Training optimization for hybrid MIMO communication systems

Channel estimation is conceived for hybrid multiple-input multiple-output (MIMO) communication systems. Bothmean square error minimization and mutual information maxi-mization are used as our performance metrics and a pair of low-complexity channel estimation schemes are proposed. In eachscheme, the training sequence and the analog matrices of thetransmitter and receiver are jointly optimized. We commence bydesigning the optimal training sequences and analog matrices forthe first scheme. Upon relying on the resultant optimal structures,the training optimization problems are substantially simplifiedand the nonconvexity resulting from the analog matrices canbe overcome. In the second scheme, the channel estimation anddata transmission share the same analog matrices, which bene-ficially reduces the overhead of optimizing the associated analogmatrices. Therefore, a composite channel matrix is estimatedinstead of the true channel matrix. By exploiting the statisticaloptimization framework advocated, the analog matrices can bedesigned independently of the training sequence. Based on theresultant analog matrices, the training sequence can then beefficiently designed according to diverse channel statistics andperformance metrics. Finally, we conclude by quantifying theperformance benefits of the proposed estimation schemes.