利用并联CNN-LSTM的调制样式识别算法

为了提高基于卷积神经网络的调制样式识别算法性能，利用CNN的空间特征提取能力和LSTM时序特征提取能力，设计了CNN-LSTM并联网络，上支路由一层卷积层和一层池化层组成，下支路使用单层LSTM网络。直接将同向分量和正交分量作为输入数据，上下支路提取信号的空间和时间特征，提高特征表达能力。对BPSK、QPSK、8PSK、16QAM、32QAM、16APSK、32APSK 等7种信号的调制样式识别仿真实验结果表明:算法无需人为设计特征参数，减少人为因素影响，同时该算法在低信噪比下具有较好的识别性能。 

Modulation Recognition Algorithm By Usign Parallel CNN-LSTM

To improve the performance of modulation type recognition algorithm based on convolutional neural network, CNN-LSTM parallel network is designed by CNN spatial feature extraction ability and LSTM time series feature extraction ability. The upper branch consists of a pooling layer and a convolution layer, and the lower branch uses a single-layer LSTM network. The in-phase component and quadrature component are directly used as input data, and the upper and lower branches extract the spatial and temporal characteristics of the signal respectively, to improve the feature expression ability. The experimental results of modulation type recognition for 7 kinds of signals, such as BPSK, QPSK, 8PSK, 16QAM, 32QAM, 16APSK and 32APSK, show that the algorithm does not need to artificially design the characteristic parameters and reduces the influence of human factors. At the same time, the algorithm has good recognition performance at lower SNR.