Joint UAV Trajectory and Transceiver Optimization for Over-the-Air Computation Systems.

This paper investigates an unmanned aerial vehicle (UAV) aided over-the-air computation (AirComp) system, where the UAV is deployed as a flying base station to swiftly compute functional values of the data distributed at multiple ground sensors via AirComp within multiple time slots. Subject to the individual transmit power constraints of each ground sensor, we aim to minimize the computational mean-squared error (MSE) of AirComp, by optimizing the UAV's trajectory over multiple slots, the ground sensors' transmit coefficients, and the UAV's de-noising factors per slot. Due to the complicated variable coupling, the resultant AirComp design problem is non-convex. As such, we decompose the original AirComp design problem into two low-dimensional subproblems, one for obtaining multiple groups of ground sensors to determine the UAV's trajectory over time, and the other for optimizing the ground sensors' transmit coefficients and the UAV's receive de-noising factors for AirComp. For the first subproblem, we use the K-means algorithm to group ground sensors, and then the UAV's hovering point at each time slot is determined based on each group of ground sensors. For the second subproblem, we recast it as a convex problem and then employ the Lagrange duality method to obtain the optimal solution. Numerical results show that the proposed scheme achieves a significant computational MSE performance gain over the alternative benchmark schemes.