Deep Bi-LSTM Detection for FTN-GFDM in Underserved Communication Scenarios

This work proposes a deep Bi-LSTM for symbol detection in FTN-GFDM systems. The estimator learns the nonlinear mapping between matched filter outputs and transmitted symbols, handling ISI and colored noise. The network is trained offline using synthetically generated data and evaluated over AWGN and TIFS channels. Results show that the Bi-LSTM achieves competitive BER performance compared to the SD, while offering fixed and low complexity during inference, making it suitable for real-time and resource-constrained applications.