Deep Learning Aided NOMA for User Fairness in 5G

Abstract

User fairness is an important consideration for the upcoming wireless communication systems. Non Orthogonal Multiple Access (NOMA) is a potential promoter of fifth-generation (5G) wireless communication. The advantages of NOMA in 5G are to improve the efficiency of spectrum and increase the number of users. But the main disadvantage of this system is high computational complexity. To overcome this disadvantage, in our work we put forward a new technique of DNN (Deep Neural Network) aided NOMA system. User fairness in the sense of max-min fairness is attained by joint power allocation and beamforming. This is a non-convex problem, because of the higher dimensionality of variables and complexity of the problem. A substandard solution is obtained in the form of perfect power assigning and beamforming vector. From the results, we can clear that the DL aided NOMA has better performance characteristics than the conventional NOMA system.