A novel deep learning automatic modulation classifier with fusion of multichannel information using GRU

EURASIP Journal on Wireless Communications and Networking

Table 1 Configuration of the proposed architecture

	Output volume	Description
input	N \(\times\) 4	N = 128, 256, 512, 1024
GRU1	N \(\times\) 128	128 hidden neurons
GRU2	128 \(\times\) 1 \(\times\) 128	128 hidden neurons
Conv1	32 \(\times\) 1 \(\times\) 62	\((32,1,7),s=2,p=1\)
Pooling1	32 \(\times\) 1 \(\times\) 31	max-pooling, \(s=2\)
Conv2	64 \(\times\) 1 \(\times\) 14	\((64,1,7),s=2,p=1\)
Pooling2	64 \(\times\) 1 \(\times\) 7	max-pooling, \(s=2\)
Conv3	128 \(\times\) 1	\((128,1,7),s=1,p=0\)
FC1	64	Fully connected with 64 neurons
FC2	24	Fully connected with 24 neurons