def calc_avg_amps(refs_all, data_shape, nchans=4, altcol=True):
"""Calculate amplifier averages
Save the average reference value for each amplifier in each frame.
Assume by default that alternating columns are offset from each other,
so we save two arrays: self.refs_amps_avg1 and self.refs_amps_avg2.
Each array has a size of (namp, ngroup).
Parameters
----------
refs_all : ndarray
The top and/or bottom references pixels order
in a shape (nz, nref_rows, nx)
data_shape : tuple
Shape of the data array: (nz, ny, nx).
nchans : int
Number of amplifier output channels.
altcol : bool
Calculate separate reference values for even/odd columns?
Default=True.
"""
nz, ny, nx = data_shape
chsize = int(nx / nchans)
if altcol:
refs_amps_avg1 = []
refs_amps_avg2 = []
for ch in range(nchans):
# Channel indices
ich1 = ch * chsize
ich2 = ich1 + chsize
# Slice out alternating columns
refs_ch1 = refs_all[:, :, ich1:ich2 - 1:2].reshape((nz, -1))
refs_ch2 = refs_all[:, :, ich1 + 1:ich2:2].reshape((nz, -1))
# Take the resistant mean
chavg1 = robust.mean(refs_ch1, axis=1)
chavg2 = robust.mean(refs_ch2, axis=1)
refs_amps_avg1.append(chavg1)
refs_amps_avg2.append(chavg2)
return (np.array(refs_amps_avg1), np.array(refs_amps_avg2))
else:
refs_amps_avg = []
for ch in range(nchans):
# Channel indices
ich1 = ch * chsize
ich2 = ich1 + chsize
# Slice out alternating columns
refs_ch = refs_all[:, :, ich1:ich2].reshape((nz, -1))
# Take the resistant mean and reshape for broadcasting
chavg = robust.mean(refs_ch, axis=1).reshape(-1, 1, 1)
refs_amps_avg.append(chavg)
return np.array(refs_amps_avg)
def calc_avg_amps(refs_all, data_shape, nchans=4, altcol=True):
"""
Save the average reference value for each amplifier in each frame.
This assumes that alternating columns are offset from each other,
so we save two arrays: self.refs_amps_avg1 and self.refs_amps_avg2. Each array
has a size of (namp, ngroup).
top_ref (bool) : Include top reference rows when correcting channel offsets.
bot_ref (bool) : Include bottom reference rows when correcting channel offsets.
"""
nz, ny, nx = data_shape
chsize = int(nx / nchans)
if altcol:
refs_amps_avg1 = []
refs_amps_avg2 = []
for ch in range(nchans):
# Channel indices
ich1 = ch * chsize
ich2 = ich1 + chsize
# Slice out alternating columns
refs_ch1 = refs_all[:, :, ich1:ich2 - 1:2].reshape((nz, -1))
refs_ch2 = refs_all[:, :, ich1 + 1:ich2:2].reshape((nz, -1))
# Take the resistant mean
chavg1 = robust.mean(refs_ch1, axis=1)
chavg2 = robust.mean(refs_ch2, axis=1)
refs_amps_avg1.append(chavg1)
refs_amps_avg2.append(chavg2)
return (np.array(refs_amps_avg1), np.array(refs_amps_avg2))
else:
refs_amps_avg = []
for ch in range(nchans):
# Channel indices
ich1 = ch * chsize
ich2 = ich1 + chsize
# Slice out alternating columns
refs_ch = refs_all[:, :, ich1:ich2].reshape((nz, -1))
# Take the resistant mean and reshape for broadcasting
chavg = robust.mean(refs_ch, axis=1).reshape(-1, 1, 1)
refs_amps_avg.append(chavg)
return np.array(refs_amps_avg)
def correct_amp_refs(self, supermean=False):
"""Correct amplifier offsets
Use values in ``self.refs_amps_avg`` to correct amplifier offsets.
Parameters
----------
supermean : bool
Add back the overall mean of the reference pixels.
"""
# Check to make sure refs_amps_avg is valid
if (self.refs_amps_avg is None):
_log.warning(
'self.refs_amps_avg is set to None. No offsets applied.')
return
#raise ValueError('self.refs_amps_avg is set to None')
# Supermean
# the average of the average is the DC level of the output channel
smean = robust.mean(refs_all) if supermean else 0.0
nchans = self.detector.nout
chsize = self.detector.chsize
nz, ny, nx = self.data.shape
for ch in range(nchans):
# Channel indices
ich1 = int(ch * chsize)
ich2 = int(ich1 + chsize)
# In-place subtraction of channel averages
if self.altcol:
for i in range(nz):
self.data[i, :,
ich1:ich2 - 1:2] -= self.refs_amps_avg[0][ch, i]
self.data[i, :,
ich1 + 1:ich2:2] -= self.refs_amps_avg[1][ch, i]
else:
for i in range(nz):
self.data[i, :, ich1:ich2] -= self.refs_amps_avg[ch, i]
# Add back supermean
if supermean: self.data += smean
def correct_amp_refs(self, supermean=False):
"""
Use values from calc_avg_amps to correct amplifier offsets.
supermean (bool) : Add back the overall mean of the reference pixels.
"""
# Check to make sure refs_amps_avg is valid
if (self.refs_amps_avg is None):
raise ValueError('self.refs_amps_avg is set to None')
# Supermean
# the average of the average is the DC level of the output channel
smean = robust.mean(refs_all) if supermean else 0.0
nchans = self.detector.nout
chsize = self.detector.chsize
nz, ny, nx = self.data.shape
for ch in range(nchans):
# Channel indices
ich1 = ch * chsize
ich2 = ich1 + chsize
# In-place subtraction of channel averages
if self.altcol:
for i in range(nz):
self.data[i, :,
ich1:ich2 - 1:2] -= self.refs_amps_avg[0][ch, i]
self.data[i, :,
ich1 + 1:ich2:2] -= self.refs_amps_avg[1][ch, i]
else:
for i in range(nz):
self.data[i, :, ich1:ich2] -= self.refs_amps_avg[ch, i]
# Add back supermean
if supermean: self.data += smean
def reffix_amps(cube,
nchans=4,
in_place=True,
altcol=True,
supermean=False,
top_ref=True,
bot_ref=True,
ntop=4,
nbot=4,
**kwargs):
"""Correct amplifier offsets
Matches all amplifier outputs of the detector to a common level.
This routine subtracts the average of the top and bottom reference rows
for each amplifier and frame individually.
By default, reference pixel corrections are performed in place since it's
faster and consumes less memory.
Parameters
----------
cube : ndarray
Input datacube. Can be two or three dimensions (nz,ny,nx).
nchans : int
Number of output amplifier channels in the detector. Default=4.
altcol : bool
Calculate separate reference values for even/odd columns.
supermean : bool
Add back the overall mean of the reference pixels.
in_place : bool
Perform calculations in place. Input array is overwritten.
top_ref : bool
Include top reference rows when correcting channel offsets.
bot_ref : bool
Include bottom reference rows when correcting channel offsets.
ntop : int
Specify the number of top reference rows.
nbot : int
Specify the number of bottom reference rows.
"""
if not in_place:
cube = np.copy(cube)
# Check the number of dimensions are valid.
ndim = len(cube.shape)
if ndim == 2:
ny, nx = cube.shape
nz = 1
cube = cube.reshape((nz, ny, nx))
elif ndim == 3:
nz, ny, nx = cube.shape
else:
raise ValueError('Input data can only have 2 or 3 dimensions. \
Found {} dimensions.'.format(ndim))
chsize = int(nx / nchans)
# Number of reference rows to use
# Set nt or nb equal to 0 if we don't want to use either
nt = ntop if top_ref else 0
nb = nbot if bot_ref else 0
if (nt + nb) == 0:
print("No reference pixels available for use. Returning...")
return
# Slice out reference pixels
refs_bot = cube[:, :nb, :]
refs_top = cube[:, -nt:, :]
if nt == 0:
refs_all = refs_bot
elif nb == 0:
refs_all = refs_top
else:
refs_all = np.hstack((refs_bot, refs_top))
assert refs_all.shape[1] == (nb + nt)
# Supermean
# the average of the average is the DC level of the output channel
smean = robust.mean(refs_all) if supermean else 0.0
# Calculate avg reference values for each frame and channel
refs_amps_avg = calc_avg_amps(refs_all,
cube.shape,
nchans=nchans,
altcol=altcol)
for ch in range(nchans):
# Channel indices
ich1 = ch * chsize
ich2 = ich1 + chsize
# In-place subtraction of channel medians
if altcol:
for i in range(nz):
cube[i, :, ich1:ich2 - 1:2] -= refs_amps_avg[0][ch, i]
cube[i, :, ich1 + 1:ich2:2] -= refs_amps_avg[1][ch, i]
else:
for i in range(nz):
cube[i, :, ich1:ich2] -= refs_amps_avg[ch, i]
# Add back supermean
if supermean: cube += smean
cube = cube.squeeze()
return cube
