基于深度学习的MIMO系统联合优化

Deep learning-based joint optimization for MIMO systems

自动编码器神经网络可将通信系统重新构建为端到端的任务,从而实现整个系统的联合优化。针对基于深度学习的2用户与4用户多输入多输出(multiple-input multiple-output,MIMO)系统联合优化问题,提出将自动编码器运用到系统中,将整个通信系统的发射端和接收端视为自动编码器的编码和译码部分,利用交叉熵损失加权和函数进行训练学习,从而获得优化的系统模型,并进一步分析得出每个用户的误比特率及所有用户的平均误比特率。实验结果表明,基于自动编码器所构建的MIMO通信系统相比于传统的通信系统具有更优的系统性能。

Reconstructing the communication system as an end-to-end task,the autoencoder neural network enables the joint optimization of the whole system. In order to deal with the joint optimization problem of MIMO( multiple-input multiple-output) systems based on deep learning with two users and four users respectively,the autoencoder was applied to the communication system. The encoding and decoding parts of autoencoder were considered as the transmitter and receiver of the overall communication system respectively. The cross-entropy loss weighted sum functions were applied to train and learn,so that the optimized system model was obtained and the bit error ratio of each user and average bit error ratio of all users were further analyzed and achieved. Compared with the traditional communication system,the experiment results show that MIMO communication system based on the autoencoder demonstrates better system performance.