FTN-Based Non-Orthogonal Signal Detection Technique With Machine Learning in Quasi-Static Multipath Channel
Myung-Sun Baek, Eui‐Suk Jung, Young Soo Park, Yong-Tae Lee
Abstract
Faster-than-Nyquist (FTN) signaling is a useful communication technique for high spectral efficiency. However, since orthogonality between symbols is destroyed when a symbol rate faster than the Nyquist rate is used, intersymbol interference (ISI) is inevitably generated. Interference cancellation and signal detection processes are required to reduce the effect of ISI on the FTN receiver. Furthermore, FTN signaling in the multipath fading channel is a complicated problem because nonlinear interference from FTN signaling is combined with the additional nonlinear interference from the multipath fading channel. This paper proposes a deep learning-based signal detection technique for the FTN signal received through a multipath fading channel. The proposed technology considers the normal FTN signaling-based transmitter without additional signal processing schemes such as precoding and power allocation. We designed a recurrent neural network (RNN)-based deep learning structure and applied it to the FTN signaling-based system. For the RNN structure, both unidirectional long short-term memory (Uni-LSTM) and bidirectional LSTM (Bi-LSTM) architectures were considered. Simulation results showed that both Uni-LSTM and Bi-LSTM performed better than the conventional BCJR algorithm when the FTN factor <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\tau \,\,=$ </tex-math></inline-formula> 0.8 and 0.7. Furthermore, the Bi-LSTM architecture achieved the best performances in all simulations. The proposed LSTM-based FTN detector can detect transmitted symbols without estimating channel coefficients and SNR.