import h5py
import numpy as np
import os, random
import tensorflow as tf
from tensorflow.keras.layers import Input, Reshape, ZeroPadding2D, Conv2D, Dropout, Flatten, Dense, Activation, MaxPooling2D, \
AlphaDropout
from tensorflow.keras import layers
import tensorflow.keras.models as Model
# from tensorflow.keras.regularizers import *
# from tensorflow.keras.optimizers import adam
# import seaborn as sns
# import keras
import matplotlib.pyplot as plt
#plt.rcParams.update({'figure.max_open_warning': 0})
# import gc
from tensorflow.compat.v1 import ConfigProto
from tensorflow.compat.v1 import InteractiveSession
%matplotlib inline
os.environ["KERAS_BACKEND"] = "tensorflow"
# 防止出现GPU内存爆炸
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ['CUDA_VISIBLE_DEVICES'] = "0,1" # 选择哪一块gpu
config = ConfigProto()
config.allow_soft_placement = True # 如果你指定的设备不存在，允许TF自动分配设备
config.gpu_options.per_process_gpu_memory_fraction = 0.7 # 分配百分之七十的显存给程序使用，避免内存溢出，可以自己调整
config.gpu_options.allow_growth = True # 按需分配显存，这个比较重要
session = InteractiveSession(config=config)
"""数据集处理"""
import keras.backend.tensorflow_backend as tfback
print("tf.__version__ is", tf.__version__)
print("tf.keras.__version__ is:", tf.keras.__version__)
def _get_available_gpus():
"""Get a list of available gpu devices (formatted as strings).
# Returns
A list of available GPU devices.
"""
# global _LOCAL_DEVICES
if tfback._LOCAL_DEVICES is None:
devices = tf.config.list_logical_devices()
tfback._LOCAL_DEVICES = [x.name for x in devices]
return [x for x in tfback._LOCAL_DEVICES if 'device:gpu' in x.lower()]
tfback._get_available_gpus = _get_available_gpus
############################################
# 由于硬件限制，无法使用完整数据集，因此我从完整数据集中抽取出部分数据，并分割成24个部分
# 每部分对应一种调制，有1200*26=31200条数据
# 因此，目前数据集大小为748800*1024*2
############################################
for i in range(0, 24): # 24个数据集文件
########打开文件#######
filename = 'ExtractDataset/part' + str(i) + '.h5'
print(filename)
f = h5py.File(filename, 'r')
########读取数据#######
X_data = f['X'][:]
Y_data = f['Y'][:]
Z_data = f['Z'][:]
f.close()
#########分割训练集和测试集#########
# 每读取到一个数据文件就直接分割为训练集和测试集，防止爆内存
n_examples = X_data.shape[0]
n_train = int(n_examples * 0.7) # 70%训练样本
train_idx = np.random.choice(range(0, n_examples), size=n_train, replace=False) # 随机选取训练样本下标
test_idx = list(set(range(0, n_examples)) - set(train_idx)) # 测试样本下标
if i == 0:
X_train = X_data[train_idx]
Y_train = Y_data[train_idx]
Z_train = Z_data[train_idx]
X_test = X_data[test_idx]
Y_test = Y_data[test_idx]
Z_test = Z_data[test_idx]
else:
X_train = np.vstack((X_train, X_data[train_idx]))
Y_train = np.vstack((Y_train, Y_data[train_idx]))
Z_train = np.vstack((Z_train, Z_data[train_idx]))
X_test = np.vstack((X_test, X_data[test_idx]))
Y_test = np.vstack((Y_test, Y_data[test_idx]))
Z_test = np.vstack((Z_test, Z_data[test_idx]))
print('训练集X维度：', X_train.shape)
print('训练集Y维度：', Y_train.shape)
print('训练集Z维度：', Z_train.shape)
print('测试集X维度：', X_test.shape)
print('测试集Y维度：', Y_test.shape)
print('测试集Z维度：', Z_test.shape)
##查看数据是否正常
sample_idx = 8736 # 随机下标
print('snr:', Z_train[sample_idx])
print('Y', Y_train[sample_idx])
plt_data = X_train[sample_idx].T
data_fig = plt.figure(figsize=(15, 5))
plt.plot(plt_data[0])
plt.plot(plt_data[1], color='red')
plt.show()
"""建立模型"""
classes = ['32PSK',
'16APSK',
'32QAM',
'FM',
'GMSK',
'32APSK',
'OQPSK',
'8ASK',
'BPSK',
'8PSK',
'AM-SSB-SC',
'4ASK',
'16PSK',
'64APSK',
'128QAM',
'128APSK',
'AM-DSB-SC',
'AM-SSB-WC',
'64QAM',
'QPSK',
'256QAM',
'AM-DSB-WC',
'OOK',
'16QAM']
data_format = 'channels_first'
def residual_stack(Xm, kennel_size, Seq, pool_size):
# 1*1 Conv Linear
Xm = Conv2D(32, (1, 1), padding='same', name=Seq + "_conv1", kernel_initializer='glorot_normal',
data_format=data_format)(Xm)
# Residual Unit 1
Xm_shortcut = Xm
Xm = Conv2D(32, kennel_size, padding='same', activation="relu", name=Seq + "_conv2",
kernel_initializer='glorot_normal', data_format=data_format)(Xm)
Xm = Conv2D(32, kennel_size, padding='same', name=Seq + "_conv3", kernel_initializer='glorot_normal',
data_format=data_format)(Xm)
Xm = layers.add([Xm, Xm_shortcut])
Xm = Activation("relu")(Xm)
# Residual Unit 2
Xm_shortcut = Xm
Xm = Conv2D(32, kennel_size, padding='same', activation="relu", name=Seq + "_conv4",
kernel_initializer='glorot_normal', data_format=data_format)(Xm)
X = Conv2D(32, kennel_size, padding='same', name=Seq + "_conv5", kernel_initializer='glorot_normal',
data_format=data_format)(Xm)
Xm = layers.add([Xm, Xm_shortcut])
Xm = Activation("relu")(Xm)
# MaxPooling
Xm = MaxPooling2D(pool_size=pool_size, strides=pool_size, padding='valid', data_format=data_format)(Xm)
return Xm
in_shp = X_train.shape[1:] # 每个样本的维度[1024,2]
# input layer
Xm_input = Input(in_shp)
Xm = Reshape([1, 1024, 2], input_shape=in_shp)(Xm_input)
# Residual Srack
Xm = residual_stack(Xm, kennel_size=(3, 2), Seq="ReStk0", pool_size=(2, 2)) # shape:(512,1,32)
Xm = residual_stack(Xm, kennel_size=(3, 1), Seq="ReStk1", pool_size=(2, 1)) # shape:(256,1,32)
Xm = residual_stack(Xm, kennel_size=(3, 1), Seq="ReStk2", pool_size=(2, 1)) # shape:(128,1,32)
Xm = residual_stack(Xm, kennel_size=(3, 1), Seq="ReStk3", pool_size=(2, 1)) # shape:(64,1,32)
Xm = residual_stack(Xm, kennel_size=(3, 1), Seq="ReStk4", pool_size=(2, 1)) # shape:(32,1,32)
Xm = residual_stack(Xm, kennel_size=(3, 1), Seq="ReStk5", pool_size=(2, 1)) # shape:(16,1,32)
#############################################################################
# 多次尝试发现减少一层全连接层能使loss下降更快
# 将AlphaDropout设置为0.3似乎比0.5效果更好
#############################################################################
# Full Con 1
Xm = Flatten(data_format=data_format)(Xm)
Xm = Dense(128, activation='selu', kernel_initializer='glorot_normal', name="dense1")(Xm)
Xm = AlphaDropout(0.3)(Xm)
# Full Con 2
Xm = Dense(len(classes), kernel_initializer='glorot_normal', name="dense2")(Xm)
# SoftMax
Xm = Activation('softmax')(Xm)
# Create Model
model = Model.Model(inputs=Xm_input, outputs=Xm)
adam = tf.keras.optimizers.Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=None, decay=0.0, amsgrad=False)
model.compile(loss='categorical_crossentropy', optimizer=adam)
model.summary()
"""训练模型"""
#############################################################################
# 当val_loss连续10次迭代不再减小或总迭代次数大于100时停止
# 将最小验证损失的模型保存
#############################################################################
print(tf.test.gpu_device_name())
filepath = 'Models/ResNet_Model_72w.h5'
history = model.fit(X_train,
Y_train,
batch_size=1000,
epochs=100,
verbose=2,
validation_data=(X_test, Y_test),
# validation_split = 0.3,
callbacks=[
tf.keras.callbacks.ModelCheckpoint(filepath, monitor='val_loss', verbose=0,
save_best_only=True),
tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=10, verbose=0, mode='auto')
])
# we re-load the best weights once training is finished
model.load_weights(filepath)
val_loss_list = history.history['val_loss']
loss_list = history.history['loss']
plt.plot(range(len(loss_list)), val_loss_list)
plt.plot(range(len(loss_list)), loss_list)
plt.show()
# ##########从loss走势来看，预计loss还能继续下降，故再训练一次#######
# history = model.fit(X_train,
# Y_train,
# batch_size=1000,
# epochs=100,
# verbose=2,
# validation_data=(X_test, Y_test),
# # validation_split = 0.3,
# callbacks=[
# keras.callbacks.ModelCheckpoint(filepath, monitor='val_loss', verbose=0, save_best_only=True, mode='auto'),
# keras.callbacks.EarlyStopping(monitor='val_loss', patience=10, verbose=0, mode='auto')
# ])
#
# # we re-load the best weights once training is finished
# model.load_weights(filepath)
"""**测试**"""
def plot_confusion_matrix(cm, title='Confusion matrix', cmap=plt.cm.Blues, labels=[]):
plt.figure(figsize=(10, 10))
plt.imshow(cm, interpolation='nearest', cmap=cmap)
plt.title(title)
plt.colorbar()
tick_marks = np.arange(len(labels))
plt.xticks(tick_marks, labels, rotation=45)
plt.yticks(tick_marks, labels)
plt.tight_layout()
plt.ylabel('True label')
plt.xlabel('Predicted label')
plt.show()
#plt.clf()
# Plot confusion matrix
batch_size = 1024
test_Y_hat = model.predict(X_test, batch_size=1024)
conf = np.zeros([len(classes), len(classes)])
confnorm = np.zeros([len(classes), len(classes)])
for i in range(0, X_test.shape[0]):
j = list(Y_test[i, :]).index(1)
k = int(np.argmax(test_Y_hat[i, :]))
conf[j, k] = conf[j, k] + 1
for i in range(0, len(classes)):
confnorm[i, :] = conf[i, :] / np.sum(conf[i, :])
plot_confusion_matrix(confnorm, labels=classes)
for i in range(len(confnorm)):
print(classes[i], confnorm[i, i])
acc = {}
Z_test = Z_test.reshape((len(Z_test)))
SNRs = np.unique(Z_test)
for snr in SNRs:
X_test_snr = X_test[Z_test == snr]
Y_test_snr = Y_test[Z_test == snr]
pre_Y_test = model.predict(X_test_snr)
conf = np.zeros([len(classes), len(classes)])
confnorm = np.zeros([len(classes), len(classes)])
for i in range(0, X_test_snr.shape[0]): # 该信噪比下测试数据量
j = list(Y_test_snr[i, :]).index(1) # 正确类别下标
j = classes.index(classes[j])
k = int(np.argmax(pre_Y_test[i, :])) # 预测类别下标
k = classes.index(classes[k])
conf[j, k] = conf[j, k] + 1
for i in range(0, len(classes)):
confnorm[i, :] = conf[i, :] / np.sum(conf[i, :])
plt.figure()
plot_confusion_matrix(confnorm, labels=classes, title="ConvNet Confusion Matrix (SNR=%d)" % (snr))
cor = np.sum(np.diag(conf))
ncor = np.sum(conf) - cor
print("Overall Accuracy %s: " % snr, cor / (cor + ncor))
acc[snr] = 1.0 * cor / (cor + ncor)
plt.show() # 仅仅有下面这个plt.clf()能出现标注，但是没有图片，难道是因为被clf()了吗？
#plt.clf() # 不关闭，容易导致 warning,且导致程序无法继续执行下去了，直接终止，让模型无法训练所有epoch，（RuntimeWarning: More than 20 figures have been opened.）
plt.plot(acc.keys(), acc.values())
plt.ylabel('ACC')
plt.xlabel('SNR')
plt.show()
snr: [26]
Y [1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
Model: "model"
Layer (type) Output Shape Param # Connected to
input_1 (InputLayer) [(None, 1024, 2)] 0
reshape (Reshape) (None, 1, 1024, 2) 0 input_1[0][0]
ReStk0_conv1 (Conv2D) (None, 32, 1024, 2) 64 reshape[0][0]
ReStk0_conv2 (Conv2D) (None, 32, 1024, 2) 6176 ReStk0_conv1[0][0]
ReStk0_conv3 (Conv2D) (None, 32, 1024, 2) 6176 ReStk0_conv2[0][0]
add (Add) (None, 32, 1024, 2) 0 ReStk0_conv3[0][0]
ReStk0_conv1[0][0]
activation (Activation) (None, 32, 1024, 2) 0 add[0][0]
ReStk0_conv4 (Conv2D) (None, 32, 1024, 2) 6176 activation[0][0]
add_1 (Add) (None, 32, 1024, 2) 0 ReStk0_conv4[0][0]
activation[0][0]
activation_1 (Activation) (None, 32, 1024, 2) 0 add_1[0][0]
max_pooling2d (MaxPooling2D) (None, 32, 512, 1) 0 activation_1[0][0]
ReStk1_conv1 (Conv2D) (None, 32, 512, 1) 1056 max_pooling2d[0][0]
ReStk1_conv2 (Conv2D) (None, 32, 512, 1) 3104 ReStk1_conv1[0][0]
ReStk1_conv3 (Conv2D) (None, 32, 512, 1) 3104 ReStk1_conv2[0][0]
add_2 (Add) (None, 32, 512, 1) 0 ReStk1_conv3[0][0]
ReStk1_conv1[0][0]
activation_2 (Activation) (None, 32, 512, 1) 0 add_2[0][0]
ReStk1_conv4 (Conv2D) (None, 32, 512, 1) 3104 activation_2[0][0]
add_3 (Add) (None, 32, 512, 1) 0 ReStk1_conv4[0][0]
activation_2[0][0]
activation_3 (Activation) (None, 32, 512, 1) 0 add_3[0][0]
max_pooling2d_1 (MaxPooling2D) (None, 32, 256, 1) 0 activation_3[0][0]
ReStk2_conv1 (Conv2D) (None, 32, 256, 1) 1056 max_pooling2d_1[0][0]
ReStk2_conv2 (Conv2D) (None, 32, 256, 1) 3104 ReStk2_conv1[0][0]
ReStk2_conv3 (Conv2D) (None, 32, 256, 1) 3104 ReStk2_conv2[0][0]
add_4 (Add) (None, 32, 256, 1) 0 ReStk2_conv3[0][0]
ReStk2_conv1[0][0]
activation_4 (Activation) (None, 32, 256, 1) 0 add_4[0][0]
ReStk2_conv4 (Conv2D) (None, 32, 256, 1) 3104 activation_4[0][0]
add_5 (Add) (None, 32, 256, 1) 0 ReStk2_conv4[0][0]
activation_4[0][0]
activation_5 (Activation) (None, 32, 256, 1) 0 add_5[0][0]
max_pooling2d_2 (MaxPooling2D) (None, 32, 128, 1) 0 activation_5[0][0]
ReStk3_conv1 (Conv2D) (None, 32, 128, 1) 1056 max_pooling2d_2[0][0]
ReStk3_conv2 (Conv2D) (None, 32, 128, 1) 3104 ReStk3_conv1[0][0]
ReStk3_conv3 (Conv2D) (None, 32, 128, 1) 3104 ReStk3_conv2[0][0]
add_6 (Add) (None, 32, 128, 1) 0 ReStk3_conv3[0][0]
ReStk3_conv1[0][0]
activation_6 (Activation) (None, 32, 128, 1) 0 add_6[0][0]
ReStk3_conv4 (Conv2D) (None, 32, 128, 1) 3104 activation_6[0][0]
add_7 (Add) (None, 32, 128, 1) 0 ReStk3_conv4[0][0]
activation_6[0][0]
activation_7 (Activation) (None, 32, 128, 1) 0 add_7[0][0]
max_pooling2d_3 (MaxPooling2D) (None, 32, 64, 1) 0 activation_7[0][0]
ReStk4_conv1 (Conv2D) (None, 32, 64, 1) 1056 max_pooling2d_3[0][0]
ReStk4_conv2 (Conv2D) (None, 32, 64, 1) 3104 ReStk4_conv1[0][0]
ReStk4_conv3 (Conv2D) (None, 32, 64, 1) 3104 ReStk4_conv2[0][0]
add_8 (Add) (None, 32, 64, 1) 0 ReStk4_conv3[0][0]
ReStk4_conv1[0][0]
activation_8 (Activation) (None, 32, 64, 1) 0 add_8[0][0]
ReStk4_conv4 (Conv2D) (None, 32, 64, 1) 3104 activation_8[0][0]
add_9 (Add) (None, 32, 64, 1) 0 ReStk4_conv4[0][0]
activation_8[0][0]
activation_9 (Activation) (None, 32, 64, 1) 0 add_9[0][0]
max_pooling2d_4 (MaxPooling2D) (None, 32, 32, 1) 0 activation_9[0][0]
ReStk5_conv1 (Conv2D) (None, 32, 32, 1) 1056 max_pooling2d_4[0][0]
ReStk5_conv2 (Conv2D) (None, 32, 32, 1) 3104 ReStk5_conv1[0][0]
ReStk5_conv3 (Conv2D) (None, 32, 32, 1) 3104 ReStk5_conv2[0][0]
add_10 (Add) (None, 32, 32, 1) 0 ReStk5_conv3[0][0]
ReStk5_conv1[0][0]
activation_10 (Activation) (None, 32, 32, 1) 0 add_10[0][0]
ReStk5_conv4 (Conv2D) (None, 32, 32, 1) 3104 activation_10[0][0]
add_11 (Add) (None, 32, 32, 1) 0 ReStk5_conv4[0][0]
activation_10[0][0]
activation_11 (Activation) (None, 32, 32, 1) 0 add_11[0][0]
max_pooling2d_5 (MaxPooling2D) (None, 32, 16, 1) 0 activation_11[0][0]
flatten (Flatten) (None, 512) 0 max_pooling2d_5[0][0]
dense1 (Dense) (None, 128) 65664 flatten[0][0]
alpha_dropout (AlphaDropout) (None, 128) 0 dense1[0][0]
dense2 (Dense) (None, 24) 3096 alpha_dropout[0][0]
activation_12 (Activation) (None, 24) 0 dense2[0][0]
Total params: 139,192
Trainable params: 139,192
Non-trainable params: 0
/device:GPU:0
Train on 262080 samples, validate on 112320 samples
Epoch 1/100
262080/262080 - 121s - loss: 2.4649 - val_loss: 2.2968
Epoch 2/100
262080/262080 - 115s - loss: 2.0031 - val_loss: 1.8899
Epoch 3/100
262080/262080 - 114s - loss: 1.8912 - val_loss: 1.8477
Epoch 4/100
262080/262080 - 115s - loss: 1.8314 - val_loss: 2.0882
Epoch 5/100
262080/262080 - 114s - loss: 1.7901 - val_loss: 1.9614
Epoch 6/100
262080/262080 - 115s - loss: 1.7454 - val_loss: 2.4555
Epoch 7/100
262080/262080 - 114s - loss: 1.7270 - val_loss: 1.6827
Epoch 8/100
262080/262080 - 114s - loss: 1.6809 - val_loss: 1.7736
Epoch 9/100
262080/262080 - 114s - loss: 1.6645 - val_loss: 1.7251
Epoch 10/100
262080/262080 - 115s - loss: 1.6398 - val_loss: 1.7696
Epoch 11/100
262080/262080 - 114s - loss: 1.6191 - val_loss: 1.6233
Epoch 12/100
262080/262080 - 114s - loss: 1.6071 - val_loss: 1.6533
Epoch 13/100
262080/262080 - 114s - loss: 1.5874 - val_loss: 1.7658
Epoch 14/100
262080/262080 - 114s - loss: 1.5763 - val_loss: 1.6382
Epoch 15/100
262080/262080 - 114s - loss: 1.5481 - val_loss: 1.6054
Epoch 16/100
262080/262080 - 115s - loss: 1.5422 - val_loss: 1.7639
Epoch 17/100
262080/262080 - 115s - loss: 1.5170 - val_loss: 1.5619
Epoch 18/100
262080/262080 - 114s - loss: 1.5031 - val_loss: 1.5510
Epoch 19/100
262080/262080 - 117s - loss: 1.4890 - val_loss: 1.5124
Epoch 20/100
262080/262080 - 114s - loss: 1.4919 - val_loss: 1.5309
Epoch 21/100
262080/262080 - 114s - loss: 1.4662 - val_loss: 1.5423
Epoch 22/100
262080/262080 - 114s - loss: 1.4601 - val_loss: 1.4762
Epoch 23/100
262080/262080 - 114s - loss: 1.4541 - val_loss: 1.5190
Epoch 24/100
262080/262080 - 114s - loss: 1.4459 - val_loss: 1.5118
Epoch 25/100
262080/262080 - 114s - loss: 1.4465 - val_loss: 1.5295
Epoch 26/100
262080/262080 - 114s - loss: 1.4353 - val_loss: 1.4769
Epoch 27/100
262080/262080 - 114s - loss: 1.4302 - val_loss: 1.5943
Epoch 28/100
262080/262080 - 114s - loss: 1.4219 - val_loss: 1.5141
Epoch 29/100
262080/262080 - 114s - loss: 1.4198 - val_loss: 1.4589
Epoch 30/100
262080/262080 - 114s - loss: 1.4136 - val_loss: 1.5035
Epoch 31/100
262080/262080 - 114s - loss: 1.4133 - val_loss: 1.4398
Epoch 32/100
262080/262080 - 114s - loss: 1.4115 - val_loss: 1.5046
Epoch 33/100
262080/262080 - 114s - loss: 1.5056 - val_loss: 1.4821
Epoch 34/100
262080/262080 - 114s - loss: 1.4238 - val_loss: 1.5173
Epoch 35/100
262080/262080 - 114s - loss: 1.4083 - val_loss: 1.5039
Epoch 36/100
262080/262080 - 114s - loss: 1.3920 - val_loss: 1.4710
Epoch 37/100
262080/262080 - 114s - loss: 1.3912 - val_loss: 1.5334
Epoch 38/100
262080/262080 - 115s - loss: 1.3856 - val_loss: 1.4302
Epoch 39/100
262080/262080 - 114s - loss: 1.3772 - val_loss: 1.5279
Epoch 40/100
262080/262080 - 114s - loss: 1.3770 - val_loss: 1.4703
Epoch 41/100
262080/262080 - 114s - loss: 1.3725 - val_loss: 1.4585
Epoch 42/100
262080/262080 - 115s - loss: 1.3612 - val_loss: 1.4944
Epoch 43/100
262080/262080 - 115s - loss: 1.3561 - val_loss: 1.5215
Epoch 44/100
262080/262080 - 115s - loss: 1.3536 - val_loss: 1.4955
Epoch 45/100
262080/262080 - 116s - loss: 1.3903 - val_loss: 1.4142
Epoch 46/100
262080/262080 - 116s - loss: 1.3444 - val_loss: 1.4782
Epoch 47/100
262080/262080 - 115s - loss: 1.3412 - val_loss: 1.4662
Epoch 48/100
262080/262080 - 115s - loss: 1.3293 - val_loss: 1.4285
Epoch 49/100
262080/262080 - 115s - loss: 1.3182 - val_loss: 1.7751
Epoch 50/100
262080/262080 - 115s - loss: 1.3378 - val_loss: 1.5361
Epoch 51/100
262080/262080 - 115s - loss: 1.3149 - val_loss: 1.5202
Epoch 52/100
262080/262080 - 115s - loss: 1.3127 - val_loss: 1.4762
Epoch 53/100
262080/262080 - 115s - loss: 1.3057 - val_loss: 1.4900
Epoch 54/100
262080/262080 - 115s - loss: 1.3036 - val_loss: 1.5167
Epoch 55/100
262080/262080 - 115s - loss: 1.3059 - val_loss: 1.5198
32PSK 0.7243589743589743
16APSK 0.6083333333333333
32QAM 0.6581196581196581
FM 0.6679487179487179
GMSK 0.5927350427350427
32APSK 0.37286324786324787
OQPSK 0.015384615384615385
8ASK 0.5585470085470086
BPSK 0.5912393162393162
8PSK 0.5544871794871795
AM-SSB-SC 0.46025641025641023
4ASK 0.42991452991452994
16PSK 0.5912393162393162
64APSK 0.42457264957264956
128QAM 0.12799145299145298
128APSK 0.41196581196581195
AM-DSB-SC 0.4025641025641026
AM-SSB-WC 0.27735042735042736
64QAM 0.6102564102564103
QPSK 0.7651709401709401
256QAM 0.4895299145299145
AM-DSB-WC 0.7542735042735043
OOK 0.8143162393162393
16QAM 0.6284188034188034
Overall Accuracy -20: 0.03875432525951557
Overall Accuracy -18: 0.04708362614195362
Overall Accuracy -16: 0.04984929283561326
Overall Accuracy -14: 0.0497790183763666
Overall Accuracy -12: 0.0672365988909427
Overall Accuracy -10: 0.08615456455755312
Overall Accuracy -8: 0.125553226182157
Overall Accuracy -6: 0.17679558011049723
Overall Accuracy -4: 0.2675736961451247
Overall Accuracy -2: 0.37135526008864006
Overall Accuracy 0: 0.4640093786635405
Overall Accuracy 2: 0.5496642741375318
Overall Accuracy 4: 0.6252042007001167
Overall Accuracy 6: 0.7041172365666434
Overall Accuracy 8: 0.7706422018348624
Overall Accuracy 10: 0.8340817242927705
Overall Accuracy 12: 0.821840433043069
Overall Accuracy 14: 0.8342541436464088
Overall Accuracy 16: 0.841743119266055
Overall Accuracy 18: 0.8339930151338766
Overall Accuracy 20: 0.8390909090909091
Overall Accuracy 22: 0.825836216839677
Overall Accuracy 24: 0.8296792060927763
Overall Accuracy 26: 0.8382006594441828
Overall Accuracy 28: 0.8410351201478743
Overall Accuracy 30: 0.8266853277350128