Dimensionality Reduction in Wireless Physical Layer Key Generation

Achieving security in the wireless networks by extracting secure bits using physical layer key generation is one of the emerging ideas. However, for the networks with the resource constraint nodes, the security algorithm should be less computationally complex. This can be achieved by reducing the number of numerical computations or by reducing the dimensionality of input data. In this work, we propose to apply dimensionality reduction by the principal component analysis on the input vector of received signal strength for key generation. We select the principal components extracted from the input vector based on information content & cross correlation, and instead of considering all the column of input vector we consider the column vectors corresponding to a few dominant principal components only, for secure key generation. The bit disagreement performance of the system is improved along with the reduction in several numerical calculations corresponding to the rejected input vectors at the cost of few extra computations. This reduces the computational power significantly as verified by simulation and experimental results with sufficient key randomness.