.abs(propagated_minus_defocus))**2
base_factor_nom = 2 * (self.image_nominal - image_nominal)
base_factor_plus = 2 * (self.image_plus_defocus - image_plus_defocus)
base_factor_minus = 2 * (self.image_minus_defocus -
image_minus_defocus)
# Helper stuff
Ec_nom, Ec_plus, Ec_minus = np.conj(pupil_nominal), np.conj(
pupil_plus_defocus), np.conj(pupil_minus_defocus)
FE_nom = norm_pix * fftshift(fft2(pupil_nominal))
FE_plus = norm_pix * fftshift(fft2(pupil_plus_defocus))
FE_minus = norm_pix * fftshift(fft2(pupil_minus_defocus))
FEc_nom, FEc_plus, FEc_minus = np.conj(FE_nom), np.conj(
FE_plus), np.conj(FE_minus)
# print('Common: %f sec' %(timer() - grad_start))
g = np.zeros(N_zern)
for k in range(N_zern):
Z_k = self.model_matrix[:, :, k]
fourier_factor_nom = self.helper_grad(Z_k, pupil_nominal, Ec_nom,
FE_nom, FEc_nom)
fourier_factor_plus = self.helper_grad(Z_k, pupil_plus_defocus,
Ec_plus, FE_plus, FEc_plus)
fourier_factor_minus = self.helper_grad(Z_k, pupil_minus_defocus,
Ec_minus, FE_minus,
FEc_minus)
grad_nom = np.sum(base_factor_nom * self.t_exp *
fourier_factor_nom)
grad_plus = np.sum(base_factor_plus * self.t_exp *
fourier_factor_plus)
grad_minus = np.sum(base_factor_minus * self.t_exp *
fourier_factor_minus)
g[k] = np.real(grad_nom + grad_plus + grad_minus) / (self.t_exp**2)
return g
print('\nML model guesses:')
print(guessed[0, :])
print('\nTrue values are :')
print(true_coef)
""" Compare the guess and the true NCPA """
_true_ncpa = np.dot(H_matrix, zern_coef)
guessed_coef = np.zeros_like(zern_coef)
guessed_coef[1:] = guessed
_guessed_ncpa = np.dot(H_matrix, guessed_coef)
_res_ncpa = _true_ncpa - _guessed_ncpa
PV = np.max(_res_ncpa) - np.min(_res_ncpa)
mu = np.mean(_res_ncpa)
nn = _res_ncpa.shape[0]
RMS = np.sqrt(1. / nn * np.sum((_res_ncpa - mu)**2))
true_ncpa = 1e3 * zern.invert_mask(_true_ncpa, pupil_mask)
guessed_ncpa = 1e3 * zern.invert_mask(_guessed_ncpa, pupil_mask)
res_ncpa = true_ncpa - guessed_ncpa
print('\nDefocus = %.1f [mm]' % defocus)
print('PV residual NPCA = %.1f [nm]' % (1e3 * PV))
print('RMS residual NPCA = %.1f [nm]\n' % (1e3 * RMS))
plt.figure()
plt.imshow(true_ncpa, cmap='jet')
plt.colorbar()
plt.title('True NCPA map [nm]')
plt.figure()
plt.imshow(guessed_ncpa, cmap='jet')
plt.colorbar()
plt.subplot(122)
plt.imshow(pop.crop_array(image, N_pix), extent=extends, cmap='jet')
plt.xlabel('X [mm]')
plt.title('Ideal PSF')
# Show an example of NCPA map
zern_model = zern.ZernikeNaive(mask=aper_mask)
zern_coef = np.zeros(10)
zern_coef[4] = 0.25
zz = zern_model(coef=zern_coef,
rho=rho,
theta=theta,
mode='Standard',
print_option=None)
phase = zern.invert_mask(zz, aper_mask)
H = zern_model.model_matrix
plt.figure()
plt.imshow(zern.invert_mask(H[:, 4], aper_mask), extent=[-1, 1, -1, 1])
plt.xlim([-1.1 * eps, 1.1 * eps])
plt.ylim([-1.1 * eps, 1.1 * eps])
plt.colorbar()
# Investigate why there's a difference in Cost
# when you use a slightly different defocus than
# the one Zemax says we used
j = []
focus = np.linspace(0.1, 0.2, 20)
rho, theta = rho[aperture_mask], theta[aperture_mask]
rho_max = np.max(rho)
extends = [-rho_max, rho_max, -rho_max, rho_max]
# Compute the Zernike series
coef = randgen.normal(size=N_zern)
z = zern.ZernikeNaive(mask=aperture_mask)
phase_map = z(coef=coef,
rho=rho,
theta=theta,
normalize_noll=False,
mode='Jacobi',
print_option=None)
phase_map = zern.rescale_phase_map(phase_map, peak=1)
phase_2d = zern.invert_mask(phase_map, aperture_mask)
plt.figure()
plt.imshow(phase_2d, extent=extends, cmap='jet')
plt.title("Zernike Series (%d polynomials)" % N_zern)
plt.xlabel('x')
plt.ylabel('y')
plt.colorbar()
# Compute the Power Spectral Density of the Zernike map
phase_f = fftshift(fft2(phase_2d))
power_f = (np.abs(phase_f) / N / N)**2
spatial_frequencies = fftfreq(N, d=rho_spacing)
freq_plus = spatial_frequencies[:N // 2]
# Slice the 2D result for all 4 sub-axis: +X, +Y, -X, -Y
