def handle_control_demod(self, keys):
# Check for keys we care about
if "freeze" in keys:
self.run_mode = "freeze"
if self.demodtype == 'neural':
self.logger.info(
"neural demod {}: freezing model and saving state".format(
self.uuid_str))
torch.save(self.demod.model.state_dict(),
"demod_neural_weights-{}.mdl".format(self.uuid_str))
print("Saving final demod constellation @{}".format(
self.demod_update_cnt))
data_vis = gen_demod_grid(points_per_dim=100,
min_val=-2.5,
max_val=2.5)['data']
labels_si_g = self.demod.demodulate(data_c=data_vis,
mode='exploit')
np.savez("neural_demod_constellation_{:05d}_{}".format(
self.demod_update_cnt, time.strftime('%Y%m%d_%H%M%S')),
iq=data_vis,
labels=labels_si_g)
elif "train" in keys:
self.run_mode = "train"
self.logger.info("neural demod {}: resuming training".format(
self.uuid_str))
def train_preamble(self, data_c):
self.update(inputs=self.preamble_si,
actions=[],
data_for_rewards=data_c,
mode='echo')
if self.packet_cnt % self.log_constellation_interval == 0:
print("Saving demod constellation @{}".format(self.packet_cnt))
data_vis = gen_demod_grid(points_per_dim=40,
min_val=-2.5,
max_val=2.5)['data']
labels_si_g = self.demodulate(data_c=data_vis, mode='exploit')
numpy.savez("neural_demod_constellation_{:05d}_{}".format(
self.packet_cnt, time.strftime('%Y%m%d_%H%M%S')),
iq=data_vis,
labels=labels_si_g)
def handle_control(self, msg):
# Test whether message is a dict, per UHD requirements
if not pmt.is_dict(msg):
self.logger.info(
"{} received non-dict control message, ignoring".format(
self.uuid_str))
return
try:
keys = pmt.to_python(pmt.dict_keys(msg))
except pmt.wrong_type as e:
self.logger.debug(
"{} received pair instead of dict, fixing".format(
self.uuid_str))
msg = pmt.dict_add(pmt.make_dict(), pmt.car(msg), pmt.cdr(msg))
keys = pmt.to_python(pmt.dict_keys(msg))
print("KEYS: {}".format(keys))
# Check for keys we care about
if "freeze" in keys:
self.run_mode = "freeze"
self.logger.info(
"neural demod {}: freezing model and saving state".format(
self.uuid_str))
torch.save(self.model.state_dict(),
"demod_neural_weights-{}".format(self.uuid_str))
print("Saving final demod constellation @{}".format(
self.packet_cnt))
data_vis = gen_demod_grid(points_per_dim=40,
min_val=-2.5,
max_val=2.5)['data']
labels_si_g = self.demodulate(data_c=data_vis, mode='exploit')
numpy.savez("neural_demod_constellation_{:05d}_{}".format(
self.packet_cnt, time.strftime('%Y%m%d_%H%M%S')),
iq=data_vis,
labels=labels_si_g)
elif "train" in keys:
self.run_mode = "train"
self.logger.info("neural demod {}: resuming training".format(
self.uuid_str))
def demod_train_preamble(self, data_c, idxecho):
try:
preamble_si = util_data.bits_to_integers(
self.get_preamble_hist(idxecho), self.bits_per_symbol)
self.demod.update(inputs=preamble_si,
actions=[],
data_for_rewards=data_c,
mode='echo')
if self.demod_packet_cnt % self.log_interval == 0:
print("Saving demod constellation @{}".format(
self.demod_packet_cnt))
data_vis = gen_demod_grid(points_per_dim=100,
min_val=-2.5,
max_val=2.5)['data']
labels_si_g = self.demod.demodulate(data_c=data_vis,
mode='exploit')
np.savez("neural_demod_constellation_{:05d}_{}".format(
self.demod_packet_cnt, time.strftime('%Y%m%d_%H%M%S')),
iq=data_vis,
labels=labels_si_g)
except KeyError as e:
self.logger.info(
"Unable to train demodulator with stored index {}".format(
idxecho))