Rate Analysis and Deep Neural Network Detectors for SEFDM FTN Systems

2021/03/03 by Chorti, Arsenia, Picard, David
#FOS: Computer and information sciences #FOS: Electrical engineering #Information Theory (cs.IT) #Signal Processing (eess.SP) #electronic engineering #information engineering

paper · doi:10.48550/arxiv.2103.02306

Abstract

In this work we compare the capacity and achievable rate of uncoded faster than Nyquist (FTN) signalling in the frequency domain, also referred to as spectrally efficient FDM (SEFDM). We propose a deep residual convolutional neural network detector for SEFDM signals in additive white Gaussian noise channels, that allows to approach the Mazo limit in systems with up to 60 subcarriers. Notably, the deep detectors achieve a loss less than 0.4-0.7 dB for uncoded QPSK SEFDM systems of 12 to 60 subcarriers at a 15% spectral compression.

Related

• Deep Learning Based Detection for Spectrally Efficient FDM Systems
• A Novel CNN Based Standalone Detector for Faster-than-Nyquist Signaling
• Receiver Design for Faster-than-Nyquist Signaling: Deep-learning-based Architectures
• Faster-Than-Nyquist Equalization with Convolutional Neural Networks
• Low Complexity Lookup Table Aided Soft Output Semidefinite Relaxation based Faster-than-Nyquist Signaling Detector