Research Article

Automatic Modulation Recognition Based on Hybrid Neural Network

Table 1

MCBL network parameters.

LayerOutput volumeDescription (or remarks)

Input(2, 256, 1)
Input padding(2, 260, 1)Zero padding (0, 2)
Conv1(2, 258, 16)(16, (1, 3))
Dropout1(2, 258, 16)Dropout 0.6
Zero padding1(2, 262, 16)Zero padding (0, 2)
Conv2(2, 258, 16)(16, (1, 5))
Dropout2(2, 258, 16)Dropout 0.6
Zero padding2(2, 262, 16)Zero padding (0, 2)
Conv3(2, 256, 16)(16, (1, 3))
Dropout3(2, 256, 16)Dropout 0.6
Zero padding3(2, 260, 16)Zero padding (0, 2)
Reshape(520, 16)Dimension transform
Bi_LSTM(100)merge_mode = concat
GA_Dense(2, 256, 64)64
Dense1(64)64
GA_Conv4(2, 254, 16)(16, (1, 3))
A_Dense(64)64
GA_Conv5(2, 252, 1)(1, (1, 3))
Multiply1(64)
Flatten(504)
Dropout4(64)Dropout 0.6
GA_Dense2(64)512
Multiply2(64)
Dense2(10)10
FC layer(10)Softmax