Antenna Selection With Erroneous Covariance Matrices Under Secrecy Constraints

We maximize achievable secrecy rate while performing antenna selection (AS) when we do not have perfect availability of instantaneous channel covariance matrices of the legitimate (L) and eavesdropper/ wiretapper (E) nodes. Instead, we have at our disposal corrupted estimates of the channel covariance matrices. The error component of the estimated matrices is assumed to be weighted by a norm-bounded error vector. For a class of norms, irrespective of the distribution of the error vector, we devise a so-called convex inner approximation (CIA) semidefinite program (SDP)-based solution that yields a transmit precoder with the desired sparsity as dictated by the number of antennas to be selected. Our numerical results reveal that the CIA procedure works close to exhaustive results and possesses amenable polynomial complexity properties. We conclude our numerical investigations by demonstrating the effectiveness of AS under the circumstances considered in this paper.