def preprocess_test_data(self, iq_data, sample_rate,rbw=None):
## getting spectrogram
if rbw:
_, time, fft_d = iq2fft(iq_data, sample_rate, rbw)
else:
_, time, fft_d = iq2fft(iq_data, sample_rate, self.rbw)
## scaling spectrogram
if use_scaling:
fft_d = scale_test_vectors(fft_d, self.scaler)
return (time, fft_d)
def save_alot_spec(data_dir):
plots_path = os.path.join(data_dir, 'samples')
if not os.path.exists(plots_path):
os.mkdir(plots_path)
data_name = os.path.basename(data_dir)
sample_rate = get_xhdr_sample_rate(data_dir)
glob_data_iq = load_raw_data(data_dir)
freqs, _, fft_d = iq2fft(glob_data_iq, sample_rate, rbw)
glob_means = np.mean(fft_d, axis=0)
glob_stds = np.std(fft_d, axis=0)
del fft_d
gc.collect()
basic_len = get_basic_block_len(sample_rate)
starts = np.random.randint(0, glob_data_iq.shape[0] - basic_len, (5, ))
print('random starts as indices:')
print(starts)
for ind, j in enumerate(starts):
data_iq = glob_data_iq[j:j + basic_len, :]
_, time, fft_d = iq2fft(data_iq, sample_rate, rbw)
means = np.mean(fft_d, axis=0)
stds = np.std(fft_d, axis=0)
f, axes = plt.subplots(nrows=2, ncols=1, sharex=True)
plt.sca(axes[0])
plt.plot(freqs, glob_means, color='black', linewidth=3)
plt.plot(freqs, glob_means - glob_stds, color='red', linewidth=3)
plt.plot(freqs, glob_means + glob_stds, color='red', linewidth=3)
plt.plot(freqs, means, color='green', linewidth=1)
plt.plot(freqs, means - stds, color='green', linewidth=1)
plt.plot(freqs, means + stds, color='green', linewidth=1)
plt.sca(axes[1])
plt.imshow(fft_d,
aspect='auto',
origin='lower',
extent=[freqs[0], freqs[-1], time[0], time[-1]])
f.suptitle(data_name)
f.suptitle('file: ' + data_name + \
'\nstart index = ' + str(j))
f.set_size_inches(8, 6.5, forward=True)
# plt.show()
# input('continue? [y]')
fig_path = os.path.join(plots_path, data_name + '_sample_' + str(ind))
save_fig(f, fig_path)
plt.close(f)
print('workin on file ' + data_name +
' - {}/{}'.format(i, 5 * num_records))
def preprocess_train_data(self, iq_data,sample_rate,rbw=None):
scaler_path = os.path.join(self.model_path, "train_scaler.pkl")
## getting spectrogram
if rbw:
self.freqs, time, fft_d = iq2fft(iq_data, sample_rate, rbw)
else:
self.freqs, time, fft_d = iq2fft(iq_data, sample_rate, self.rbw)
## scaling spectrogram
if use_scaling:
(fft_d, scaler) = scale_train_vectors(fft_d, scaler_path, rng=feature_range)
self.scaler = scaler
return (time, fft_d)
def preprocess_test_data(self, iq_data, sample_rate,rbw=None):
## getting spectrogram
if rbw:
self.rbw = rbw
self.freqs, time, fft_list = iq2fft(iq_data, sample_rate, self.rbw, mode=['real', 'imag'])
fft_d = np.stack(fft_list, axis=-1)
return (time, fft_d)
def preprocess_test_data(self, iq_data, sample_rate):
self.scaler_path = os.path.join(self.model_path, "train_scaler.pkl")
self.freqs, time, fft_test = iq2fft(iq_data, sample_rate, self.rbw)
## scaling spectrogram
if use_scaling:
fft_test = scale_test_vectors(fft_test, self.scaler_path)
## getting spectrogram
return (time, fft_test)
def preprocess_test_data(self, iq_data, sample_rate, rbw=None):
self.whiten_path = os.path.join(self.model_path, "zca_scaler.pkl")
## whitening
if use_whitening:
iq_data = whiten_test_data(iq_data, self.whiten_path)
## getting spectrogram
freqs, time, fft_d = iq2fft(iq_data, sample_rate, self.rbw)
## scaling spectrogram
if use_scaling:
fft_d = scale_test_vectors(fft_d, scaler=self.scaler)
return ((freqs, time), fft_d)
def preprocess_train(self, iq_data, sample_rate):
params_path = os.path.join(self.model_path, "model_params.pkl")
## getting spectrogram
self.freqs, time, fft_list = iq2fft(iq_data, sample_rate, self.rbw, mode=['real', 'imag'])
fft_iq = np.stack(fft_list, axis=-1)
self.means = np.mean(fft_iq, axis=0)
self.stds = np.std(fft_iq, axis=0)
self.gen_gaussians()
params_dic = {'freqs': self.freqs,
'means': self.means,
'stds': self.stds,
'gaussians': self.gaussians}
persist_object(params_dic, params_path)
self.loaded = True
def plot_prediction(self, iq_data_basic_block, sample_rate):
## get only basic_block_len
basic_len = get_basic_block_len(sample_rate, basic_time)
if basic_len != iq_data_basic_block.shape[0]:
raise("iq_data not 1 [msec]...")
_, pred_matrix = self.predict_basic_block(iq_data_basic_block, sample_rate)
pred_matrix[0,0] = 0
pred_matrix[-1,-1] = 7
_, time, fft_d = iq2fft(iq_data_basic_block, sample_rate, self.rbw)
imshow_limits = [self.freqs[0], self.freqs[-1], time[0], time[-1]]
fig, axes = plt.subplots(nrows=1, ncols=2, sharex=True, sharey=True)
plt.sca(axes[0])
pred_im = plt.imshow(pred_matrix, aspect='auto', origin='lower', extent=imshow_limits)
plt.colorbar(pred_im)
plt.sca(axes[1])
fft_im = plt.imshow(fft_d, aspect='auto', origin='lower', extent=imshow_limits)
plt.colorbar(fft_im)
def plot_prediction(self, iq_data_basic_block, sample_rate):
## get only basic_block_len
basic_len = get_basic_block_len(sample_rate, basic_time)
if basic_len != iq_data_basic_block.shape[0]:
raise ("iq_data too long...")
pred_freqs, pred_time, pred_matrix = self.predict_basic_block(
iq_data_basic_block, sample_rate)
pred_matrix[-1, -1] = sigma_ae
freqs, time, fft_d = iq2fft(iq_data_basic_block, sample_rate, self.rbw)
fig, axes = plt.subplots(nrows=1, ncols=2, sharex=True, sharey=True)
plt.sca(axes[0])
plt.imshow(pred_matrix,
aspect='auto',
origin='lower',
extent=[
pred_freqs[0], pred_freqs[-1], pred_time[0],
pred_time[-1]
])
plt.sca(axes[1])
plt.imshow(fft_d,
aspect='auto',
origin='lower',
extent=[freqs[0], freqs[-1], time[0], time[-1]])
def save_alot_spec(data_dir):
plots_path = data_dir
if not os.path.exists(plots_path):
os.mkdir(plots_path)
data_name = os.path.basename(data_dir)
sample_rate = get_xhdr_sample_rate(data_dir)
glob_data_iq = load_raw_data(data_dir)
basic_len = get_basic_block_len(sample_rate)
starts = np.random.randint(0, glob_data_iq.shape[0] - basic_len,
(sample_per_file, ))
print('random starts as indices:')
print(starts)
for j in starts:
gc.collect()
f, axes = plt.subplots(nrows=1,
ncols=len(complex2scalar_mode) + 1,
sharey=True)
## ploting spectrogram
data_iq = glob_data_iq[j:j + basic_len, :]
freqs, time, fft_d = iq2fft(data_iq, sample_rate, rbw)
plt.sca(axes[0])
plt.imshow(fft_d.T,
aspect='auto',
origin='lower',
extent=[time[0], time[-1], freqs[0], freqs[-1]])
plt.title('spectrogram')
## getting index of intrest freqs
intrest_freqs_index = []
for f_0 in intrest_freqs:
intrest_freqs_index.append(np.argmin(np.abs(freqs - f_0)))
## ploting histograms of power
_, _, ffts = iq2fft(glob_data_iq,
sample_rate,
rbw,
mode=complex2scalar_mode)
for l, fft_d in enumerate(ffts):
plt.sca(axes[l + 1])
for freq_ind in intrest_freqs_index:
freq_col = fft_d[:, freq_ind]
hist, bins_edge = np.histogram(freq_col, bins=200)
hist = (1e6 / np.max(hist)) * hist
plt.plot(bins_edge[:-1], hist + freqs[freq_ind])
print('mode: ' + complex2scalar_mode[l] +
' --- hist for freq: ' + str(freqs[freq_ind]))
plt.title('hist of: ' + complex2scalar_mode[l])
f.suptitle(data_name)
f.set_size_inches(12, 6.5, forward=True)
# plt.show()
# input('continue? [y]')
#
fig_path = os.path.join(plots_path, data_name + '_hists')
save_fig(f, fig_path)
plt.close(f)
print('workin on file ' + data_name +
' - {}/{}'.format(i, sample_per_file * num_records))
