Recurrent Neural Network Assisted Equalization for FTN Signaling
Shihao Lai, Mingqi Li · 2020
In this paper, we consider the application of neural network for equalization of Faster-than-Nyquist (FTN) Signaling. First, we formulate the detection problem as a supervised regression task in machine learning framework. Then a recurrent neural network (RNN) called Bi-directional long short-term memory (Bi-LSTM) is proposed to characterize the feature of inter-symbol interference (ISI) introduced in FTN Signaling. Moreover, we describe a “mismatch SNR” strategy for building the training Dataset that can effectively help to prevent overfitting. Numerical results prove that the BER performance of the proposed neural network based detector is close to the theoretical optimal maximum likelihood sequence estimation (MLSE) when symbol rate within the Mazo Limit, and Bi-LSTM could be a more realistic scheme compare with MLSE when symbol rate exceeds the Mazo Limit.