Abstract: Radio frequency fingerprint identification (RFFI) has been widely adopted in IoT device management and wireless security due to its ability to authenticate devices based on hardware-induced imperfections without higher-layer dependencies. However, its growing identifiability also raises concerns about potential privacy leakage. To address this, this paper proposes a radio frequency fingerprint concealment (RFFC) framework to suppress the exposure of device identity features. Furthermore, a novel RFF modeling network based on the vector decomposed gated recurrent unit (VDGRU) is developed. This network reconstructs the original baseband signal by modeling the inverse mapping of the transmitter chain and reuses the inverse model as a forward compensator to embed inverse RFF features, thereby mitigating fingerprint characteristics after transmission. The approach employs amplitude-driven GRU modeling for nonlinear distortion and integrates a phase reconstruction mechanism to enhance both concealment and signal fidelity. Experimental evaluations across multiple transmitters, signal-to-noise ratios, and multipath channel conditions demonstrate that the proposed method significantly degrades identification accuracy in various mainstream RFFI systems. It also achieves superior concealment in terms of confusion matrices, feature distribution visualization, and signal linearity metrics, exhibiting strong generalization and robustness.