Channel equalization has been widely used to eliminate inter-symbol interference (ISI) and improve transmission performance in fading channel. In this paper, we propose a novel model of joint channel equalization and decoding based on recurrent neural network (RNN) in order to recover information messages interfered by channel distortion. By returning the output of decoder to the input of equalizer, an iterative equalizing and decoding process is achieved. Simulation over linear channel shows our method offers performance near that of maximum likelihood (ML) equalizer with knowledge of perfect channel state information (CSI). With less than 2/3 of the parameters, the proposed model has more than 0.5 dB gain over the CNN + NND-Joint model (and three other models) over nonlinear channel.
Paper
Full text
Joint Channel Equalization and Decoding with One Recurrent Neural Network
Semantic Scholar · Computer Science · 2019
Abstract
Channel equalization has been widely used to eliminate inter-symbol interference (ISI) and improve transmission performance in fading channel. In this paper, we propose a novel model of joint channel equalization and decoding based on recurrent neural network (RNN) in order to recover information messages interfered by channel distortion. By returning the output of decoder to the input of equalizer, an iterative equalizing and decoding process is achieved. Simulation over linear channel shows our method offers performance near that of maximum likelihood (ML) equalizer with knowledge of perfect channel state information (CSI). With less than 2/3 of the parameters, the proposed model has more than 0.5 dB gain over the CNN + NND-Joint model (and three other models) over nonlinear channel.