def gt_blast_center_noise_uneven(
sensor_epoch_s: np.array,
noise_std_loss_bits: float = 2,
frequency_center_hz: Optional[float] = None) -> np.ndarray:
"""
Construct the GT explosion pulse of Garces (2019) for even or uneven sensor time
in Gaussion noise with SNR in bits re signal STD.
This is a very flexible variation.
:param sensor_epoch_s: array with timestamps for signal in epoch seconds
:param noise_std_loss_bits: number of bits below signal standard deviation. Default is 2
:param frequency_center_hz: center frequency in Hz. Optional
:return: numpy array with anti-aliased GT explosion pulse with Gaussian noise
"""
time_duration_s = sensor_epoch_s[-1] - sensor_epoch_s[0]
if frequency_center_hz:
pseudo_period_s = 1 / frequency_center_hz
else:
pseudo_period_s = time_duration_s / 4.
# Convert to seconds
time_center_s = sensor_epoch_s - sensor_epoch_s[0] - time_duration_s / 2.
sig_gt = gt_blast_period_center(time_center_s, pseudo_period_s)
sig_noise = white_noise_fbits(np.copy(sig_gt), noise_std_loss_bits)
gt_white = sig_gt + sig_noise
# AA filter
gt_white_aa = antialias_halfNyquist(gt_white)
return gt_white_aa
def gt_blast_center_fast(
frequency_peak_hz: float = 6.3,
sample_rate_hz: float = 100.,
noise_std_loss_bits: float = 16) -> Tuple[np.ndarray, np.ndarray]:
"""
Fast computation of GT pulse with noise
:param frequency_peak_hz: peak frequency, nominal 6.3 Hz for 1 tonne TNT
:param sample_rate_hz: sample rate, nominal 100 Hz
:param noise_std_loss_bits: noise loss relative to signal variance
:return: centered time in seconds, GT pulse with white noise
"""
duration_s = 16 / frequency_peak_hz # 16 cycles for 6th octave (M = 14)
pseudo_period_s = 1 / frequency_peak_hz
duration_points = int(duration_s * sample_rate_hz)
time_center_s = np.arange(duration_points) / sample_rate_hz
time_center_s -= time_center_s[-1] / 2.
sig_gt = gt_blast_period_center(time_center_s, pseudo_period_s)
sig_noise = white_noise_fbits(sig_gt, noise_std_loss_bits)
gt_white = sig_gt + sig_noise
# AA filter
gt_white_aa = antialias_halfNyquist(gt_white)
return time_center_s, gt_white_aa
sig_wf_red = np.concatenate(
[np.zeros(head_points), sig_wf_red,
np.zeros(head_points)])
sig_time_s = np.arange(len(sig_wf_red)) / sig_wf_sample_rate_hz
sig_duration_s = np.max(sig_time_s)
# Blueshift
sig_wf_blue = np.flipud(sig_wf_red)
# Choose origin and red/blue shift
sig_wf_epoch_s = sig_time_s + run_time_epoch_s
sig_wf = np.copy(np.imag(sig_wf_red))
# Antialias filter synthetic
synthetics.antialias_halfNyquist(synth=sig_wf)
# Export to wav directory
if do_save_wave:
wav_sample_rate_hz = 8000.
export_filename = os.path.join(output_wav_directory,
wav_filename + "_8kz.wav")
synth_wav = 0.9 * np.real(sig_wf) / np.max(np.abs((np.real(sig_wf))))
scipy.io.wavfile.write(export_filename, int(wav_sample_rate_hz),
synth_wav)
# Frame to mic start and end and plot
event_reference_time_epoch_s = sig_wf_epoch_s[0]
# The min_frequency_hz is needed for STFT
max_time_s, min_frequency_hz = scales.from_duration(
time_s = np.arange(time_points)/frequency_sample_rate_hz
time_half_s = np.max(time_s)/2.
time_shifted_s = time_s - time_half_s
# Build signal, no noise
sig_gt = kaboom.gt_blast_period_center(time_center_s=time_shifted_s,
pseudo_period_s=pseudo_period_s)
# Add white noise
# Variance computed from transient, stressing at bit_loss=1
bit_loss = 6
sig_noise = synth.white_noise_fbits(sig=sig_gt,
std_bit_loss=bit_loss)
gt_white = sig_gt + sig_noise
# AA filter of signal with noise
sig_n = synth.antialias_halfNyquist(synth=gt_white) # With noise
# Compute complex wavelet transform of real signal + noise
cwtm, _, _, frequency_hz = \
atoms.cwt_chirp_from_sig(sig_wf=sig_n,
frequency_sample_rate_hz=frequency_sample_rate_hz,
band_order_Nth=order_Nth)
# Reconstruction coefficients
_, reconstruct = \
atoms_inverse.morlet2_reconstruct(band_order_Nth=order_Nth,
scale_frequency_center_hz=frequency_hz,
frequency_sample_rate_hz=frequency_sample_rate_hz)
# Scaled wavelet coefficients
f_x_cwtm = utils.d1tile_x_d2(d1=reconstruct,
d2=cwtm.real)
# Inverse to verify reconstruction (only works for real, returns Hilbert on imaginary
# Select signal
sig_gt = kaboom.gt_blast_period_center(time_center_s=time_shifted_s,
pseudo_period_s=pseudo_period_s)
sig_gt_hilbert = kaboom.gt_hilbert_blast_period_center(time_center_s=time_shifted_s,
pseudo_period_s=pseudo_period_s)
sig_complex = sig_gt + sig_gt_hilbert*1j
# Add white noise
# Variance computed from transient
bit_loss = 1
sig_noise = synth.white_noise_fbits(sig=sig_gt,
std_bit_loss=bit_loss)
gt_white = sig_gt + sig_noise
gt_white_hilbert = sig_gt_hilbert + sig_noise
# AA filter
noise = synth.antialias_halfNyquist(synth=sig_noise)
sig_n = synth.antialias_halfNyquist(synth=gt_white)
sig_n_hilbert = synth.antialias_halfNyquist(synth=gt_white_hilbert)
# Analytic record
sig_n_complex = sig_n + sig_n_hilbert*1j
# Compute complex wavelet transform
cwtm, _, _, frequency_hz = \
atoms.cwt_chirp_from_sig(sig_wf=sig_n,
frequency_sample_rate_hz=frequency_sample_rate_hz,
band_order_Nth=order_Nth)
# For noise
cwtm_noise, _, _, _ = \
atoms.cwt_chirp_from_sig(sig_wf=noise,
frequency_sample_rate_hz=frequency_sample_rate_hz,
band_order_Nth=order_Nth)
order_number_input = 12
EVENT_NAME = "Tone Test"
station_id_str = 'Synthya'
run_time_epoch_s = utils.datetime_now_epoch_s()
mic_sig_sample_rate_hz = 800.
sig_frequency_hz = 50.
sig_duration_s = 5.
# Construct synthetic tone with max unit amplitude
mic_sig_epoch_s = np.arange(int(mic_sig_sample_rate_hz * sig_duration_s)) / mic_sig_sample_rate_hz + run_time_epoch_s
mic_sig = np.sin(2*np.pi*sig_frequency_hz*mic_sig_epoch_s)
mic_sig += synthetics.white_noise_fbits(sig=mic_sig, std_bit_loss=4.)
mic_sig *= utils.taper_tukey(mic_sig_epoch_s, fraction_cosine=0.1) # add taper
synthetics.antialias_halfNyquist(mic_sig) # Antialias filter synthetic
# Frame to mic start and end and plot
event_reference_time_epoch_s = mic_sig_epoch_s[0]
max_time_s, min_frequency_hz = scales.from_duration(band_order_Nth=order_number_input,
sig_duration_s=sig_duration_s)
print('\nRequest Order N=', order_number_input)
print('Lowest frequency in hz that can support this order for this signal duration is ', min_frequency_hz)
print('Scale with signal duration and to Nyquist, default G2 base re F1')
# Select plot frequencies
fmin = np.ceil(min_frequency_hz)
fmax = 400
# TFR SECTION