Deep learning (DL) based automatic modulation classification (AMC) has gained popularity for next-generation wireless communication systems. However, these DL-based AMC models are vulnerable to adversarial examples, which can cause false predictions with high confidence, leading to unreliable and non-robust communication networks. In this paper, we propose a data mapping-based adversarial defense scheme to address this issue. This scheme uses random split, time-domain flips, and phase rotations as three methods of data mapping on the input examples, effectively mitigating the impact of adversarial perturbations on the model's output and ensuring reliable model inference. Evaluation results on the RML2016.10a dataset demonstrate that the proposed defense scheme can effectively resist various white-box attacks and improve the robustness of the AMC model without requiring fine-tuning or incremental training. This scheme therefore offers a secure solution for intelligent communication networks.