Radio-Frequency Fingerprint Identification (RFFI) is crucial for device authentication, utilizing unique RF fingerprints resulting from hardware-level imperfections to enhance wireless communication security. However, conventional RFFI approaches face significant limitations in addressing domain shift phenomena, particularly within resource-constrained IoT deployments. This study introduces CCWNet, an innovative one-dimensional (1D) deep learning framework that integrates multi-scale wavelet transformations with complex-valued convolutional operators to deliver robust and computationally efficient RFFI capabilities. The systematic integration of Dual-Tree Complex Wavelet Transform (DTCWT) within the network architecture enables effective extraction of fine-grained, non-modulated RF fingerprints while maintaining computational efficiency. Rigorous empirical evaluations utilizing real-world WiFi and ADS-B datasets substantiate CCWNet’s consistent superiority across both short-term and extended temporal cross-domain scenarios. Notably, CCWNet attains 99.55% accuracy in same-domain evaluations while sustaining 85.60% accuracy under cross-domain conditions throughout a 40-day observation period. Furthermore, the architecture achieves inference times of approximately one second per sample on Raspberry Pi hardware. These results underscore CCWNet as a highly accurate, lightweight, and practical end-to-end solution for edge-deployed IoT security.