Channel Estimation and Equalization based on Deep BLSTM for FBMC-OQAM Systems

Channel estimation and equalization is one of the challenges of the filter bank multicarrier (FBMC) systems because of the existence of intrinsic imaginary interference. In this paper, we try to solve this challenge from a learning-based perspective. Based on the study of the intrinsic relationship between bidirectional long short-term memory (BLSTM) recurrent neural networks (RNN) and the FBMC radio signals, we propose a novel deep BLSTM network based channel estimation and equalization scheme, abbreviated as BLSTM-CE scheme. In the BLSTM-CE scheme, the input and output of some traditional independent modules are seen as an unknown nonlinear mapping and use a deep BLSTM network to approximate it. Numerical simulation shows that our proposed BLSTM-CE scheme can obtain perfect channel estimation performance in the FBMC systems.