def test_ccovf_fft_vs_convolution(demean, adjusted, reset_randomstate):
x = np.random.normal(size=128)
y = np.random.normal(size=128)
F1 = ccovf(x, y, demean=demean, adjusted=adjusted, fft=False)
F2 = ccovf(x, y, demean=demean, adjusted=adjusted, fft=True)
assert_almost_equal(F1, F2, decimal=7)
def phase_align(reference, target, roi, res=100):
'''
Cross-correlate data within region of interest at a precision of 1./res
if data is cross-correlated at native resolution (i.e. res=1) this function
can only achieve integer precision
Args:
reference (1d array/list): signal that won't be shifted
target (1d array/list): signal to be shifted to reference
roi (tuple): region of interest to compute chi-squared
res (int): factor to increase resolution of data via linear interpolation
Returns:
shift (float): offset between target and reference signal
'''
# convert to int to avoid indexing issues
ROI = slice(int(roi[0]), int(roi[1]), 1)
# interpolate data onto a higher resolution grid
x, r1 = highres(reference[ROI], kind='linear', res=res)
x, r2 = highres(target[ROI], kind='linear', res=res)
# subtract mean
r1 -= r1.mean()
r2 -= r2.mean()
# compute cross covariance
cc = ccovf(r1, r2, demean=False, unbiased=False)
# determine if shift if positive/negative
if np.argmax(cc) == 0:
cc = ccovf(r2, r1, demean=False, unbiased=False)
mod = -1
else:
mod = 1
# often found this method to be more accurate then the way below
return np.argmax(cc) * mod * (1. / res)
# interpolate data onto a higher resolution grid
x, r1 = highres(reference[ROI], kind='linear', res=res)
x, r2 = highres(target[ROI], kind='linear', res=res)
# subtract off mean
r1 -= r1.mean()
r1 -= r2.mean()
# compute the phase-only correlation function
product = np.fft.fft(r1) * np.fft.fft(r2).conj()
cc = np.fft.fftshift(np.fft.ifft(product))
# manipulate the output from np.fft
l = reference[ROI].shape[0]
shifts = np.linspace(-0.5 * l, 0.5 * l, l * res)
# plt.plot(shifts,cc,'k-'); plt.show()
return shifts[np.argmax(cc.real)]
matplotlib.rcParams.update({'font.size': 16})
#####################################################################################
# 1D CCF
#####################################################################################
# Load calibrated data
print("Loading data: {0}-imageplane-dynspectrum-calibrated.stokesI.txt".format(src))
dynspec_I = np.loadtxt("{0}-imageplane-dynspectrum-calibrated.stokesI.txt".format(src))
# TODO: Weight the time average by fscrunch, rather than doing a simple mean
pulse1spectra = np.mean(dynspec_I[ccfstart1:ccfstop1+1], 0)
pulse2spectra = np.mean(dynspec_I[ccfstart2:ccfstop2+1], 0)
# CCF calculation for individual bins
print("Computing CCF between user-selected pulse1 and pulse2")
# Calculate the 1D CCF across frequency between the two user-selected time ranges using statsmodels.tsa.stattools.ccovf
ccf = st.ccovf(pulse1spectra, pulse2spectra)
# Set up figure and axes
ccf_fig, ccf_ax = plt.subplots(figsize=(7,7))
# Plot the CCF
ccf_ax.plot(ccf, label='CCF (bins {0}-{1} x {2}-{3})'.format(ccfstart1,ccfstop1,ccfstart2,ccfstop2))
plt.legend()
ccf_fig.savefig("{0}-CCF_bin{1}-{2}x{3}-{4}.png".format(src,ccfstart1,ccfstop1,ccfstart2,ccfstop2), bbox_inches='tight')
def test_compare_acovf_vs_ccovf(demean, adjusted, fft, reset_randomstate):
x = np.random.normal(size=128)
F1 = acovf(x, demean=demean, adjusted=adjusted, fft=fft)
F2 = ccovf(x, x, demean=demean, adjusted=adjusted, fft=fft)
assert_almost_equal(F1, F2, decimal=7)