Learnable Pilot Design for OFDM-based Deep Joint Source-Channel Coding
Recent advances in deep joint source-channel coding (DeepJSCC) have shown strong robustness to channel variations in wireless image transmission. However, its integration with orthogonal frequency-division multiplexing (OFDM) typically relies on fixed pilot patterns that are not optimized for receiver- side channel processing. This work introduces Pilot-Efficient DeepJSCC-OFDM (PE-JSCC), which learns pilot placement under a strictly fixed number of pilot symbols via a Gumbel- Top-k-based selector. The learned pilot placement improves end- to-end performance over the conventional Block-JSCC, Comb- JSCC, and Scatter-JSCC baselines. To support high mobility, the receiver incorporates a lightweight axial-attention-based inter- carrier interference (ICI) canceler and a CNN-based channel estimator. Simulation results show that PE-JSCC consistently outperforms fixed pilot patterns across diverse SNRs and remains robust over a wide mobility range.