A Research And FPGA Implementation Of Channel Estimation Based On Deep Learning In SCMA System

Sparse code multiple access(SCMA)technology extremely improves the quality of communication system,which is regarded as the candidate of the fifth generation of wire-less communications.Channel estimation is the important process in SCMA communi-cation system.The accuracy and complexity of channel estimation have a great impact in the whole system.The SCMA system is sensitive to the precision of channel estima-tion,but the traditional and classic channel estimation algorithms have high complexity and low accuracy,which leads to the limitations of these algorithms in the application of SCMA.It is neccessary to study more effective channel estimation algorithms which is used for SCMA system.Deep learning has lots of advantages,such as low computational complexity and high-speed parallelism.At present,the combination of deep learning and wireless communication is a promising and developing research direction.Therefore,more and more researches have achieved remarkable results in this direction.Inspired by deep learning technology,this paper attempts to introduce this method into channel estimation in SCMA communication system.Firstly,the deep learning theory is introduced in this paper,the channel estimation algorithms are studied in the existing literature,both advantages and disadvantages of the mainstream channel estimation algorithms are summarized,and the channel estimation technology are introduced based on deep learning into the SCMA communication system on the basis of fully understanding the characteristics of each algorithm.This arcticle re-gards the channel estimation in SCMA as regression fitting in depth learning.The neural network learns the nonlinear mapping relationship between the training sequence and the channel estimation parameters,so that the network can track and predict the channel state according to the training sequence in the receiving channel to complete the task of accurate channel parameter estimation.According to the relevant algorithm,the paper designed the structure of the neural network competely,which includes BP neural network and convo-lutional neural network.In order to meet the demands of the hardware implementation and the data-format in the wireless communication system,this paper focused on optimizing and promoting the structure of the networks.Whatsmore,the communication simulation software is used to describe BP neural network and convolutional neural network,and an application scenario of the SCMA communication system using these neural networks in multi-path channel.The final simulation results show that the designed channel estimator can predict the wireless channel state in SCMA communication,and the channel estima-tion algorithm based on deep learning can maintain a low computational complexity in comparison with the classic channel estimation algorithm.Above,further improve the performance of SCMA communication system.While verifying the application of deep learning technology in 5G,this article pays more attention to the practical feasibility of this optimization method.Therefore,this paper uses vivado and Modelsim design software to complete the related function imple-mentation and verification.Firstly,the implementation process of the key modules of the FPGA to realize SCMA communication system is introduced.Then,the FPGA structure and design skills of the deep learning channel estimator are introduced in detail.Finally,the two modules are cascaded to simulate the test and compared with the simulation results of the software.The simulation results show that the SCMA communication system can meet the requirements of the BER,timing and resource consumption.