Channel Estimation for Secret Key Generation

Abstract

Generating secret keys from radio channel measurements has been shown to allow private communications. Previously described methods for performing wireless key generation have not come close to achieving the theoretical upper bounds on key rates theory for this technique. This paper demonstrates how using a Kalman filter based on an auto-regressive (AR) model for the channel process, channel gain measurements can be converted into a sequence of independent Gaussian vectors. Methods for processing these vectors so they are compatible with existing secret key quantization and error reconciliation techniques are also presented. It is shown how the mutual information in these vectors is near that of the theoretical upper bounds. Finally, it is shown that most of the available secret key bits can be extracted using practical quantization and error reconciliation techniques.