Optimal dropout policies for networked control systems

SUMMARY By realizing the feedback paths over communication networks, we get a class of networked control systems (NCSs), where the network's quality‐of‐service (QoS) is commonly characterized by the average dropout rate of feedback data packets. The control performance of an NCS however, is determined not only by the average dropout rate but also by the dropout pattern of feedback data packets. This paper provides a systematic way to determine the optimal dropout pattern (policy) under a given average dropout rate, where the performance is measured by the output signal power under an exogenous white noise. By modeling the finite‐memory dropout policies with the general Markov chain, this paper formulates the optimal dropout policy design into the optimization of parameters of a dropout Markov chain. That optimization is first solved by an augmented Lagrangian gradient method, which may be stuck at local optima because of the problem's non‐convexity. To compensate this weakness, we apply the branch‐and‐bound method to the optimization whose constraints are bilinear. The branch‐and‐bound method can approach the global optimal solution with any desired tolerance in finite steps. The obtained optimal dropout policy may be interpreted as a network's QoS constraint whose enforcement provides a hard guarantee on the control system's performance. An example is used to illustrate the effectiveness of the achieved results.