Design by uncertainty: Towards the use of measurement uncertainty in real-time systems

Real-time systems usually incorporate a wide variety of challenges: A control engineer, for example, aims for the highest possible control quality achievable. Here, one key element is to minimise the uncertainty of the measurements. This, to put it simple, is the noise of sensor data, which has a negative effect on control. Although measurement uncertainty is well treated in control theory, it is usually ignored in common real-time architectures where temporal properties are the prevalent criteria. Consequently, the communication between control engineers and real-time specialists is rather one way, revolving around deadlines and sampling periods. However, on closer examination, many real-time properties are derived from the measurement uncertainty aspired by the control engineer. Conversely, temporal variations, virtually inevitable in practice, can be represented as measurement uncertainty as well. Tackling the measurement uncertainty should therefore be an interdisciplinary task: The control system respects the actual run-time conditions instead of estimating them. Likewise, the real-time system considers measurement uncertainty rather than blindly sticking to deadlines. In this paper we present an uncertainty-centric approach to leverage measurement uncertainty in real-time architectures, not only at design time but also at run-time. Using measurement uncertainty as an explicit interface minimises the gap between real-time specialists and control engineers and facilitates a modular and flexible system design. Our preliminary results are promising and show the ease of use and the applicability to existing systems.