A Superimposed Pilot Design Method for Semantic Transmission Over Doubly Selective Channel

Semantic communication has emerged as a promising paradigm, attracting significant research interest due to its potential to enhance communication efficiency. This paper extends semantic communication to doubly selective channel scenarios, addressing the challenges of time-varying and frequency-selective fading. Inspired by the noise resilience of semantic transmission and the principle of semantic inequality, a superimposed pilot design method is proposed for semantic transmission over doubly selective channels. An optimization problem is theoretically analyzed to determine the optimal power allocation ratio between pilot and semantic data symbols under channel estimation errors in doubly selective channels, without considering semantic importance. This analysis provides insights into pilot power and coherence time, serving as a benchmark for semantic performance optimization. Additionally, considering semantic symbol inequality, a suite of deep learning-based models including semantic interleaving, power adaptation, and channel compensation are designed to enable unequal power allocation that prioritizes important semantic symbols, maximizing semantic performance. Numerical results validate the effectiveness of this method, demonstrating substantial improvements in transmission quality over traditional orthogonal pilot schemes in doubly selective channels, particularly under low SNR conditions. Execution time analysis further highlights the computational efficiency of this approach, achieving a favorable balance between performance and complexity.