Keys for symmetric encryption can be established by generating key bits from the mutually-observed randomness present within estimates of a reciprocal wireless channel. We consider this approach for key establishment in multiple-input multiple-output (MIMO) systems and provide a channel estimation strategy that increases the key rate, or the mutual information between the channel estimates at the two nodes. The technique applies MIMO transmit and receive beamformers derived from the eigenvectors of the covariance matrix and then iteratively optimizes the power allocated to each transmit beamformer. Simulation results demonstrate the iterative behavior and the achieved performance of the proposed algorithm.

Establishing secret keys from the commonly-observed randomness of reciprocal wireless propagation channels has recently received considerable attention. We consider such key establishment between two multiantenna nodes that use beamforming for communication, showing that the upper bound on the number of key bits that can be generated from the channel observations can be maximized by properly probing the channel. Specifically, we demonstrate that the eigenvectors of the channel spatial covariance matrix should be used as beamformer weights during channel estimation and we optimize the energy allocated to channel estimation for each beamformer weight under a total energy constraint. Finally, by assuming that the channel covariance is separable, we illustrate implementation of the technique for practical beamforming systems and more advanced signal models that incorporate antenna mutual coupling.

The performance of techniques for establishing secret encryption keys from the reciprocal electromagnetic propagation between two wireless radios is generally measured by comparing the number of key bits generated to the key rate, or the information theoretic bound on the number of key bits that can be determined for a single independent channel estimate. While typically this bound is computed by assuming that the radios have no information regarding the channel, recent work has demonstrated an approach for computing this bound for nodes using beamformed antenna arrays that know the spatial covariance of a wide-sense stationary channel. Since this bound does not directly indicate how to perform practical channel estimation to obtain estimates that can achieve the bound, the prior work demonstrates an approximation that indicates a practical channel estimation procedure while yielding a mutual information near that of the key rate in rich multipath fading environments. This paper demonstrates appreciable degradation in the mutual information achieved using this approximation when the channel includes a line-of-sight component and proposes a modified approximation technique that performs well over a broad range of propagation conditions.