Long-Short term Memory based Channel Prediction for SISO System

Learning the characteristics of a wireless channel is one of the most fundamental and challenging issues in the wireless communication. Many conventional signal processing techniques have been developed so far to estimate the channel state information. In this paper, we present a novel approach based on deep learning using long short term memory (LSTM) neural network to predict the fast-varying Rayleigh fading channel. Having known application of LSTM neural network in the field of time-series prediction problems, we have applied LSTM to predict the future state of the channel by providing the past channel state values. The predicted values of channel is used to recover the transmitted symbol from noisy signal received at the receiver end. We have optimized the configuration of LSTM network by correctly tuning the hyper-parameters so that we get a correct prediction of channel which is in turn used for reconstruction of signal at receiver end. The simulated Bit Error Rate vs. Signal-to-Noise ratio plot proves that LSTM based channel predictor works better than currently used interpolation based channel predictor.