DCGAN信道下的端到端通信系统设计

针对通信系统中长序列建模存在维度诅咒的问题，提出了一种基于深度卷积生成对抗网络(Deep Convolutional Generative Adversarial Networks, DCGAN)信道建模的端到端通信系统改进方案。该方案将卷积神经网络(Convolutional Neural Network, CNN)和条件生成对抗网络(Conditional Generative Adversarial Network, CGAN)结合，利用CNN与全连接层(Fully Connected Layer, FC)的局部连接特性对传输长序列的信道进行建模。通过对参数重新设计及网络结构调整，获得了适应不同调制方式和信道类型的学习网络，将其应用端到端通信系统中，作为收发机之间梯度反向传播的桥梁。仿真实验表明，改进的DCGAN能够以减小的网络规模以及计算量成功地实现长序列建模，并且表现出良好的泛化能力。此外，将建模结果运用到端到端通信系统设计中，可以获得与传统数字调制系统相近的误比特率性能。

Design of an end-to-end communication system in DCGAN channel

For the problem of the curse of dimensionality in long-sequence modeling in communication systems,an improved end-to-end communication system design scheme based on Deep Convolutional Generative Adversarial Network(DCGAN) channel modeling is proposed.The solution combines the Convolutional Neural Network(CNN) with Conditional Generative Adversarial Network(CGAN),and uses the local connection characteristics of CNN and Fully Connected Layer(FC) to model the channel that transmits long sequences.By redesigning the parameters and adjusting the network structure,a learning network that adapts to different modulation methods and channel types is obtained and applied to the end-to-end system where the CNN network is used at the transceiver end as a bridge for gradient back propagation between the transmitter and the receiver.Simulation experiments show that the improved DCGAN network can successfully implement long-sequence modeling with a reduced network scale and computational complexity,and has shown good generalization capabilities.In addition,by applying the modeling results to the end-to-end communication system design,a bit error rate performance similar to that of traditional digital modulation system can be achieved.