NERecon: Neural-Enhanced Reconciliation for Secure Physical Layer Key Generation

Physical-layer secret key generation (PKG) is a promising approach for enabling two wireless nodes to establish shared secret keys by leveraging the characteristics of radio channel parameters. A key component of PKG is reconciliation, which ensures that keys generated by the two parties are identical despite noise-induced discrepancies. Traditional reconciliation methods often require multiple iterations, limiting their efficiency and scalability. Recent studies have explored deep learning as a tool for reconciliation, showing initial promise but falling short in evaluating diverse architectures and noise conditions. In this work, we introduce NERecon, a comprehensive benchmark for the neural-enhanced reconciliation method. By employing a deep learning-based framework, NERecon effectively resolves bit mismatches with high precision. Extensive experiments under various noise levels demonstrate the robustness of NERecon, achieving an average key agreement rate of 93.6%, 86.3% and 78.6% at a noise level of 0.1, 0.2, and 0.3 respectively. These results highlight the potential of deep learning to revolutionize reconciliation in PKG systems, paving the way for more secure and efficient wireless communication.