Adaptive Pulse Shaping and Equalization for OFDM in Time-Frequency Doubly Selective Channels

Orthogonal Frequency Division Multiplexing (OFDM) underpins modern wireless communication due to its resilience against multi-path fading and computationally efficient implementations. However, on one hand, in scenarios with time-frequency doubly selective fading channels, OFDM systems face significant challenges, as channel variation induces inter-carrier interference (ICI) that disrupts subcarrier orthogonality. On the other hand, the limited length of the cyclic prefix (CP), often constrained to a fraction of the symbol duration, may be insufficient to fully mitigate inter-symbol interference (ISI) in scenarios with large time spreads. Extending CP length would reduce spectral efficiency and increase latency, making it incompatible with the demands of high-efficiency, low-latency systems. In this paper, we propose an autoencoder based OFDM architecture integrating adaptive pulse shaping with an equalization neural network (PS-EQNet) that jointly addresses ISI caused by insufficient CP and ICI due to channel dynamics through learnable time-frequency filters. Additionally, we introduce a detection network tailored for simplified channels, which reduces computational complexity compared to existing schemes while maintaining robust performance. Simulation results confirm that the proposed PS-EQNet OFDM system significantly relaxes CP requirements, enhances spectral efficiency (SE), and achieves reliable bit error rate (BER) performance in time-frequency selective channels, establishing a flexible trade-off among SE, BER, and computational complexity.