Compared with orthogonal multiple access,non-orthogonal multiple access technology has attracted widely attention because of its high spectral efficiency and high channel capacity,which can be applied in various sceneries of 5G.Non-orthogonal multiple access technology allows interference between transmitted signals of different users,so that more user access can be supported on an equal number of orthogonal resources,and the receiver removes channel noise and interference of other user signals with a specific detection algorithm,and restores the original signal of each user.At present,there are many non-orthogonal multiple access technologies,which can be mainly classified into power domain and codeword domain.The non-orthogonal multiple access technology of codeword domain can flexibly support different configurations of users and resources.In this paper,we focus on the non-orthogonal multiple access technology of codeword domain,and study the low density signature code division multiple access(LDS CDMA)and sparse code multiple access(SCMA)technology.At the same time,combined with artificial intelligence technology,we have studied the off-line SCMA codebook generation performance of self-encoder through neural network.LDS CDMA is a non-orthogonal multiple access technology of codeword domain that implements multiuser non-orthogonal resource allocation using a specific sparse signature.Although different users cannot be allocated completely orthogonal signature,with user interference between transmission signals,a less complex belief propagation algorithm can be applied at the receiver for detection,since each signature is a sparse sequence.The signature matrix has an important impact on system performance,and the signature sequences of different users are designed through joint optimization.Joint signature matrix design is a complex optimization problem,including the user's transmission signal on different resources and the value of the complex factor in the sequence,especially in the case of an increase in the number of users.In this regard,this paper proposes a sparse signature matrix design method based on quasi-cyclic(QC)matrix expansion of the optimized signature of small-scale users,which can improve the maximum number of users supported by the whole system under the condition that the ratio of user number to resource number is fixed.At the same time,we found that in the case of an increase in the number of users,a larger dimension of the sparse signature matrix can effectively improve system performance.SCMA is a new non-orthogonal multiple access method for codeword domain,which is a broader extension of LDS CDMA.Similarly,a message passing algorithm can be applied at the receiver to detect different users' signals.The difference is that SCMA combines the user's modulation with spreading process,assigns a codebook to each user,and obtains a new gain.Similarly,its codebook design is a complicated problem.This paper proposes a new codebook design method based on QAM decomposition,reasonably arranges the participating users and their sub-constellations on each resource,and further correction of codes for different users.This method can not only achieve better performance,but also have a lower complexity detection algorithm.The LDPC code is a channel coding method close to the Shannon limit,which can effectively improve the correct rate of recovering user signals at the receiver of the non-orthogonal multiple access system.In addition,in joint non-orthogonal multiple access detection and LDPC decoding models,the introduction of outer iterations can further improve overall system performance.Artificial intelligence(AI)has shown great potential in more and more fields,and it also has many integration points with communication technology.By applying the neural network tool to the design problem of the codebook of SCMA system,with computer randomly generating the information bits of the users,the codebook with better performance can be obtained through offline learning.In this paper,by simulating the codebook obtained by the neural network self-encoder under the parameters of different user numbers and different modulation orders,and using the message passing algorithm to detect,it is found that the neural network self-encoder method has higher robustness and the obtained codebook has better performance in some cases than the codebook designed by the traditional method. |