DFCE-Net: Doubly Fractional Channel Estimation via Encoder-Decoder CNNs for OTFS Systems

Orthogonal Time Frequency Space (OTFS) modulation is a promising technology for high-mobility 6G communications, yet its practical implementation faces challenges due to fractional delay and Doppler shifts, which cause energy leakage, SNR loss, and phase rotation mismatches in the delay-Doppler (DD) domain. Traditional compressed sensing methods suffer from grid mismatch and computational inefficiency in such doubly dispersive channels. The proposed CNN-based doubly fractional channel estimation network (DFCE-Net) introduces: (1) geometric-aware encoder-decoder enabling single-stage sparse reconstruction and spread texture learning, (2) multi-scale modules resolving delay-Doppler ambiguities, and (3) phase-constrained hybrid loss. The framework significantly outperforms OMP and SBL in NMSE across multipath scenarios and SNR variations. The proposed framework offers a high-precision, real-time solution for OTFS channel estimation, advancing reliability in high-mobility applications such as vehicular and satellite communications.