The communication graph design problem for a linear discrete time-varying sensor network system is considered. The design goal is to minimize the upper bound for the anisotropic norm of the input-to-esti- mation error system. The sensors are supposed to be objects with dropouts of given probability values. Exog- enous disturbances are considered to be sequences of random vectors with bounded anisotropy of sequence fragments. The tuning of the adjacency matrix with a fixed estimation model is reduced to a convex optimi- zation problem.