Performance Characterization Using AoI in a Single-loop Networked Control System
The main motivation behind extending the notion of Age-of-Information (AoI) to Networked Control Systems (NCS) stems from recent results suggesting that AoI can be used to reformulate some of the traditional NCS problems with a new perspective. In fact, lower AoI in a NCS results in lower estimation/control cost. This is intuitive as to improve estimation/control performance it is often desirable to have access to fresher state information, which is conceptualized as AoI metric. However, AoI is rather a new concept and hence less understood compared to well-established notions, e.g. delay. Therefore, it has very recently attracted some attention to characterize the relationship between the AoI and performance metrics in networked systems. Despite some progress, there are many challenges still to address in order to develop the fundamental relationship between AoI and the conventional estimation/control performance metrics. Towards this end, we study an infinite horizon joint optimization problem of sampling and scheduling in a single-loop NCS with the objective of minimizing the mean square estimation error. Under some mild assumptions on information structures we show that the optimal controller can be computed separably while the remaining cost function is characterized as a non-decreasing function of AoI. Further, we identify a narrow class of LTI control systems for which minimizing the mean square estimation error coincides exactly with minimizing the expected AoI. Finally, noting that minimizing the non-decreasing function of AoI in our problem setting is a stochastic combinatorial optimization problem and is hard to solve, we study heuristic algorithms by extending existing algorithms in the AoI literature.
READ FULL TEXT