Signal Recovery Technique Using Recurrent Neural Network in Interference Environment

The interference management is getting more important in 5G and beyond 5G systems because cell density increases significantly and multiple types of cells are considered. High interferences significantly degrade the spectral efficiency as well as energy efficiency. The interference mitigation techniques will be the main research challenges in 5G and beyond 5G heterogeneous networks. Deep learning is one of new bloods in beyond 5G system. Many research groups are investigating to apply deep learning in physical layer and network layer. We expect to improve the network performance as well as create new services in beyond 5G systems. Recurrent neural network (RNN) is suitable for predicting data in time sequence. In this paper, a novel signal recovery technique using RNN is proposed for mitigating interferences. The purpose of the proposed algorithm is to recover the received signals that are wiped out by interference. The performances of the proposed technique are evaluated and analyzed. In the simulation, we predicted the lost 50 subcarriers of OFDM channel estimation symbols. After having enough training of LSTM network, we obtained the RMSE value 0.24596 between the predicted value and the observed value.