NOMA System Performance Improvement Using Chaos and Deep Learning

Abstract

Non-orthogonal multiple access (NOMA) is one of the key technology of 5G system to enhance the capacity and spectral efficiency. However, the fast changing wireless channel makes the ideal power allocation be a challenging task in practice. A chaotic shape-forming filter (CSF) based NOMA system is proposed and a deep learning (DL) based method is used to estimate the user channel gains for power allocation in this paper. The contributions of the work lie in: 1) The CSF and the corresponding matched filter (MF) enhance the noise resistance performance of the NOMA system. 2) The autocorrelation function (ACF) of the superposition signal composed by the chaotic signals generated by the CSF of different users is proved to be the same as the ACF of the base function of the CSF, which is an interesting and important base point to use the previous theoretical result for blind channel identification. 3) For the flat fading channel assumed in the NOMA system, an analytical formula of the channel gain, sole parameter in this type of the channel, with respect to the ACFs of the received signal and the base function of the CSF is derived for the first time, which provides the underlying mechanism to use deep neural network (DNN) for channel gain prediction. Simulation results show that 1) the chaos-based NOMA using the CSF and corresponding MF achieves better performance as compared to the conventional binary phase shift keying NOMA (BPSK-NOMA) with root-raised cosine (RRC) filter; 2) the DNN with simplified structure and the less input neuron as compared to that for the frequency selective fading channel is capable of achieving superior performance for both blind channel identification and NOMA system in the sense of low bit error rate (BER).