基于卷积神经网络自编码器结构的空时分组传输方案

为提高现有端到端通信系统的泛化能力和可靠性，提出了一种基于卷积神经网络的空时分组码（Space-Time Block Coding，STBC）多输入多输出通信系统物理层方案。该方案将通信系统物理层表述、调制和解调过程联合起来形成端到端自编码器系统，引入多层一维卷积层，分别构建发射机和接收机，并扩展为多天线模式。为进一步提高系统可靠性，合理规划网络结构和参数，联合信号的调制和编码方案，优化了系统模型。仿真实验表明，针对瑞利相关衰落下多输入多输出（Multiple-Input Multiple-Output，MIMO）信道应用场景，训练模型可以实现传统STBC系统的误码性能，两发两收系统在发送端相关系数为0和0.9时分别优于传统系统0.5 dB和1 dB。此外，经过优化后的系统可获得采用卷积编码的性能改善效果，其两发两收不同工作方式优于传统1/2码率卷积编码STBC系统1~3 dB。

A Transmission Scheme of CNN-based Autoencoder for STBC Systems

In order to improve the generalization ability and reliability of existing end-to-end communication systems,a physical layer scheme of space-time block coding(STBC) multiple-input multiple-output(MIMO) communication systems based on convolutional neural network(CNN) is proposed.This scheme combines the physical layer representation,modulation,and demodulation processes of communication systems to form an end-to-end autoencoder system.In the autoencoder structure,multiple layers of one-dimensional convolutional layers(Conv1D) are introduced to construct the transmitter and receiver respectively,and expanded to multiple antenna mode.The network structure and parameters are reasonably planned to further improve the system reliability.The modulation and coding schemes of the signals are combined to optimize the system model.Simulation experiments show that for MIMO channels application scenarios based on Rayleigh related fading ,the training model can achieve the error performance of the traditional STBC system.The two-transmitting and two-receiving system outperforms the traditional system by 0.5 dB and 1dB when the correlation coefficients at the transmit end are 0 and 0.9,respectively.Furthermore,the optimized system can obtain the performance improvement effect of using convolutional coding,which is about 1??3 dB better than the traditional 1/2 code rate convolutional coding STBC systems in the case of various modes for two transmitters and two receivers.