def __init__(self, nu, dnu, phi, weights=None):
"""
RMSynth(nu, dnu, phi, weights=None, isl2=False)
Initializes the RMSynth class. For an image, this needs only be done
once. Each LOS may be inverted individually using the
compute_dirty_image method.
Inputs:
nu- A vector containing frequency(Hz) values. Either way they
must be ordered from lowest to highest value.
dnu- Width of the frequency channels (in Hz).
phi- The requested phi axis. Use numpy.arange to construct.
weights-A vector containing weights. Must be the same length as nu.
Values should be between 0 and 1. If no weights vector is
given, it will be assumed that each value has weight 1.
Outputs:
None
"""
# parameters used for gridding
self.m = 6 # number of grid cells overwhich the GCF spans
self.alpha = 1.5 # oversampling ratio
self.dphi = phi[1] - phi[0]
self.nphi = len(phi)
self.dl2 = 1. / self.nphi / self.dphi / self.alpha
self.nl2 = self.alpha * self.nphi
stdout.write('RMSynth Initializing... ')
stdout.flush()
# equal weighting for all frequencies by default.
if weights is None:
weights = numpy.ones(len(nu))
if len(nu) != len(weights):
msg = 'The length of weight and frequency vectors must be equal!'
raise IndexError(msg)
# for now store the ungridded l2s (for diagnostics)
self.l2_nonuni = self.convert_nu_to_l2(nu, dnu)
self.weights_nonuni = numpy.flipud(weights)
# another gridding parameter, derived from m and alpha
self.beta = numpy.pi *\
numpy.sqrt((self.m / self.alpha) ** 2. *
(self.alpha - 0.5) ** 2 - 0.8)
self.l20 = self.compute_l20()
self.l2i = self.l2_nonuni[0] - self.m * self.dl2 * numpy.pi
# this axis is actually \lambda^2/\pi
self.l2 = numpy.arange(0, self.nl2, 1) * self.dl2 + self.l2i / numpy.pi
self.l2_beam = numpy.arange(0, self.nl2 * 2., 1) * self.dl2 / 2. \
+ self.l2i / numpy.pi
self.phi = -self.dphi * self.nphi * 0.5 \
+ numpy.arange(0, self.nphi, 1) * self.dphi
self.grid_corr = G.gridnorm(self.phi, self.dl2, self.m, self.beta)
self.tndx_phi = int(0.5 * self.nphi * (self.alpha - 1))
[self.rmsf, self.rmsf_phi] = self.compute_rmsf()
stdout.write('Complete.\n')
def compute_rmsf(self):
"""
RMSynth.compute_rmsf()
Inverts the given polarization vector to arrive at the dirty dispersion
function.
Inputs:
None
Outputs:
1- A map containing the RMSF. A numpy array of complex values.
2- A map containing the RMSF phi axis.
"""
# RMSF image size must be twice that of the df image for CLEANing
phi_center = 0.5 * ((numpy.max(self.phi) + self.dphi) +
numpy.min(self.phi))
phi_range = (numpy.max(self.phi) + self.dphi) - numpy.min(self.phi)
rmsf_phi = phi_center - phi_range +\
numpy.arange(2 * self.nphi) * self.dphi
try:
# Convolve weights with the GCF
[l2_grid, w_grid] = G.grid_1d(self.weights_nonuni,
self.l2_nonuni / numpy.pi,
self.dl2 / 2., self.m, self.alpha)
# Put the convolved points on a grid
weights4rmsf = G.sample_grid(self.l2_beam, l2_grid, w_grid)
except TypeError:
print type(self.weights)
print len(self.weights)
print type(self.l2_nonuni)
print len(self.l2_nonuni)
print self.dphi
print self.nphi * 2.
raise
rmsf = numpy.fft.fftshift(numpy.fft.fft(weights4rmsf))
tndx_phi = int(self. nphi * (self.alpha - 1))
rmsf = rmsf[tndx_phi:tndx_phi + 2 * self.nphi]
#for indx in range(2 * self.nphi):
#rot = 2. * (self.l20 - self.l2i) * rmsf_phi[indx]
## shift results by the L20 and by the initial l2 value to account
## for the fact that we start at L2!=0 as FFT expects
#rmsf[indx] = numpy.complex(math.cos(rot), math.sin(rot)) *\
#rmsf[indx]
rot = 2. * (self.l20 - self.l2i) * rmsf_phi
# shift results by the L20 and by the initial l2 value to account
# for the fact that we start at L2!=0 as FFT expects
rmsf *= numpy.exp(complex(0, 1) * rot)
gridnorm = G.gridnorm(rmsf_phi, self.dl2 / 2., self.m, self.beta)
rmsf = rmsf / gridnorm
# The normalization should be as such that the rmsf has a peak
# value of 1
self.K = 1. / abs(rmsf[self.nphi])
# normalize the rmsf
rmsf = rmsf * self.K
# adjust the normalization
# because the rmsf has twice as many pixels as the image
self.K = 2. * self.K
return rmsf, rmsf_phi
