MIMO Precoding Methods Based On Deep Learning

Multiple-Input Multiple-output(MIMO)can availably enhance the throughput and reliability of wireless communication systems through its spatial multiplexing and diversity capabilities,it is one of the key technologies in the new generation of mobile communication systems.Precoding can effectively suppress interference between data streams by pre-processing the signal to be sent at the transmitting end,thereby improving system capacity and resource utilization.Therefore,precoding is an important technical means for MIMO system performance to be realized.Traditional precoding technology generally uses statistical analysis and advanced signal processing technology to design fixed algorithms,which have poor dynamic adaptive ability to the environment and cannot fully meet the requirements of the next generation wireless communication system(the 6-th Generation Mobile communication system,6G)intelligent requirements.In recent years,with the successful application of Deep Learning(DL)technology in the fields of image recognition,natural language processing,and speech recognition.The combination of deep learning and wireless communication systems to build intelligent mobile communication systems has received widespread attention.Deep learning technology is essentially a data-driven method.It has the ability to automatically condense information characteristics from massive amounts of data,which in turn enables the communication system to have the ability to adapt and evolve.It has great advantages in reducing the complexity of communication system design and improving the performance of communication systems.potential.In this context,this article focuses on deep learning-based MIMO precoding transmission technology.The main research contents include:(1)Contraposing at the point-to-point MIMO system,a deep learning intelligent transmission structure is presented that combines the transmitter precoding network and the receiver RTN(Radio Transformer Networks,RTN)network.In the case of discrete character set input,the structure can use the given data for training,automatically adjust the parameters of the transmitter network and the receiver network,and intelligently adjust the transmitter and receiver constellations according to the channel state,so that the constellation points are automatically isolated to achieve adaptive suppression of interference among multi-antenna data streams.(2)The above-mentioned deep learning intelligent transmission structure of the transmitting end precoding network and the receiving end RTN network is further extended to a multi-user MIMO broadcast channel network and an interference channel network,and a deep learning intelligent transmission structure suitable for multi-user broadcast and interference channel scenarios is constructed respectively.Similar to the point-to-point MIMO system,the proposed multi-user deep learning structure can pre-process the data streams of multiple users through the precoding network,so that the receiving end of each user can better separate the constellation points,thereby effectively suppressing Inter-user and inter-stream interference makes the multi-user MIMO system realize transmission performance guarantee through independent learning.(3)Contraposing at the traditional SVD(Singular value decomposition,SVD)precoding algorithm,this paper proposes a SVD precoding approximation algorithm combined with convolutional neural network.The algorithm inputs the channel state information into the convolutional neural network,and obtains the low-rank approximation matrix of the channel matrix at the output end.Thereby approximating the optimal precoding matrix of the SVD precoding algorithm.At the same time,the realized sum rate is close to the optimal sum rate of the traditional SVD algorithm.