def get_mi_estimate(K, m, n, r, l, SNR, H, A, filename):
TxNet, RxNet, Normalize, V, U, R = ae_load_model(K, filename, m, n, r, l,
SNR, H, A)
L1 = 10
L2 = 1000
X_data = np.zeros((L1 * L2, K))
Y_data = np.zeros((L1 * L2, K))
MI = np.zeros(K)
for j in range(L1):
_, SignalIn, _, Noise = prepare_data(K, m, n, r, V, U, R, SNR, L2)
_, Rx = get_output(TxNet, RxNet, Normalize, K, SignalIn, Noise, H)
for k in range(K):
for t1 in range(L2):
X_data[1000 * j + t1, k] = np.array(SignalIn[k])[t1, 0]
Y_data[1000 * j + t1, k] = np.array(Rx[k])[t1, 0]
for k in range(K):
X = X_data[:, k].reshape(-1, 1)
Y = np.squeeze(Y_data[:, k])
M = mutual_info_regression(X, Y)
MI[k] = M
return MI
def train_save_model(
K,
filename,
filename1,
m,
n,
r,
l,
SNR,
H,
A,
n1,
n2,
n3,
save=False,
check=True,
):
TxNet, RxNet, Normalize, V, U, R = ae_load_model(K, filename1, m, n, r, l,
SNR, H, A)
snr = 10**(np.array(SNR) / 10)
snr = n * snr
b_s = 1000
_, SignalIn, TxOut, Noise = prepare_data(K, m, n, r, V, U, R, SNR, b_s)
MSE = []
for _ in range(n1):
TxNet, Normalize, mse = update_weights_encoder(TxNet, Normalize, K,
SignalIn, TxOut, 0.005)
print(mse)
MSE.append(mse)
for _ in range(n2):
RxNet, mse = update_weights_decoder(TxNet, Normalize, RxNet, K,
SignalIn, H, Noise, 0.001)
# RxNet, mse = update_weights_decoder(
# RxNet, K, SignalIn, H, Noise, 0.001, V, R, SNR
# )
print(mse)
MSE.append(mse)
for _ in range(n3):
TxNet, RxNet, Normalize, mse = update_weights_end(
TxNet, RxNet, Normalize, K, SignalIn, H, Noise, 0.001)
print(mse)
MSE.append(mse)
if save:
save_weights(K, filename, TxNet, RxNet, Normalize, U, V, R, MSE, check)
def get_mse_ber(K, m, n, r, l, SNR, H, A, filename):
TxNet, RxNet, Normalize, V, U, R = ae_load_model(K, filename, m, n, r, l,
SNR, H, A)
L1 = 10
L2 = 1000
if m == "Gaussian":
MSE = np.zeros((L1, K))
else:
MSE = np.zeros((L1, K))
BER = np.zeros((L1, K))
SER = np.zeros((L1, K))
for j in range(L1):
MessageIn, SignalIn, _, Noise = prepare_data(K, m, n, r, V, U, R, SNR,
L2)
# _, Rx = get_output(TxNet, RxNet, Normalize, K, SignalIn, Noise, H, V, U, R, SNR)
_, Rx = get_output(TxNet, RxNet, Normalize, K, SignalIn, Noise, H)
for k in range(K):
if m == "Gaussian":
MSE[j, k] = np.mean((Rx[k] - SignalIn[k])**2)
else:
MSE[j, k] = np.mean((Rx[k] - SignalIn[k])**2)
msg_det = PAMDemod(np.array(Rx[k]), m, 1).astype(int)
# SER
SER[j, k] = np.sum((msg_det != MessageIn[k]).astype(int)) / L2
# BER
Mc = np.array([
tuple(int(c) for c in "{:02b}".format(i))
for i in MessageIn[k]
])
M1 = np.array([
tuple(int(c) for c in "{:02b}".format(i)) for i in msg_det
])
BER[j, k] = np.sum(abs(M1 - Mc)) / (m * L2)
if m == "Gaussian":
return np.squeeze(np.mean(MSE, axis=0))
else:
return (
np.squeeze(np.mean(MSE, axis=0)),
np.squeeze(np.mean(BER, axis=0)),
np.squeeze(np.mean(SER, axis=0)),
)
def check_noise_dist(K, m, n, r, l, SNR, H, A, filename):
TxNet, RxNet, Normalize, V, U, R = ae_load_model(K, filename, m, n, r, l,
SNR, H, A)
L1 = 10
L2 = 1000
filt_noise = np.zeros(L1 * L2)
t = 0
for _ in range(L1):
_, SignalIn, _, Noise = prepare_data(K, m, n, r, V, U, R, SNR, L2)
_, Rx = get_output(TxNet, RxNet, Normalize, K, SignalIn, Noise, H)
for t1 in range(1000):
filt_noise[t] = np.array(Rx[0] - SignalIn[0])[t1, 0]
t += 1
return filt_noise
def ae_load_model(K, filename, m, n, r, l, SNR, H, A):
snr = 10**(np.array(SNR) / 10)
snr = n * snr
n_b = 1
b_s = 1000
try:
Network = pickle.load(open(filename, "rb"))
V = Network["V"]
U = Network["U"]
Rp = Network["R"]
except:
Vp, Up, Rp = generate_beamvectors(K, A, n, SNR)
V = np.transpose(np.array(Vp))[0, :, :]
U = np.transpose(np.array(Up))[0, :, :]
_, SignalIn, TxOut, Noise = prepare_data(K, m, n, r, V, U, Rp, SNR, b_s)
TxNet, RxNet, Normalize = model_init(
K,
m,
n,
l,
r,
n_b,
b_s,
H,
SignalIn,
Noise,
TxOut,
snr,
)
try:
Network = pickle.load(open(filename, "rb"))
for k in range(K):
TxNet[k].set_weights(Network["TX"][k])
RxNet[k].set_weights(Network["RX"][k])
Normalize[k].set_weights(Network["NORM"][k])
except:
print("Untrained Model Loaded")
pass
return TxNet, RxNet, Normalize, V, U, Rp