def test_filter(resp_wfm,
frequency_filters=None,
noise_threshold=None,
excit_wfm=None,
show_plots=True,
plot_title=None,
verbose=False):
"""
Filters the provided response with the provided filters.
Parameters
----------
resp_wfm : array-like, 1D
Raw response waveform in the time domain
frequency_filters : (Optional) FrequencyFilter object or list of
Frequency filters to apply to signal
noise_threshold : (Optional) float
Noise threshold to apply to signal
excit_wfm : (Optional) 1D array-like
Excitation waveform in the time domain. This waveform is necessary for plotting loops. If the length of
resp_wfm matches excit_wfm, a single plot will be returned with the raw and filtered signals plotted against the
excit_wfm. Else, resp_wfm and the filtered (filt_data) signal will be broken into chunks matching the length of
excit_wfm and a figure with multiple plots (one for each chunk) with the raw and filtered signal chunks plotted
against excit_wfm will be returned for fig_loops
show_plots : (Optional) Boolean
Whether or not to plot FFTs before and after filtering
plot_title : str / unicode (Optional)
Title for the raw vs filtered plots if requested. For example - 'Row 15'
verbose : (Optional) Boolean
Prints extra debugging information if True. Default False
Returns
-------
filt_data : 1D numpy float array
Filtered signal in the time domain
fig_fft : matplotlib.pyplot.figure object
handle to the plotted figure if requested, else None
fig_loops : matplotlib.pyplot.figure object
handle to figure with the filtered signal and raw signal plotted against the excitation waveform
"""
if not isinstance(resp_wfm, (np.ndarray, list)):
raise TypeError('resp_wfm should be array-like')
resp_wfm = np.array(resp_wfm)
show_loops = False
if excit_wfm is not None and show_plots:
if len(resp_wfm) % len(excit_wfm) == 0:
show_loops = True
else:
raise ValueError(
'Length of resp_wfm should be divisibe by length of excit_wfm')
if show_loops:
if plot_title is None:
plot_title = 'FFT Filtering'
else:
assert isinstance(plot_title, (str, unicode))
if frequency_filters is None and noise_threshold is None:
raise ValueError(
'Need to specify at least some noise thresholding / frequency filter'
)
if noise_threshold is not None:
if noise_threshold >= 1 or noise_threshold <= 0:
raise ValueError('Noise threshold must be within (0 1)')
samp_rate = 1
composite_filter = 1
if frequency_filters is not None:
if not isinstance(frequency_filters, Iterable):
frequency_filters = [frequency_filters]
if not are_compatible_filters(frequency_filters):
raise ValueError(
'frequency filters must be a single or list of FrequencyFilter objects'
)
composite_filter = build_composite_freq_filter(frequency_filters)
samp_rate = frequency_filters[0].samp_rate
resp_wfm = np.array(resp_wfm)
num_pts = resp_wfm.size
fft_pix_data = np.fft.fftshift(np.fft.fft(resp_wfm))
if noise_threshold is not None:
noise_floor = get_noise_floor(fft_pix_data, noise_threshold)[0]
if verbose:
print('The noise_floor is', noise_floor)
fig_fft = None
if show_plots:
w_vec = np.linspace(-0.5 * samp_rate, 0.5 * samp_rate, num_pts) * 1E-3
fig_fft, [ax_raw, ax_filt] = plt.subplots(figsize=(12, 8), nrows=2)
axes_fft = [ax_raw, ax_filt]
set_tick_font_size(axes_fft, 14)
r_ind = num_pts
if isinstance(composite_filter, np.ndarray):
r_ind = np.max(np.where(composite_filter > 0)[0])
x_lims = slice(len(w_vec) // 2, r_ind)
amp = np.abs(fft_pix_data)
ax_raw.semilogy(w_vec[x_lims], amp[x_lims], label='Raw')
if frequency_filters is not None:
ax_raw.semilogy(w_vec[x_lims],
(composite_filter[x_lims] + np.min(amp)) *
(np.max(amp) - np.min(amp)),
linewidth=3,
color='orange',
label='Composite Filter')
if noise_threshold is not None:
ax_raw.axhline(
noise_floor,
# ax_raw.semilogy(w_vec, np.ones(r_ind - l_ind) * noise_floor,
linewidth=2,
color='r',
label='Noise Threshold')
ax_raw.legend(loc='best', fontsize=14)
ax_raw.set_title('Raw Signal', fontsize=16)
ax_raw.set_ylabel('Magnitude (a. u.)', fontsize=14)
fft_pix_data *= composite_filter
if noise_threshold is not None:
fft_pix_data[
np.abs(fft_pix_data) <
noise_floor] = 1E-16 # DON'T use 0 here. ipython kernel dies
if show_plots:
ax_filt.semilogy(w_vec[x_lims], np.abs(fft_pix_data)[x_lims])
ax_filt.set_title('Filtered Signal', fontsize=16)
ax_filt.set_xlabel('Frequency(kHz)', fontsize=14)
ax_filt.set_ylabel('Magnitude (a. u.)', fontsize=14)
if noise_threshold is not None:
ax_filt.set_ylim(
bottom=noise_floor
) # prevents the noise threshold from messing up plots
fig_fft.tight_layout()
filt_data = np.real(np.fft.ifft(np.fft.ifftshift(fft_pix_data)))
if verbose:
print('The shape of the filtered data is {}'.format(filt_data.shape))
print('The shape of the excitation waveform is {}'.format(
excit_wfm.shape))
fig_loops = None
if show_loops:
if len(resp_wfm) == len(excit_wfm):
# single plot:
fig_loops, axis = plt.subplots(figsize=(5.5, 5))
axis.plot(excit_wfm, resp_wfm, 'r', label='Raw')
axis.plot(excit_wfm, filt_data, 'k', label='Filtered')
axis.legend(fontsize=14)
set_tick_font_size(axis, 14)
axis.set_xlabel('Excitation', fontsize=16)
axis.set_ylabel('Signal', fontsize=16)
axis.set_title(plot_title, fontsize=16)
fig_loops.tight_layout()
else:
# N loops:
raw_pixels = np.reshape(resp_wfm, (-1, len(excit_wfm)))
filt_pixels = np.reshape(filt_data, (-1, len(excit_wfm)))
print(raw_pixels.shape, filt_pixels.shape)
fig_loops, axes_loops = plot_curves(
excit_wfm, [raw_pixels, filt_pixels],
line_colors=['r', 'k'],
dataset_names=['Raw', 'Filtered'],
x_label='Excitation',
y_label='Signal',
subtitle_prefix='Col ',
num_plots=16,
title=plot_title)
return filt_data, fig_fft, fig_loops
def plot_bayesian_results(bias_sine, i_meas, i_corrected, bias_triang, resistance, r_variance, i_recon=None,
pix_pos=[0, 0], broken_resistance=True, r_max=None, res_scatter=False, **kwargs):
"""
Plots the basic Bayesian Inference results for a specific pixel
Parameters
----------
bias_sine : 1D float numpy array
Original bias vector used for experiment
i_meas : 1D float numpy array
Current measured from experiment
i_corrected : 1D float numpy array
current with capacitance and R extra compensated
i_recon : 1D float numpy array
Reconstructed current
bias_triang : 1D float numpy array
Interpolated bias
resistance : 1D float numpy array
Inferred resistance
r_variance : 1D float numpy array
Variance of the resistance
pix_pos : list of two numbers
Pixel row and column positions or values
broken_resistance : bool, Optional
Whether or not to break the resistance plots into sections so as to avoid plotting areas with high variance
r_max : float, Optional
Maximum value of resistance to plot
res_scatter : bool, Optional
Use scatter instead of line plots for resistance
Returns
-------
fig : matplotlib.pyplot figure handle
Handle to figure
"""
font_size_1 = 14
font_size_2 = 16
half_x_ind = int(0.5 * bias_triang.size)
ex_amp = np.max(bias_triang)
colors = [['red', 'orange'], ['blue', 'cyan']]
syms = [['-', '--', '--'], ['-', ':', ':']]
names = ['Forward', 'Reverse']
cos_omega_t = np.roll(bias_sine, int(-0.25 * bias_sine.size))
orig_half_pt = int(0.5 * bias_sine.size)
i_correct_rolled = np.roll(i_corrected, int(-0.25 * bias_sine.size))
st_dev = np.sqrt(r_variance)
tests = [st_dev < 10, resistance > 0]
if r_max is not None:
tests.append(resistance < r_max)
good_pts = np.ones(resistance.shape, dtype=bool)
for item in tests:
good_pts = np.logical_and(good_pts, item)
good_pts = np.where(good_pts)[0]
good_forw = good_pts[np.where(good_pts < half_x_ind)[0]]
good_rev = good_pts[np.where(good_pts >= half_x_ind)[0]]
pos_limits = resistance + st_dev
neg_limits = resistance - st_dev
fig, axes = plt.subplots(ncols=3, figsize=(15, 5))
# fig.subplots_adjust(wspace=3.5)
axes[0].set_ylabel('Resistance (G$\Omega$)', fontsize=font_size_2)
pts_to_plot = [good_forw, good_rev]
for type_ind, axis, pts_list, cols_set, sym_set, set_name in zip(range(len(names)),
axes[:2], pts_to_plot,
colors, syms, names):
axis.set_title('$R(V)$ ' + set_name + ' at Row = ' + str(pix_pos[1]) +
' Col =' + str(pix_pos[0]), fontsize=font_size_2)
single_plot = not broken_resistance
if broken_resistance:
diff = np.diff(pts_list)
jump_inds = np.argwhere(diff > 4) + 1
if jump_inds.size < 1:
single_plot = True
if not single_plot:
jump_inds = np.append(np.append(0, jump_inds), pts_list[-1])
for ind in range(1, jump_inds.size):
cur_range = pts_list[jump_inds[ind - 1]:jump_inds[ind]]
if res_scatter:
axis.scatter(bias_triang[cur_range], resistance[cur_range],
color=cols_set[0], s=30)
else:
axis.plot(bias_triang[cur_range], resistance[cur_range], cols_set[0],
linestyle=sym_set[0], linewidth=3)
axis.fill_between(bias_triang[cur_range], pos_limits[cur_range], neg_limits[cur_range],
alpha=0.25, color=cols_set[1])
if ind == 1:
axis.legend(['R(V)', 'R(V)+-$\sigma$'], loc='upper center', fontsize=font_size_1)
else:
if res_scatter:
axis.scatter(bias_triang[pts_list], resistance[pts_list],
color=cols_set[0], s=30)
else:
axis.plot(bias_triang[pts_list], resistance[pts_list], cols_set[0],
linestyle=sym_set[0], linewidth=3, label='R(V)')
axis.fill_between(bias_triang[pts_list], pos_limits[pts_list], neg_limits[pts_list],
alpha=0.25, color=cols_set[1], label='R(V)+-$\sigma$')
axis.legend(loc='upper center', fontsize=font_size_1)
axis.set_xlabel('Voltage (V)', fontsize=font_size_2)
axis.set_xlim((-ex_amp, ex_amp))
# ################### CURRENT PLOT ##########################
axes[2].plot(bias_sine, i_meas, 'green', linewidth=3, label='I$_{meas}$')
if i_recon is not None:
axes[2].plot(bias_sine, i_recon, 'c--', linewidth=3, label='I$_{recon}$')
axes[2].plot(cos_omega_t[orig_half_pt:], i_correct_rolled[orig_half_pt:],
'blue', linewidth=3, label='I$_{Bayes} Forw$')
axes[2].plot(cos_omega_t[:orig_half_pt], i_correct_rolled[:orig_half_pt],
'red', linewidth=3, label='I$_{Bayes} Rev$')
# axes[2].legend(loc='upper right', bbox_to_anchor=(-.1, 0.30), fontsize=font_size_1)
axes[2].legend(loc='best', fontsize=font_size_1)
axes[2].set_xlabel('Voltage(V)', fontsize=font_size_2)
axes[2].set_title('$I(V)$ at row ' + str(pix_pos[0]) + ', col ' + str(pix_pos[1]),
fontsize=font_size_2)
axes[2].set_ylabel('Current (nA)', fontsize=font_size_2)
set_tick_font_size(axes, font_size_1)
fig.tight_layout()
return fig
def plot_bayesian_results(bias_sine,
i_meas,
i_corrected,
bias_triang,
resistance,
r_variance,
i_recon=None,
pix_pos=[0, 0],
broken_resistance=True,
r_max=None,
res_scatter=False,
**kwargs):
"""
Plots the basic Bayesian Inference results for a specific pixel
Parameters
----------
bias_sine : 1D float numpy array
Original bias vector used for experiment
i_meas : 1D float numpy array
Current measured from experiment
i_corrected : 1D float numpy array
current with capacitance and R extra compensated
i_recon : 1D float numpy array
Reconstructed current
bias_triang : 1D float numpy array
Interpolated bias
resistance : 1D float numpy array
Inferred resistance
r_variance : 1D float numpy array
Variance of the resistance
pix_pos : list of two numbers
Pixel row and column positions or values
broken_resistance : bool, Optional
Whether or not to break the resistance plots into sections so as to avoid plotting areas with high variance
r_max : float, Optional
Maximum value of resistance to plot
res_scatter : bool, Optional
Use scatter instead of line plots for resistance
Returns
-------
fig : matplotlib.pyplot figure handle
Handle to figure
"""
font_size_1 = 14
font_size_2 = 16
half_x_ind = int(0.5 * bias_triang.size)
ex_amp = np.max(bias_triang)
colors = [['red', 'orange'], ['blue', 'cyan']]
syms = [['-', '--', '--'], ['-', ':', ':']]
names = ['Forward', 'Reverse']
cos_omega_t = np.roll(bias_sine, int(-0.25 * bias_sine.size))
orig_half_pt = int(0.5 * bias_sine.size)
i_correct_rolled = np.roll(i_corrected, int(-0.25 * bias_sine.size))
st_dev = np.sqrt(r_variance)
tests = [st_dev < 10, resistance > 0]
if r_max is not None:
tests.append(resistance < r_max)
good_pts = np.ones(resistance.shape, dtype=bool)
for item in tests:
good_pts = np.logical_and(good_pts, item)
good_pts = np.where(good_pts)[0]
good_forw = good_pts[np.where(good_pts < half_x_ind)[0]]
good_rev = good_pts[np.where(good_pts >= half_x_ind)[0]]
pos_limits = resistance + st_dev
neg_limits = resistance - st_dev
fig, axes = plt.subplots(ncols=3, figsize=(15, 5))
# fig.subplots_adjust(wspace=3.5)
axes[0].set_ylabel('Resistance (G$\Omega$)', fontsize=font_size_2)
pts_to_plot = [good_forw, good_rev]
for type_ind, axis, pts_list, cols_set, sym_set, set_name in zip(
range(len(names)), axes[:2], pts_to_plot, colors, syms, names):
axis.set_title('$R(V)$ ' + set_name + ' at Row = ' + str(pix_pos[1]) +
' Col =' + str(pix_pos[0]),
fontsize=font_size_2)
single_plot = not broken_resistance
if broken_resistance:
diff = np.diff(pts_list)
jump_inds = np.argwhere(diff > 4) + 1
if jump_inds.size < 1:
single_plot = True
if not single_plot:
jump_inds = np.append(np.append(0, jump_inds), pts_list[-1])
for ind in range(1, jump_inds.size):
cur_range = pts_list[jump_inds[ind - 1]:jump_inds[ind]]
if res_scatter:
axis.scatter(bias_triang[cur_range],
resistance[cur_range],
color=cols_set[0],
s=30)
else:
axis.plot(bias_triang[cur_range],
resistance[cur_range],
cols_set[0],
linestyle=sym_set[0],
linewidth=3)
axis.fill_between(bias_triang[cur_range],
pos_limits[cur_range],
neg_limits[cur_range],
alpha=0.25,
color=cols_set[1])
if ind == 1:
axis.legend(['R(V)', 'R(V)+-$\sigma$'],
loc='upper center',
fontsize=font_size_1)
else:
if res_scatter:
axis.scatter(bias_triang[pts_list],
resistance[pts_list],
color=cols_set[0],
s=30)
else:
axis.plot(bias_triang[pts_list],
resistance[pts_list],
cols_set[0],
linestyle=sym_set[0],
linewidth=3,
label='R(V)')
axis.fill_between(bias_triang[pts_list],
pos_limits[pts_list],
neg_limits[pts_list],
alpha=0.25,
color=cols_set[1],
label='R(V)+-$\sigma$')
axis.legend(loc='upper center', fontsize=font_size_1)
axis.set_xlabel('Voltage (V)', fontsize=font_size_2)
axis.set_xlim((-ex_amp, ex_amp))
# ################### CURRENT PLOT ##########################
axes[2].plot(bias_sine, i_meas, 'green', linewidth=3, label='I$_{meas}$')
if i_recon is not None:
axes[2].plot(bias_sine,
i_recon,
'c--',
linewidth=3,
label='I$_{recon}$')
axes[2].plot(cos_omega_t[orig_half_pt:],
i_correct_rolled[orig_half_pt:],
'blue',
linewidth=3,
label='I$_{Bayes} Forw$')
axes[2].plot(cos_omega_t[:orig_half_pt],
i_correct_rolled[:orig_half_pt],
'red',
linewidth=3,
label='I$_{Bayes} Rev$')
# axes[2].legend(loc='upper right', bbox_to_anchor=(-.1, 0.30), fontsize=font_size_1)
axes[2].legend(loc='best', fontsize=font_size_1)
axes[2].set_xlabel('Voltage(V)', fontsize=font_size_2)
axes[2].set_title('$I(V)$ at row ' + str(pix_pos[0]) + ', col ' +
str(pix_pos[1]),
fontsize=font_size_2)
axes[2].set_ylabel('Current (nA)', fontsize=font_size_2)
set_tick_font_size(axes, font_size_1)
fig.tight_layout()
return fig
def test_filter(resp_wfm, frequency_filters=None, noise_threshold=None, excit_wfm=None, show_plots=True,
plot_title=None, verbose=False):
"""
Filters the provided response with the provided filters.
Parameters
----------
resp_wfm : array-like, 1D
Raw response waveform in the time domain
frequency_filters : (Optional) FrequencyFilter object or list of
Frequency filters to apply to signal
noise_threshold : (Optional) float
Noise threshold to apply to signal
excit_wfm : (Optional) 1D array-like
Excitation waveform in the time domain. This waveform is necessary for plotting loops. If the length of
resp_wfm matches excit_wfm, a single plot will be returned with the raw and filtered signals plotted against the
excit_wfm. Else, resp_wfm and the filtered (filt_data) signal will be broken into chunks matching the length of
excit_wfm and a figure with multiple plots (one for each chunk) with the raw and filtered signal chunks plotted
against excit_wfm will be returned for fig_loops
show_plots : (Optional) Boolean
Whether or not to plot FFTs before and after filtering
plot_title : str / unicode (Optional)
Title for the raw vs filtered plots if requested. For example - 'Row 15'
verbose : (Optional) Boolean
Prints extra debugging information if True. Default False
Returns
-------
filt_data : 1D numpy float array
Filtered signal in the time domain
fig_fft : matplotlib.pyplot.figure object
handle to the plotted figure if requested, else None
fig_loops : matplotlib.pyplot.figure object
handle to figure with the filtered signal and raw signal plotted against the excitation waveform
"""
if not isinstance(resp_wfm, (np.ndarray, list)):
raise TypeError('resp_wfm should be array-like')
resp_wfm = np.array(resp_wfm)
show_loops = False
if excit_wfm is not None and show_plots:
if len(resp_wfm) % len(excit_wfm) == 0:
show_loops = True
else:
raise ValueError('Length of resp_wfm should be divisibe by length of excit_wfm')
if show_loops:
if plot_title is None:
plot_title = 'FFT Filtering'
else:
assert isinstance(plot_title, (str, unicode))
if frequency_filters is None and noise_threshold is None:
raise ValueError('Need to specify at least some noise thresholding / frequency filter')
if noise_threshold is not None:
if noise_threshold >= 1 or noise_threshold <= 0:
raise ValueError('Noise threshold must be within (0 1)')
samp_rate = 1
composite_filter = 1
if frequency_filters is not None:
if not isinstance(frequency_filters, Iterable):
frequency_filters = [frequency_filters]
if not are_compatible_filters(frequency_filters):
raise ValueError('frequency filters must be a single or list of FrequencyFilter objects')
composite_filter = build_composite_freq_filter(frequency_filters)
samp_rate = frequency_filters[0].samp_rate
resp_wfm = np.array(resp_wfm)
num_pts = resp_wfm.size
fft_pix_data = np.fft.fftshift(np.fft.fft(resp_wfm))
if noise_threshold is not None:
noise_floor = get_noise_floor(fft_pix_data, noise_threshold)[0]
if verbose:
print('The noise_floor is', noise_floor)
fig_fft = None
if show_plots:
w_vec = np.linspace(-0.5 * samp_rate, 0.5 * samp_rate, num_pts) * 1E-3
fig_fft, [ax_raw, ax_filt] = plt.subplots(figsize=(12, 8), nrows=2)
axes_fft = [ax_raw, ax_filt]
set_tick_font_size(axes_fft, 14)
r_ind = num_pts
if isinstance(composite_filter, np.ndarray):
r_ind = np.max(np.where(composite_filter > 0)[0])
x_lims = slice(len(w_vec) // 2, r_ind)
amp = np.abs(fft_pix_data)
ax_raw.semilogy(w_vec[x_lims], amp[x_lims], label='Raw')
if frequency_filters is not None:
ax_raw.semilogy(w_vec[x_lims], (composite_filter[x_lims] + np.min(amp)) * (np.max(amp) - np.min(amp)),
linewidth=3, color='orange', label='Composite Filter')
if noise_threshold is not None:
ax_raw.axhline(noise_floor,
# ax_raw.semilogy(w_vec, np.ones(r_ind - l_ind) * noise_floor,
linewidth=2, color='r', label='Noise Threshold')
ax_raw.legend(loc='best', fontsize=14)
ax_raw.set_title('Raw Signal', fontsize=16)
ax_raw.set_ylabel('Magnitude (a. u.)', fontsize=14)
fft_pix_data *= composite_filter
if noise_threshold is not None:
fft_pix_data[np.abs(fft_pix_data) < noise_floor] = 1E-16 # DON'T use 0 here. ipython kernel dies
if show_plots:
ax_filt.semilogy(w_vec[x_lims], np.abs(fft_pix_data)[x_lims])
ax_filt.set_title('Filtered Signal', fontsize=16)
ax_filt.set_xlabel('Frequency(kHz)', fontsize=14)
ax_filt.set_ylabel('Magnitude (a. u.)', fontsize=14)
if noise_threshold is not None:
ax_filt.set_ylim(bottom=noise_floor) # prevents the noise threshold from messing up plots
fig_fft.tight_layout()
filt_data = np.real(np.fft.ifft(np.fft.ifftshift(fft_pix_data)))
if verbose:
print('The shape of the filtered data is {}'.format(filt_data.shape))
print('The shape of the excitation waveform is {}'.format(excit_wfm.shape))
fig_loops = None
if show_loops:
if len(resp_wfm) == len(excit_wfm):
# single plot:
fig_loops, axis = plt.subplots(figsize=(5.5, 5))
axis.plot(excit_wfm, resp_wfm, 'r', label='Raw')
axis.plot(excit_wfm, filt_data, 'k', label='Filtered')
axis.legend(fontsize=14)
set_tick_font_size(axis, 14)
axis.set_xlabel('Excitation', fontsize=16)
axis.set_ylabel('Signal', fontsize=16)
axis.set_title(plot_title, fontsize=16)
fig_loops.tight_layout()
else:
# N loops:
raw_pixels = np.reshape(resp_wfm, (-1, len(excit_wfm)))
filt_pixels = np.reshape(filt_data, (-1, len(excit_wfm)))
print(raw_pixels.shape, filt_pixels.shape)
fig_loops, axes_loops = plot_curves(excit_wfm, [raw_pixels, filt_pixels], line_colors=['r', 'k'],
dataset_names=['Raw', 'Filtered'], x_label='Excitation',
y_label='Signal', subtitle_prefix='Col ', num_plots=16,
title=plot_title)
return filt_data, fig_fft, fig_loops