Robust-SemCom: An Adversarial-Resilient Semantic Communication Framework for Contextual Text Transmission

• We leverage LLaMA for both semantic encoding and
decoding to capture deep contextual dependencies and
to mitigate adversarial attacks in semantic communication.
• To address the limitations of strict lexical matching in
Token Error Rate (TER) and Jaccard Similarity, we
augment these metrics to prioritize semantic consistency
over literal word-for-word comparison.
• The performance of the proposed system is rigorously
evaluated across both ideal channels and adversarial attack environments to demonstrate its resilience against
semantic noise and disturbances.