Use jitter

This commit is contained in:
Lukas Klass 2020-05-06 16:59:51 +02:00
parent c1f2e28775
commit fb2238b0b7

View File

@ -4,6 +4,7 @@ from astropy import units as u
from .IPSF import IPSF
from scipy.optimize import newton
from scipy.special import j0, j1
from scipy.signal import fftconvolve
from ...lib.helpers import error
@ -13,9 +14,28 @@ class Airy(IPSF):
"""
@u.quantity_input(wl="length", d_aperture="length")
def __init__(self, wl: u.Quantity, d_aperture: u.Quantity):
def __init__(self, f_number: float, wl: u.Quantity, d_aperture: u.Quantity, osf: float, pixel_size: u.Quantity):
"""
Initialize a new PSF from a airy disk.
Parameters
----------
f_number : float
The working focal number of the optical system
wl : Quantity
The central wavelength which is used for calculating the PSF
d_aperture : Quantity
The diameter of the telescope's aperture.
osf : float
The oversampling factor to be used for oversampling the PSF with regards to the pixel size.
pixel_size : Quantity
The size of a pixel as length-quantity.
"""
self.__f_number = f_number
self.__wl = wl
self.__d_aperture = d_aperture
self.__osf = osf
self.__pixel_size = pixel_size
def calcReducedObservationAngle(self, contained_energy: Union[str, int, float, u.Quantity],
jitter_sigma: u.Quantity = None) -> u.Quantity:
@ -44,9 +64,11 @@ class Airy(IPSF):
elif contained_energy.lower() == "fwhm":
# Width of the FWHM of the airy disk
reduced_observation_angle = 1.028
contained_energy = 0.4738 * u.dimensionless_unscaled
elif contained_energy.lower() == "min":
# Width of the first minimum of the airy disk
reduced_observation_angle = 1.22 * 2
contained_energy = 0.8377 * u.dimensionless_unscaled
else:
# Try to parse the encircled energy to float
reduced_observation_angle = 0
@ -54,15 +76,58 @@ class Airy(IPSF):
contained_energy = float(contained_energy) / 100.0 * u.dimensionless_unscaled
# Calculate the width numerically from the integral of the airy disk
# See also https://en.wikipedia.org/wiki/Airy_disk#Mathematical_formulation
reduced_observation_angle = 2 * newton(lambda x: 1 - j0(np.pi * x) ** 2 - j1(np.pi * x) ** 2 -
contained_energy, 1, tol=1e-6)
reduced_observation_angle = 2 * newton(
lambda x: 1 - j0(np.pi * x) ** 2 - j1(np.pi * x) ** 2 - contained_energy, 1, tol=1e-6)
except ValueError:
error("Could not convert encircled energy to float.")
elif type(contained_energy) == u.Quantity:
# Calculate the width numerically from the integral of the airy disk
reduced_observation_angle = 2 * newton(
lambda x: 1 - j0(np.pi * x) ** 2 - j1(np.pi * x) ** 2 - contained_energy.value, 1, tol=1e-6)
else:
# Calculate the width numerically from the integral of the airy disk
reduced_observation_angle = 2 * newton(lambda x: 1 - j0(np.pi * x) ** 2 - j1(np.pi * x) ** 2 -
contained_energy.value, 1, tol=1e-6)
contained_energy = contained_energy * u.dimensionless_unscaled
reduced_observation_angle = 2 * newton(
lambda x: 1 - j0(np.pi * x) ** 2 - j1(np.pi * x) ** 2 - contained_energy.value, 1, tol=1e-6)
if jitter_sigma is not None and type(contained_energy) == u.Quantity:
# Convert jitter to reduced observation angle in lambda / d_ap
jitter_sigma = jitter_sigma.to(u.rad).value * self.__d_aperture / self.__wl.to(u.m)
# Calculate necessary grid length to accommodate the psf and 3-sigma of the gaussian
grid_width = (reduced_observation_angle / 2 + 3 * jitter_sigma)
# Calculate the reduced observation angle of a single detector pixel
reduced_observation_angle_pixel = (self.__pixel_size / (
self.__f_number * self.__d_aperture) * self.__d_aperture / self.__wl)
# Calculate the width of each grid element
dx = reduced_observation_angle_pixel / self.__osf
# Calculate the necessary number of points on the grid
n_points = np.ceil(grid_width / dx).value
# Calculate the corresponding x-coordinates of each grid element
x = np.arange(1, n_points + 1) * dx
# Calculate the psf from an airy disk for each element on the grid
psf = (2 * j1(np.pi * x) / (np.pi * x))**2
# Calculate the integral of the undisturbed airy disk in order to scale teh result of the convolution
total = np.sum(psf * x) * dx * 2 * np.pi
# Mirror the PSF to the negative x-domain
psf = np.concatenate((np.flip(psf), np.array([1]), psf))
# Calculate a gaussian kernel
kernel = 1 / (2 * np.pi * jitter_sigma ** 2) * np.exp(
- np.concatenate((np.flip(x), np.array([0]), x)) ** 2 / (2 * jitter_sigma ** 2))
# Normalize the kernel
kernel = kernel / np.sum(kernel)
# Convolve the PSF with gaussian kernel
psf = fftconvolve(np.pad(psf, int(n_points), mode="constant", constant_values=0),
kernel, mode="same")
# Reduce the PSF to the positive x-domain
psf = psf[int((psf.shape[0] - 1) / 2):]
# Calculate the rolling integral of the PSF
psf_int = np.cumsum(psf * np.arange(psf.shape[0])) * reduced_observation_angle_pixel.value / self.__osf
# Scale the integral of the disturbed PSF equal to the undisturbed PSF
psf_int = psf_int / psf_int[-1] * total / (4 / np.pi)
# Calculate the reduced observation angle
reduced_observation_angle = np.argmax(
psf_int > contained_energy) * reduced_observation_angle_pixel.value / self.__osf * 2
return reduced_observation_angle * u.dimensionless_unscaled
def mapToGrid(self, grid: np.ndarray) -> np.ndarray: