Pilot-Free Semantic Communication Systems for Frequency-Selective Fading Channels

With the paradigm shift from traditional communication systems to machine learning-driven communication systems, deep learning-based semantic communication is considered as a potential approach to improve transmission efficiency. However, the time-varying and multipath effects of wireless channels significantly influence the performance of existing semantic communication systems. The conventional solutions to overcome channel impairments, i.e., pilot-driven channel estimation and channel equalization approaches, will complicate the system and introduce extra overhead. Therefore, this paper proposes a deep learning-driven pilot-free semantic communication framework for frequency-selective fading channels. The convolutional neural network-based channel feature extraction and data recovery modules are introduced in the receiver to implicitly extract channel information directly from the received signal and recover the semantic information. The simulation demonstrates that the proposed network can be applied to various frequency-selective fading channels and outperforms existing schemes in terms of computational complexity and semantic transmission accuracy.