Signal Detection Scheme Research And Design Of 3D Massive MIMO Systems Under Non-Gaussian Noise Channel

Wireless communication is in a period of rapid development,users are eager for faster Internet access and instant services,data traffic therefore rapid growth.Researchers propose Fifth Generation(5G)mobile communications technology to deal with these problems.Among them,3D Massive MIMO technology plays a key role.It makes full use of the elevation angle to make the base station more accurate to distinguish users which is based on Massive MIMO.It greatly improves system performance.Signal detection is one of the key issues.At present,there have been many signal detection methods,and most of the signal detection methods are mainly focused on the Additive White Gaussian Noise model,for the case of non-Gaussian Noise,the detection performance may be greatly reduced.Therefore,it is very important to study the problem of signal detection in the case of non-Gaussian noise.In addition,the rise of DL provides a new way to solve a large number of related problems in 3D Massive MIMO.Therefore,this paper studies and designs a signal detection scheme for 3D Massive MIMO systems with impulse noise in combination with deep learning and channel characteristics.The main work of this paper is as follows:First,aiming at the presence of impulse noise in 3D Massive MIMO systems,this paper proposes a deep learning signal detection scheme based on correlation entropy loss function based on channel angular domain sparsity.The deep neural network method is introduced into the signal detection problem of the 3D Massive MIMO system.Then,in view of the problem that the loss function design in the existing deep neural network algorithm is extremely sensitive to impulse noise,the correntropy loss function is used to deal with outliers to detect the signal.The problem is extended from the Gaussian noise environment to the non-Gaussian noise environment,which better solves the signal detection problem in the impulse noise environment.In addition,this paper designs a channel compression method based on the sparsity of the angular domain channel from the perspective of reducing the computational complexity of the signal detection method.A signal detection scheme based on angular domain compression channel in 3D Massive MIMO system is designed,and the feasibility and effectiveness of the above scheme are verified and analyzed through experimental simulation.Second,this paper proposes a deep learning signal detection scheme based on distributed likelihood function and normalized flow network is proposed to solve the problem of non-Gaussian noise signal detection 3D Massive MIMO system.Therefore,a signal detection method based on deep neural network is designed.The method uses the combination of distributed likelihood function and normalizing flow network to recover the signal.Normalized flow network can use a certain priori knowledge of noise to convert the original logarithmic likelihood estimation of distribution to the logarithmic likelihood of known potential variables by designing the potential variables corresponding to noise.The scheme can change the design of the corresponding potential variables when facing different noise conditions.The proposed scheme can solve the problem of signal detection in the channel with different non-Gaussian noises,and greatly improves the performance of 3D Massive MIMO system in the non-Gaussian noise channel.Finally,the effectiveness of the proposed scheme is analyzed by simulation.