Deep learning for channel estimation in non-orthogonal multiple access scheme

Abstract

Non-orthogonal multiple access (NOMA) has a great potential in the fifth generation (5G) communication systems and has drawn increasing attention because of the capability of increasing spectral efficiency and supporting the large number of connections. However, the unsteady channel characteristic of wireless communication system has severely restricted the performance of NOMA system. The conventional channel estimation method cannot guarantee real-time detection of the sharply changing channel conditions. In addition, the high computing complexity and overhead should also be taken into account in practical implementation. In order to break these limitations mentioned above, a novel deep neural network (DNN) aided NOMA system is proposed in this dissertation, introducing deep-learning (DL) technology into existing NOMA systems. The DNN could not only substitute some communication modules such as encoder, detector, etc. but also act as a channel estimator which could acquire the perfect channel state information (CSI) in a rapidly changing channel environment. The introduction of DL technology reduces the computation complexity and improves the performance of NOMA system.

Index Terms: Non-orthogonal multiple access (NOMA), channel state information (CSI), deep learning (DL), deep neural network (DNN)