Research On Deep Learning-based Channel Estimation And Signal Detection In MIMO Systems

Multiple-input multiple-output(MIMO)technology can greatly increase channel capacity and transmission rate without increasing the channel width,so it is widely used in various scenarios.In order to take full advantage of MIMO,efficient channel estimation and signal detection are necessary.Most traditional communication algorithms seek a balance between complexity and accuracy,thus there is an urgent need to introduce new methods for exploration.At this time,deep learning has played an important role in many fields.The introduction of this method into wireless communication systems to promote artificial intelligence wireless communication technology has become a new generation of research boom.In view of the advantages of deep learning in various aspects of the wireless communication field,this article takes multi-user MIMO system as the research object,and uses deep learning to solve the channel estimation and signal detection problems in the MIMO system.In order to further improve the performance of MIMO signal detection,this thesis uses the traditional deep learning network to jointly solve the problem of channel estimation and signal detection in the physical layer of the MIMO system.First,a channel estimation and signal detection network based on data-driven Full Con is designed.The transmission signal is directly recovered through the received pilot signal and data signal.This method only needs to implicitly estimate the channel,thereby avoiding additional errors in the channel estimation link.Experimental results show that the signal detection performance of the network is better than traditional detection algorithms,and no statistical information of noise is required.The data-driven Full Con network likes a black box model,and its principle is to obtain the ability to recover the signal through a large amount of data training.In order to be able to combine the existing technology of communication,this thesis proposes a MIMO channel estimation and signal detection network based on data-driven network named Md Net.Md Net is composed of two subnetworks,the channel estimation sub-network and the signal detection sub-network.In the channel estimation sub-network,the lowest-square estimation method is used to obtain preliminary rough estimation channel information,and then the noise is filtered through the deep neural network net1 to obtain more accurate channel information.In the signal detection sub-network,the Md Net signal detection sub-network uses the idea of deep learning network hierarchy to simulate the expanded projection gradient detection algorithm,and obtains higher detection performance by adding trainable neural network parameters.The simulation results show that,when the number of antennas is small or the signal-to-noise ratio is low,Full Con has a powerful noise filtering function,and its bit error rate performance is the best.However,with the increase in the number of MIMO antennas,Full Con must increase the training parameters a lot to improve the detection performance,which makes the training speed of Full Con decrease,and Md Net shows excellent characteristics at this time.Since the Md Net training parameters are only related to the number of training layers,in the case of a large number of antennas,fewer training parameters can still be maintained,so the training can be completed quickly,and the bit error rate performance is better than traditional algorithms.