I found that the paper, "On the convergence conditions of distributed dyanmic state estimation using sensor networks: A unified framework", proposed the almost identical framework of ours. It should not be surprising, since Covariance Intersection can be naturally integrated in Kalman filter. However, they gave a convergence criterion based on the paper "Kullback-Leibler average, consensus on probability densities, and distributed state estimation with guaranteed stability," where some assumptions are unnecessary in discrete-time dynamic system. I think it is better to interpret the whole things in the classic control theory.