A Survey on Noise Detection and Correction in Semantic Communication
Seongjin Choi, Junsuk Oh, Tung Son, Seonghun Hong, Ju‐Young Kim, Sungrae Cho · 2025
Semantic Communication (SemCom) has gained significant attention as a communication paradigm that reduces data transmission overhead while ensuring effective semantic information transfer. However, unlike noise in traditional communication systems, SemCom introduces new types of semantic noise, including semantic distortion, knowledge mismatch, and adversarial perturbations. This paper investigates semantic noise detection and correction techniques in SemCom and explores key approaches, including deep learning-based semantic restoration, knowledge-driven error detection, and adaptive coding schemes such as Hybrid Automatic Repeat Request (HARQ). Through this study, it is demonstrated that deep learning and knowledge-driven approaches enhance the robustness and efficiency of SemCom. Experimental evaluations using BLEU and BERT Score confirm that the proposed techniques improve the reliability of SemCom compared to conventional methods