Open World Lifelong Radio Frequency Fingerprint Identification Based on Stacked Structure

As an important device identity authentication technique, radio frequency fingerprint identification (RFFI) has been used in various wireless networks to enhance wireless security. Most of the recent studies follow a closed-set and static assumption, however, in practical open world scenarios, new unseen devices may emerge, and the recognition model needs to evolve incrementally. This paper proposes a novel stacked open world lifelong RFFI framework SOWL. By serially stacking a modified intra-class splitting method that integrates uncertainty and confidence measurements and a hybrid lifelong learning method that cooperates regularization and dynamic structure, open world lifelong RFFI is achieved. Moreover, we further propose a parallel stacked structure based on forward and backward transfer between previous models and the new model, which can greatly reduce the calculation consumption and make the RFFI system more practice-oriented. Numerical experimental results on real collected signal dataset demonstrate the effectiveness of our proposed framework.