Changed cpp ProximalDecoder implementation to always return x_hat

This commit is contained in:
Andreas Tsouchlos 2022-11-29 01:04:06 +01:00
parent 80964ebb91
commit 56fd051385
5 changed files with 194 additions and 34 deletions

View File

@ -67,10 +67,12 @@ public:
* This function assumes a BPSK modulated signal and an AWGN channel.
* @param y Vector of received values. (y = x + w, where 'x' is element of
* [-1, 1]^n and 'w' is noise)
* @return Most probably sent codeword (element of [0, 1]^n). If decoding
* fails, the returned value is 'None'
* @return \b std::pair of the form (x_hat, num_iter), x_hat is the most
* probably sent codeword and num_iter is the number of iterations that were
* performed. If the parity check fails and no valid codeword is reached,
* num_iter is -1
*/
std::pair<std::optional<RowVectori<t_n>>, int>
std::pair<RowVectori<t_n>, int>
decode(const Eigen::Ref<const RowVectord<t_n>>& y) {
if (y.size() != mH.cols())
throw std::runtime_error("Length of vector must match H matrix");
@ -95,12 +97,19 @@ public:
}
}
return {std::nullopt, mK};
return {x_hat, -1};
}
/// Private members are not private in order to make the class easily
/// picklable
// private:
/**
* @brief Get the values of all member variables necessary to recreate an
* exact copy of this class. Used for pickling
* @return \b std::tuple
*/
auto get_decoder_state() const {
return std::tuple(mK, mOmega, mGamma, mEta, mH);
}
private:
const int mK;
const double mOmega;
const double mGamma;

View File

@ -17,7 +17,13 @@ using namespace pybind11::literals;
.def("decode", &ProximalDecoder<m, n>::decode, "x"_a.noconvert()) \
.def(py::pickle( \
[](const ProximalDecoder<m, n>& a) { \
return py::make_tuple(a.mH, a.mK, a.mOmega, a.mGamma, a.mEta); \
MatrixiR<m, n> H; \
int K; \
double omega; \
double gamma; \
double eta; \
std::tie(H, K, omega, gamma, eta) = a.get_decoder_state(); \
return py::make_tuple(H, K, omega, gamma, eta); \
}, \
[](py::tuple t) { \
return ProximalDecoder<m, n>{ \

View File

@ -18,39 +18,39 @@ def format_yticks(previous_yticks):
def main():
sns.set_theme()
# titles = [
# "$n=7$, $m=4$",
# "$n=31$, $m=20$",
# "$n=31$, $m=5$",
# "$n=96$, $m=48$",
# "$n=204$, $m=102$",
# "$n=408$, $m=204$",
# ]
#
# filenames = [
# "sim_results/2d_dec_fails_w_log_k_lin_bch_7_4alist.csv",
# "sim_results/2d_dec_fails_w_log_k_lin_bch_31_11alist.csv",
# "sim_results/2d_dec_fails_w_log_k_lin_bch_31_26alist.csv",
# "sim_results/2d_dec_fails_w_log_k_lin_963965alist.csv",
# "sim_results/2d_dec_fails_w_log_k_lin_2043486alist.csv",
# "sim_results/2d_dec_fails_w_log_k_lin_40833844alist.csv",
# ]
titles = [
"$n=7$, $m=4$",
"$n=31$, $m=20$",
"$n=31$, $m=5$",
"$n=96$, $m=48$"
"$n=96$, $m=48$",
"$n=204$, $m=102$",
"$n=408$, $m=204$",
]
filenames = [
"sim_results/2d_dec_fails_w_log_k_lin_zoomed_in_bch_7_4alist.csv",
"sim_results/2d_dec_fails_w_log_k_lin_zoomed_in_bch_31_11alist.csv",
"sim_results/2d_dec_fails_w_log_k_lin_zoomed_in_bch_31_26alist.csv",
"sim_results/2d_dec_fails_w_log_k_lin_zoomed_in_963965alist.csv",
"sim_results/2d_dec_fails_w_log_k_lin_bch_7_4alist.csv",
"sim_results/2d_dec_fails_w_log_k_lin_bch_31_11alist.csv",
"sim_results/2d_dec_fails_w_log_k_lin_bch_31_26alist.csv",
"sim_results/2d_dec_fails_w_log_k_lin_963965alist.csv",
"sim_results/2d_dec_fails_w_log_k_lin_2043486alist.csv",
"sim_results/2d_dec_fails_w_log_k_lin_40833844alist.csv",
]
fig, axes = plt.subplots(2, 2, squeeze=False)
# titles = [
# "$n=7$, $m=4$",
# "$n=31$, $m=20$",
# "$n=31$, $m=5$",
# "$n=96$, $m=48$"
# ]
#
# filenames = [
# "sim_results/2d_dec_fails_w_log_k_lin_zoomed_in_bch_7_4alist.csv",
# "sim_results/2d_dec_fails_w_log_k_lin_zoomed_in_bch_31_11alist.csv",
# "sim_results/2d_dec_fails_w_log_k_lin_zoomed_in_bch_31_26alist.csv",
# "sim_results/2d_dec_fails_w_log_k_lin_zoomed_in_963965alist.csv",
# ]
fig, axes = plt.subplots(2, 3, squeeze=False)
fig.suptitle("SNR = 3dB")

View File

@ -23,7 +23,7 @@ def task_func(params):
x = noise.add_awgn(x_bpsk, SNR, n, k)
x_hat, num_iter = decoder.decode(x)
if x_hat is None:
if num_iter == -1:
dec_fails += 1
return dec_fails / num_iterations
@ -80,7 +80,7 @@ def main():
H_file = "BCH_31_26.alist"
SNR = 3
num_iterations = 10000
num_iterations = 1000
omegas = np.logspace(-0.3, -2.82, 40)
Ks = np.ceil(np.linspace(10 ** 1.3, 10 ** 2.3, 40)).astype('int32')

View File

@ -0,0 +1,145 @@
import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import signal
from timeit import default_timer
from tqdm import tqdm
from utility import codes, noise, misc
from utility.simulation.simulators import GenericMultithreadedSimulator
# from cpp_modules.cpp_decoders import ProximalDecoder
from cpp_modules.cpp_decoders import ProximalDecoder_7_3 as ProximalDecoder
def count_bit_errors(d: np.array, d_hat: np.array) -> int:
return np.sum(d != d_hat)
def task_func(params):
signal.signal(signal.SIGINT, signal.SIG_IGN)
decoder, max_iterations, SNR, n, k = params
c = np.zeros(n)
x_bpsk = c + 1
total_bit_errors = 0
total_frame_errors = 0
dec_fails = 0
num_iterations = 0
for i in range(max_iterations):
x = noise.add_awgn(x_bpsk, SNR, n, k)
x_hat, k_max = decoder.decode(x)
bit_errors = count_bit_errors(x_hat, c)
if bit_errors > 0:
total_bit_errors += bit_errors
total_frame_errors += 1
num_iterations += 1
if k_max == -1:
dec_fails += 1
if total_frame_errors > 4000:
break
BER = total_bit_errors / (num_iterations * n)
FER = total_frame_errors / num_iterations
DFR = dec_fails / (num_iterations + dec_fails)
return BER, FER, DFR, num_iterations
def simulate(H_file, SNRs, max_iterations, omega, K, gammas):
sim = GenericMultithreadedSimulator()
# Define fixed simulation params
H = codes.read_alist_file(f"res/{H_file}")
n_min_k, n = H.shape
k = n - n_min_k
# Define params different for each task
params = {}
for i, SNR in enumerate(SNRs):
for j, gamma in enumerate(gammas):
decoder = ProximalDecoder(H=H.astype('int32'), K=K, omega=omega,
gamma=gamma)
params[f"{i}_{j}"] = (decoder, max_iterations, SNR, n, k)
# Set up simulation
sim.task_params = params
sim.task_func = task_func
sim.start_or_continue()
return sim.get_current_results()
def reformat_data(results, SNRs, gammas):
data = {"BER": np.zeros(3 * 10), "FER": np.zeros(3 * 10),
"DFR": np.zeros(3 * 10), "gamma": np.zeros(3 * 10),
"SNR": np.zeros(3 * 10), "num_iter": np.zeros(3 * 10)}
for i, (key, (BER, FER, DFR, num_iter)) in enumerate(results.items()):
i_SNR, i_gamma = key.split('_')
data["BER"][i] = BER
data["FER"][i] = FER
data["DFR"][i] = DFR
data["num_iter"][i] = num_iter
data["SNR"][i] = SNRs[int(i_SNR)]
data["gamma"][i] = gammas[int(i_gamma)]
print(pd.DataFrame(data))
return pd.DataFrame(data)
def main():
# Set up simulation params
sim_name = "BER_FER_DFR"
# H_file = "96.3.965.alist"
# H_file = "204.3.486.alist"
# H_file = "204.55.187.alist"
# H_file = "408.33.844.alist"
H_file = "BCH_7_4.alist"
# H_file = "BCH_31_11.alist"
# H_file = "BCH_31_26.alist"
SNRs = np.arange(1, 6, 0.5)
max_iterations = 10000
# omega = 0.005
# K = 60
omega = 0.05
K = 60
gammas = [0.15, 0.01, 0.05]
# Run simulation
start_time = default_timer()
results = simulate(H_file, SNRs, max_iterations, omega, K, gammas)
end_time = default_timer()
print(f"duration: {end_time - start_time}")
df = reformat_data(results, SNRs, gammas)
df.to_csv(
f"sim_results/2d_dec_fails_{sim_name}_{misc.slugify(H_file)}.csv")
sns.set_theme()
ax = sns.lineplot(data=df, x="SNR", y="BER", hue="gamma")
ax.set_yscale('log')
ax.set_ylim((5e-5, 2e-0))
plt.show()
if __name__ == "__main__":
main()