For wireless networks which aim at high speed communication and computation, we propose a parallel analog function computation scheme via wireless multiple-access multiple-input-multiple-output channels. We consider that each sensor node only has imperfect channel state information from itself to the fusion center.

Modeling the channel uncertainties by the worst-case model, the robust transceiver design problem for parallel analog function computation is formulated as a non-convex optimization problem which minimizes the worst-case mean-square-error subject to individual transmit power constraints. We then propose an alternating optimization algorithm to solve the problem. Simulationresults demonstrate that the proposed robust scheme outperforms the non-robust one.