IN_SNR = []
for nl in noise_levels:
bs = []
pred_snr = []
sim_snr = []
for b in range(3, 4):
Bx = 5
Bw = 5
Ba = 8
noise_level = nl
noise_level_db = noise_level
noise_level_linear = utils.db_to_linear(noise_level_db)
f = 437.5
configurations = config.parse_config('./yaml/example.yaml')
configurations.signal.num_pos_decision = 5000
configurations.signal.amps = [1]
configurations.signal.freqs = [f]
configurations.signal.phase = [0]
configurations.noise.init_args.top = noise_level_linear
configurations.noise.init_args.steady_state = noise_level_linear
generator = signal_generator.InputSignalGenerator(
configurations.signal, configurations.noise)
signal_with_noise, _ = generator.get()
configurations.noise.init_args.top = 0
configurations.noise.init_args.steady_state = 0
generator = signal_generator.InputSignalGenerator(
configurations.signal, configurations.noise)
def sim_fx_dp_noise_model(noise_level,
model_coeff_generator,
Bx=B,
Bw=B,
Ba=B):
configurations = config.parse_config('../yaml/example.yaml')
noise_level_db = noise_level
noise_level_linear = utils.db_to_linear(noise_level_db)
configurations.signal.num_pos_decision = 1000
configurations.noise.init_args.top = noise_level_linear
configurations.noise.init_args.steady_state = noise_level_linear
generator = signal_generator.InputSignalGenerator(configurations.signal,
configurations.noise)
signal_with_noise, _ = generator.get()
configurations.noise.init_args.top = 0
configurations.noise.init_args.steady_state = 0
generator = signal_generator.InputSignalGenerator(configurations.signal,
configurations.noise)
signal_without_noise, _ = generator.get()
noise_sample = signal_with_noise - signal_without_noise
Nd = configurations.signal.num_pos_decision
N = configurations.signal.block_size
signal_with_noise = signal_with_noise[0:Nd, :, :]
signal_without_noise = signal_without_noise[0:Nd, :, :]
noise_sample = noise_sample[0:Nd, :, :]
coeff_jdht = np.zeros((Nd, N, N))
for i in range(Nd):
coeff_jdht[i, :, :] = model_coeff_generator(N) / N
var_jdht_coeff = np.var(coeff_jdht)
signal_power = np.var(signal_without_noise)
w_max = np.abs(coeff_jdht).max()
qcoeff_jdht = utils.unisign_quant(coeff_jdht, Bw, w_max)
deltaw = utils.get_delta(Bw, w_max)
sigal_max = np.abs(signal_with_noise).max()
qsignal_with_noise = utils.unisign_quant(signal_with_noise, Bx, sigal_max)
deltax = utils.get_delta(Bx, sigal_max)
result_noise = np.zeros(signal_without_noise.shape)
result_signal_without_noise = np.zeros(signal_without_noise.shape)
result_signal_with_noise = np.zeros(signal_without_noise.shape)
qresult_signal_without_noise = np.zeros(signal_without_noise.shape)
qresult_signal_with_noise = np.zeros(signal_without_noise.shape)
start = datetime.now()
for i in range(Nd):
result_signal_without_noise[i, :, :] = signal_without_noise[
i, :, :].dot(coeff_jdht[i, :, :].T)
result_noise[i, :, :] = noise_sample[i, :, :].dot(
coeff_jdht[i, :, :].T)
result_signal_with_noise[i, :, :] = signal_with_noise[i, :, :].dot(
coeff_jdht[i, :, :].T)
qresult_signal_with_noise[i, :, :] = qsignal_with_noise[i, :, :].dot(
qcoeff_jdht[i, :, :].T)
result_max = result_signal_with_noise.max()
deltay = utils.get_delta(Ba, result_max)
qresult_signal_with_noise = utils.unisign_quant(qresult_signal_with_noise,
Ba, result_max)
end = datetime.now()
total_dp_noise = qresult_signal_with_noise - result_signal_without_noise
var_total_dp_noise = total_dp_noise.var()
A = deltay**2 / 12
B = N * var_jdht_coeff * noise_level_linear
C = N / 12 * var_jdht_coeff * deltax**2
D = N / 12 * signal_without_noise.var() * deltaw**2
E = N / 12 * noise_level_linear * deltaw**2
predicted_total_var = A + B + C + D + E
predicted_signal_power = N * signal_power * var_jdht_coeff
sim_signal_power = result_signal_without_noise.var()
predicted_snr = predicted_signal_power / (A + B + C + D + E)
sim_snr = sim_signal_power / var_total_dp_noise
print('total time (s):', utils.format_float((end - start).total_seconds()),
'Bx:', Bx, 'Bw:', Bw, 'Ba:', Ba, noise_level, 'dB:',
10 * np.log10(predicted_total_var / var_total_dp_noise),
10 * np.log10(predicted_signal_power / sim_signal_power),
10 * np.log10(predicted_snr / sim_snr))
def multi_run_wrapper(args):
return experiment(*args)
def run_single_machine(usr_configs_template, search_space, host_name=None):
pool = Pool(processes=4)
total_args = []
for s in search_space:
total_args.append([usr_configs_template, s])
s = time.time()
print('staring at', datetime.now())
for _ in tqdm(pool.map(multi_run_wrapper, total_args),
total=len(total_args)):
pass
e = time.time()
print('end at', datetime.now())
print('total time in second:', e - s)
if __name__ == '__main__':
import warnings
warnings.simplefilter(action='ignore', category=FutureWarning)
usr_configs_template = config.parse_config('./yaml/template.yaml')
search_space = config.parse_search_space('./yaml/search_space.yaml')
total_args = [[usr_configs_template, s] for s in search_space]
run_single_machine(usr_configs_template, search_space, host_name=None)