#%% Perform the test
for i, esno in enumerate(ESNOs):
print("ESNO", esno)
total_errors = 0
total_frame_errors = 0
total_bits = 0
total_frames = 0
for _ in range(num_runs_max):
data = np.random.binomial(1, 0.5, wirt.DATA_SIZE).astype(np.uint8)
# Encoder
encoded = encoder.encode(data)
# Channel
received = channel_AWGN(np.pad(encoded, (20, 40), 'constant'), esno)
# Decoder
decoded = encoder.decode(received, estimate_esno=esno)
###
errors = (decoded != data)
total_errors += errors.sum()
total_frame_errors += errors.any()
total_bits += wirt.DATA_SIZE
total_frames += 1
if total_frame_errors >= num_frame_errors_max:
break
else:
def test_polar_soft_combine(filename='output/polar_soft_combine.npz'):
print("Polar compare post- vs pre-combining.")
num_runs_max = 50000
num_frame_errors = 100
ESNOs = np.arange(-12, -9, 0.25)
total_bits = 14400
K = 400
Ns = np.array([2048, 2048, 4096, 4096])
repeats = (total_bits / Ns).astype(int)
results_BER = np.empty((len(repeats), len(ESNOs)))
results_BLER = np.empty((len(repeats), len(ESNOs)))
for i, esno in enumerate(ESNOs):
print("ESNO", esno)
for n_rep in range(len(repeats)):
total_errors = 0
total_frame_errors = 0
total_bits = 0
total_frames = 0
for _ in range(num_runs_max):
###
data = np.frombuffer(np.random.bytes(K // 8), dtype=np.uint8)
data_bits = np.unpackbits(data)
###
polar_data = np.tile(polar_encode(Ns[n_rep], K, data_bits),
repeats[n_rep])
polar_data_repacked = packed_to_unpacked(
np.packbits(polar_data))
polar_data_modulated = qpsk_modulate(polar_data_repacked)
###
polar_coded = channel_AWGN(polar_data_modulated, esno)
######
polar_reshaped = polar_coded.reshape((repeats[n_rep], -1))
polar_channel_LLRs = qpsk_demodulate_soft(polar_reshaped, esno)
#####
if n_rep % 1 == 0:
polar_result = polar_decode_ssc(
Ns[n_rep], K,
polar_channel_LLRs.mean(axis=0).flatten())
######
else:
polar_results_soft_all = np.empty((repeats[n_rep], K))
for j in range(len(polar_channel_LLRs)):
polar_results_soft_all[j] = polar_decode_ssc(
Ns[n_rep],
K,
polar_channel_LLRs[j],
soft_output=True)
polar_result = polar_results_soft_all.mean(axis=0)
######
errors = (polar_result != data_bits)
total_errors += errors.sum()
total_frame_errors += errors.any()
total_bits += K
total_frames += 1
if total_frame_errors >= num_frame_errors:
break
else:
print("Timeout at {} dB".format(esno))
results_BER[n_rep, i] = total_errors / total_bits
results_BLER[n_rep, i] = total_frame_errors / total_frames
np.savez(filename,
N=Ns,
K=K,
repeats=repeats,
ESNOs=ESNOs,
results_BER=results_BER,
results_BLER=results_BLER,
legend=[
"Hard combine", "Soft combine", "Hard combine", "Soft combine"
],
config={
'num_runs_max': num_runs_max,
'num_frame_errors': num_frame_errors,
'total_bits': total_bits
})
def test_polar_compare_impl(filename='output/polar_compare_impl.npz'):
print("Polar with and without approximation.")
num_runs_max = 20000
num_frame_errors = 150
ESNOs = np.arange(-12, -9, 0.25)
total_bits = 14400
K = 400
N = 2048
reps = total_bits // N
run_types = ["None", "F approx", "SSC + F_approx"]
results_BER = np.empty((len(run_types), len(ESNOs)))
results_BLER = np.empty((len(run_types), len(ESNOs)))
for i, esno in enumerate(ESNOs):
print("ESNO", esno)
for rt_i, _ in enumerate(run_types):
total_errors = 0
total_frame_errors = 0
total_bits = 0
total_frames = 0
for _ in range(num_runs_max):
###
data = np.frombuffer(np.random.bytes(K // 8), dtype=np.uint8)
data_bits = np.unpackbits(data)
###
polar_data = np.tile(polar_encode(N, K, data_bits), reps)
polar_data_repacked = packed_to_unpacked(
np.packbits(polar_data))
polar_data_modulated = qpsk_modulate(polar_data_repacked)
###
polar_coded = channel_AWGN(polar_data_modulated, esno)
######
polar_reshaped = polar_coded.reshape((reps, -1))
polar_channel_LLRs = qpsk_demodulate_soft(polar_reshaped, esno)
#####
mean_LLRs = polar_channel_LLRs.mean(axis=0).flatten()
if rt_i == 0:
polar_result = polar_decode_alternate(N,
K,
mean_LLRs,
use_f_approx=False)
elif rt_i == 1:
polar_result = polar_decode_alternate(N,
K,
mean_LLRs,
use_f_approx=True)
elif rt_i == 2:
polar_result = polar_decode_ssc(N, K, mean_LLRs)
######
errors = (polar_result != data_bits)
total_errors += errors.sum()
total_frame_errors += errors.any()
total_bits += K
total_frames += 1
if total_frame_errors >= num_frame_errors:
break
else:
print("Timeout at {} dB".format(esno))
results_BER[rt_i, i] = total_errors / total_bits
results_BLER[rt_i, i] = total_frame_errors / total_frames
np.savez(filename,
N=N,
K=K,
repeats=reps,
ESNOs=ESNOs,
results_BER=results_BER,
results_BLER=results_BLER,
legend=run_types,
config={
'num_runs_max': num_runs_max,
'num_frame_errors': num_frame_errors,
'total_bits': total_bits
})
def test_polar_rate_vs_rep(filename='output/polar_rates_vs_rep.npz'):
num_runs_max = 50000
num_frame_errors = 100
ESNOs = np.arange(-12, -6, 0.25)
total_bits = 14400
K = 400
Ns = 2**(np.arange(9, 14))
repeats = (total_bits / Ns).astype(int)
results_BER = np.empty((len(repeats), len(ESNOs)))
results_BLER = np.empty((len(repeats), len(ESNOs)))
for i, esno in enumerate(ESNOs):
print("ESNO", esno)
for n_rep in range(len(repeats)):
total_errors = 0
total_frame_errors = 0
total_bits = 0
total_frames = 0
for _ in range(num_runs_max):
###
data = np.frombuffer(np.random.bytes(K // 8), dtype=np.uint8)
data_bits = np.unpackbits(data)
###
polar_data = np.tile(polar_encode(Ns[n_rep], K, data_bits),
repeats[n_rep])
polar_data_repacked = packed_to_unpacked(
np.packbits(polar_data))
polar_data_modulated = qpsk_modulate(polar_data_repacked)
###
polar_coded = channel_AWGN(polar_data_modulated, esno)
###
polar_reshaped = polar_coded.reshape((repeats[n_rep], -1))
polar_channel_LLRs = np.mean(qpsk_demodulate_soft(
polar_reshaped, esno),
axis=0)
polar_result = polar_decode_ssc(Ns[n_rep], K,
polar_channel_LLRs.flatten())
###
errors = (polar_result != data_bits)
total_errors += errors.sum()
total_frame_errors += errors.any()
total_bits += K
total_frames += 1
if total_frame_errors >= num_frame_errors:
break
else:
print("Timeout at {} dB".format(esno))
results_BER[n_rep, i] = total_errors / total_bits
results_BLER[n_rep, i] = total_frame_errors / total_frames
np.savez(filename,
N=Ns,
K=K,
repeats=repeats,
ESNOs=ESNOs,
results_BER=results_BER,
results_BLER=results_BLER,
legend=[f"N: {Ns[i]}, Rep: {repeats[i]}" for i in range(len(Ns))],
config={
'num_runs_max': num_runs_max,
'num_frame_errors': num_frame_errors,
'total_bits': total_bits,
})
def test_polar_combining(K=32, repetitions=2, plot=False):
num_runs = 100
ESNOs = np.arange(-13, -2, 0.5)
# First just repeating and combining soft LLRs
# Then a rate 1/2 code, repeated a number of times
# Then polar code with the lowest rate possible
results_uncoded_BER = np.empty((num_runs, len(ESNOs)))
results_uncoded_BLER = np.empty((num_runs, len(ESNOs)))
results_rate_half_BER = np.empty((num_runs, len(ESNOs)))
results_rate_half_BLER = np.empty((num_runs, len(ESNOs)))
results_rate_min_BER = np.empty((num_runs, len(ESNOs)))
results_rate_min_BLER = np.empty((num_runs, len(ESNOs)))
for n_run in range(num_runs):
if num_runs > 10 and n_run % (num_runs // 10) == 0:
print(n_run)
# Generate data
data = np.frombuffer(np.random.bytes(K // 8), dtype=np.uint8)
data_bits = np.unpackbits(data)
# Match N
uncoded_data = np.tile(data_bits, (2 * repetitions))
polar_rate_half = np.tile(polar_encode(2 * K, K, data_bits),
repetitions)
polar_rate_min = polar_encode(repetitions * 2 * K, K, data_bits)
# Pack
uncoded_repacked = packed_to_unpacked(np.packbits(uncoded_data))
polar_rate_half_repacked = packed_to_unpacked(
np.packbits(polar_rate_half))
polar_rate_min_repacked = packed_to_unpacked(
np.packbits(polar_rate_min))
# Modulate
uncoded_modulated = qpsk_modulate(uncoded_repacked)
polar_rate_half_modulated = qpsk_modulate(polar_rate_half_repacked)
polar_rate_min_modulated = qpsk_modulate(polar_rate_min_repacked)
for i, esno in enumerate(ESNOs):
# print("{}: esno {}".format(i, esno))
uncoded_channel = channel_AWGN(uncoded_modulated, esno)
uncoded_reshaped = uncoded_channel.reshape((-1, K // 2))
uncoded_channel_LLRs = np.mean(qpsk_demodulate_soft(
uncoded_reshaped, esno),
axis=0)
uncoded_bits = np.unpackbits(
unpacked_to_packed(qpsk_hard_decision(uncoded_channel_LLRs)))
results_uncoded_BER[n_run, i] = (uncoded_bits != data_bits).mean()
results_uncoded_BLER[n_run, i] = (uncoded_bits != data_bits).any()
###
polar_rate_half_channel = channel_AWGN(polar_rate_half_modulated,
esno)
polar_rate_half_reshaped = polar_rate_half_channel.reshape((-1, K))
polar_rate_half_channel_LLRs = np.mean(qpsk_demodulate_soft(
polar_rate_half_reshaped, esno),
axis=0)
polar_result_rate_half = polar_decode_ssc(
2 * K, K, polar_rate_half_channel_LLRs.flatten())
results_rate_half_BER[n_run, i] = (polar_result_rate_half !=
data_bits).mean()
results_rate_half_BLER[n_run, i] = (polar_result_rate_half !=
data_bits).any()
###
polar_rate_min_channel = channel_AWGN(polar_rate_min_modulated,
esno)
polar_rate_min_demod = qpsk_demodulate_soft(
polar_rate_min_channel, esno).flatten()
polar_result_rate_min = polar_decode_ssc(repetitions * 2 * K, K,
polar_rate_min_demod)
results_rate_min_BER[n_run, i] = (polar_result_rate_min !=
data_bits).mean()
results_rate_min_BLER[n_run, i] = (polar_result_rate_min !=
data_bits).any()
if plot:
###
plt.figure("Polar BER test")
plt.title(
f"Comparison of different rate polar codes, QPSK, AWGN channel, K={K}"
)
plt.plot(ESNOs,
results_uncoded_BER.mean(axis=0),
label=f'Repetition code R = 1 / {1/(2*repetitions)}')
plt.plot(
ESNOs,
results_rate_half_BER.mean(axis=0),
label=
f'Polar code (SC decoder) R = 1 / {1/2}, repeated = {repetitions}')
plt.plot(ESNOs,
results_rate_min_BER.mean(axis=0),
label=f'Polar code (SC decoder) R = 1 / {1/(2*repetitions)}')
plt.ylabel("BER")
plt.yscale('log')
plt.legend()
###
plt.figure("Polar BLER test")
plt.title(
f"Comparison of different rate polar codes, QPSK, AWGN channel, K={K}"
)
plt.plot(ESNOs,
results_uncoded_BLER.mean(axis=0),
label=f'Repetition code R = 1 / {1/(2*repetitions)}')
plt.plot(
ESNOs,
results_rate_half_BLER.mean(axis=0),
label=
f'Polar code (SC decoder) R = 1 / {1/2}, repeated = {repetitions}')
plt.plot(ESNOs,
results_rate_min_BLER.mean(axis=0),
label=f'Polar code (SC decoder) R = 1 / {1/(2*repetitions)}')
plt.xlabel("ESNO")
plt.ylabel("BLER")
plt.yscale('log')
plt.legend()
###
plt.show()
def test_polar_rates(filename='output/polar_various_rates.npz'):
print("Polar compare performance at different rates")
num_runs_max = 1000
num_frame_errors = 100
ESNOs = np.arange(-11, 3, 0.5)
rates = 1 / np.array([2, 3, 5, 10, 16, 20.48])
N = 2**13
K = (N * rates).astype(int)
results_BER = np.empty((len(rates), len(ESNOs)))
results_BLER = np.empty((len(rates), len(ESNOs)))
for i, esno in enumerate(ESNOs):
print("ESNO", esno)
for K_i, cur_K in enumerate(K):
total_errors = 0
total_frame_errors = 0
total_bits = 0
total_frames = 0
for _ in range(num_runs_max):
data_bits = np.random.binomial(1, 0.5, cur_K).astype(np.uint8)
polar_data = polar_encode(N, cur_K, data_bits)
polar_data_repacked = packed_to_unpacked(
np.packbits(polar_data))
polar_data_modulated = qpsk_modulate(polar_data_repacked)
polar_coded = channel_AWGN(polar_data_modulated, esno)
polar_coded_demod_soft = qpsk_demodulate_soft(
polar_coded, esno).flatten()
polar_result = polar_decode_ssc(N, cur_K,
polar_coded_demod_soft)
######
errors = (polar_result != data_bits)
total_errors += errors.sum()
total_frame_errors += errors.any()
total_bits += cur_K
total_frames += 1
if total_frame_errors >= num_frame_errors:
break
else:
print("Timeout at {} dB".format(esno))
results_BER[K_i, i] = total_errors / total_bits
results_BLER[K_i, i] = total_frame_errors / total_frames
np.savez(filename,
N=N,
K=K,
ESNOs=ESNOs,
results_BER=results_BER,
results_BLER=results_BLER,
legend=[f"Rate: {r}" for r in rates],
config={
'num_runs_max': num_runs_max,
'num_frame_errors': num_frame_errors,
'total_bits': total_bits,
})
def test_BER(filename='output/BER_compare.npz', enable_conv=False):
num_runs_max = 20000
num_frame_errors = 150
ESNOs = np.arange(-1, 10, 0.25)
N = 2**13
K = N // 2
types = [
"Uncoded", "Repetition code (Hard)", "Repetition code (Soft)",
"Polar code"
]
if enable_conv:
conv_params = (3, 7, 5)
types += "Convolutional code"
N_types = len(types)
results_BER = np.empty((N_types, len(ESNOs)))
results_BLER = np.empty((N_types, len(ESNOs)))
if enable_conv:
conv_params = (3, 7, 5)
for i, esno in enumerate(ESNOs):
print("ESNO", esno)
# A counter for the total number of frame errors, so we can stop when some of the
# schemes have reached their limit.
total_frame_errors = np.zeros(N_types, np.int)
total_bit_errors = np.zeros(N_types, np.int)
total_frames = np.zeros(N_types, np.int)
total_bits = np.zeros(N_types, np.int)
for _ in range(num_runs_max):
data = np.frombuffer(np.random.bytes(K // 8), dtype=np.uint8)
data_unpacked = packed_to_unpacked(data)
data_bits = np.unpackbits(data)
###
if total_frame_errors[0] < num_frame_errors:
data_modulated = qpsk_modulate(data_unpacked)
uncoded = channel_AWGN(data_modulated, esno)
uncoded_bits = np.unpackbits(
unpacked_to_packed(qpsk_demodulate(uncoded)))
total_frame_errors[0] += (uncoded_bits != data_bits).any()
total_bit_errors[0] += np.count_nonzero(
uncoded_bits != data_bits)
total_frames[0] += 1
total_bits[0] += K
###
if (total_frame_errors[1] < num_frame_errors) and (
total_frame_errors[2] < num_frame_errors):
rep_modulated = np.tile(data_modulated, (1, 2)).ravel()
rep_coded = channel_AWGN(rep_modulated, esno)
rep_coded_reshaped = rep_coded.reshape((-1, len(uncoded)))
rep_coded_hard = np.array([1, 2], np.uint8) @ (np.stack(
(rep_coded_reshaped.real < 0,
rep_coded_reshaped.imag < 0)).sum(axis=1) >= 1)
rep_coded_hard_bits = np.unpackbits(
unpacked_to_packed(rep_coded_hard))
rep_coded_LLRs = np.roll(qpsk_demodulate_soft(
rep_coded_reshaped, esno).mean(axis=0),
1,
axis=1)
rep_coded_soft_bits = np.unpackbits(
unpacked_to_packed(qpsk_hard_decision(rep_coded_LLRs)))
total_frame_errors[1] += (rep_coded_hard_bits !=
data_bits).any()
total_bit_errors[1] += np.count_nonzero(
rep_coded_hard_bits != data_bits)
total_frame_errors[2] += (rep_coded_soft_bits !=
data_bits).any()
total_bit_errors[2] += np.count_nonzero(
rep_coded_soft_bits != data_bits)
total_frames[1:3] += 1
total_bits[1:3] += K
###
if total_frame_errors[3] < num_frame_errors:
polar_data = polar_encode(N, K, data_bits)
polar_data_repacked = packed_to_unpacked(
np.packbits(polar_data))
polar_data_modulated = qpsk_modulate(polar_data_repacked)
polar_coded = channel_AWGN(polar_data_modulated, esno)
polar_coded_demod_soft = qpsk_demodulate_soft(
polar_coded, esno).flatten()
polar_result = polar_decode_ssc(N, K, polar_coded_demod_soft)
total_frame_errors[3] += (polar_result != data_bits).any()
total_bit_errors[3] += np.count_nonzero(
polar_result != data_bits)
total_frames[3] += 1
total_bits[3] += K
###
if enable_conv and total_frame_errors[4] < num_frame_errors:
conv_data = conv_encode(data, *conv_params)
conv_data_modulated = qpsk_modulate(
packed_to_unpacked(conv_data))
conv_coded = channel_AWGN(conv_data_modulated, esno)
conv_coded_demod = np.frombuffer(unpacked_to_packed(
qpsk_demodulate(conv_coded)),
dtype=np.uint8)
conv_decoded = conv_decode(conv_coded_demod[::2],
conv_coded_demod[1::2],
*conv_params)
total_frame_errors[4] += (np.unpackbits(conv_decoded) !=
data_bits).any()
total_bit_errors[4] += np.count_nonzero(
np.unpackbits(conv_decoded) != data_bits)
total_frames[4] += 1
total_bits[4] += K
results_BER[:, i] = total_bit_errors / total_bits
results_BLER[:, i] = total_frame_errors / total_frames
qfunc = lambda x: 0.5 * erfc(x / np.sqrt(2))
expected_uncoded_BER = qfunc(np.sqrt(10**(ESNOs / 10)))
np.savez(filename,
N=N,
K=K,
ESNOs=ESNOs,
results_BER=results_BER,
results_BLER=results_BLER,
expected_BER_uncoded=expected_uncoded_BER,
legend=types,
config={
'num_runs_max': num_runs_max,
'num_frame_errors': num_frame_errors,
})
def test_polar_limit(filename='output/polar_limit_rate05_nossc.npz'):
num_runs_max = 20000
num_frame_errors = 100
ESNOs = np.arange(-1, 5, 0.25)
SEED = 456
N = 8192
K = 4096
np.random.seed(SEED)
results_BER = np.empty((len(ESNOs)))
results_BLER = np.empty((len(ESNOs)))
results_frame_errors = np.empty((len(ESNOs)))
for i, esno in enumerate(ESNOs):
print("ESNO", esno)
total_errors = 0
total_frame_errors = 0
total_bits = 0
total_frames = 0
for _ in range(num_runs_max):
###
data = np.frombuffer(np.random.bytes(K // 8), dtype=np.uint8)
data_bits = np.unpackbits(data)
###
polar_data = polar_encode(N, K, data_bits)
polar_data_repacked = packed_to_unpacked(np.packbits(polar_data))
polar_data_modulated = qpsk_modulate(polar_data_repacked)
###
polar_coded = channel_AWGN(polar_data_modulated, esno)
###
polar_coded_demod_soft = qpsk_demodulate_soft(polar_coded,
esno).flatten()
# polar_result = polar_decode_ssc(N, K, polar_coded_demod_soft)
polar_result = polar_decode_alternate(N,
K,
polar_coded_demod_soft,
use_f_approx=False)
###
errors = (polar_result != data_bits)
total_errors += errors.sum()
total_frame_errors += errors.any()
total_frames += 1
total_bits += len(data_bits)
if total_frame_errors >= num_frame_errors:
break
else:
print("Timeout at {} dB".format(esno))
results_BER[i] = total_errors / total_bits
results_BLER[i] = total_frame_errors / total_frames
results_frame_errors[i] = total_frame_errors
#% Save file
config = {
"seed": SEED,
"num_runs": num_runs_max,
"ESNOs": ESNOs,
"Time": datetime.now(),
'num_runs_max': num_runs_max,
'num_frame_errors_max': num_frame_errors
}
np.savez(filename,
N=N,
K=K,
ESNOs=ESNOs,
results_BER=results_BER,
results_BLER=results_BLER,
frame_errors=results_frame_errors,
config=config)