Averaging Techniques for Balancing Learning and Tracking Abilities Over Fading Channels
Dong Shen, Ganggui Qu, Xinghuo Yu
Abstract
With the wide use of networks in repetitive systems, channels between a plant and a controller may experience random fading, which is a common problem in long-distance wireless communication. However, the control problem over fading channels is far from resolved. In this article, we investigate learning control over fading channels to gradually improve tracking performance. We observe that the effect of fading on input transmission greatly compromises tracking ability in practical implementations. We examine three average techniques: moving average, general average with all historical information, and forgetting-based average. The results reveal a tradeoff between learning ability and tracking ability for learning control algorithms, where learning ability refers to the convergence rate of a proposed learning algorithm, and tracking ability refers to the final tracking precision of the output to the desired reference. The convergence results for the three schemes with these averaging techniques are strictly proved. The results demonstrate that the forgetting-based average operator-based scheme can connect the other two schemes by tuning the forgetting factor. We also provide extensions of several general scenarios to expand the application range. Illustrative simulations verify the theoretical results.