Scientists:	Sebastian Gallenmüller, Prof. Dr.-Ing. Stephan M. Günther, Maurice Leclaire, M.Sc., Prof. Dr.-Ing. Georg Carle
Duration:	01.10.2016 – 30.09.2019
Funding:	DFG SPP 1914
Homepage:	http://www.spp1914.de

Description

This project aims for a measurement-based approach towards composable models for CPNs. We set up a measurement testbed that supports the evaluation of a CPN benchmark application using reproducible measurements of different wireless technologies and protocol stacks. Our approach is to use a two-wheeled inverted pendulum (TWIP), of which a Segway is a well-known instantiation, as benchmark example. This software emulation can be evaluated in combination with a variety of wireless protocol stack alternatives in an automated measurement series.

We analyze the protocol stack in detail, using composable models that we parameterize using the measurement results. Network calculus provides a framework that allows for combining performance models. Stochastic Network Calculus allows to give stochastic performance bounds on delays. Temporal network calculus allows to model the impact of error-prone wireless links and contention-based multi-access networks on latency. We investigate the influence of the software stack execution on the network service quality by different processing platforms. For effective collaboration with other projects of the Priority Program "Cyber-Physical Networking", we will share measurement data generated in the testbed.

Results

The goal of MOONSHINE's first work area was the investigation and measurement of networked control systems. The main target of this investigation was a wireless two-way inverted pendulum robot [Zoppi, 2020] [Music, 2019] [Gallenmüller, 2018]. This robot was designed to be affordable, easily replicable, and modifiable. In addition to the measurements of this robot, a wireless testbed facility was created to allow control over a wireless environment [Gallenmüller, 2019]. Therefore, the existing testbed was equipped with two rf-shielded boxes. This testing facility was used to apply a real wireless link to a software emulation. The wireless testing facility utilizes programmable attenuators to impact the channel quality selectively. Based on these results, a wireless channel emulation was created [Bergbauer, 2017].

The second work area derived models from the measurements performed in the previous work area. Measurements and the created testbed laid the foundation to derive the values needed for service curve parameter estimation. Bergbauer [Bergbauer, 2017] used these results to create models, for instance, to predict a success probability for the wireless channel. Various impacting factors for the wireless channels and their subsequent impact on the different layers of the OSI model were investigated [Gallenmüller, 2019]. A particular focus was put on high-level performance indicators for networked control systems. These results were used to predict the quality of control. The predictions led to the development and investigation of strategies to improve the robustness of networked control systems. One of these models was implemented in the NCSbench platform, allowing the compensation of sub-second connection losses for the balancing robot [Zoppi, 2020] [Music, 2019].

The third work area involved the validation and interaction with other projects of the SPP. During the runtime of the SPP, several successful collaborations with other subprojects were established. Together with the subprojects ("Optimal Co-Design of Wireless Resource Management and Multi-Loop Networked Control (Hirche, Kellerer)" and "Cooperative Consensus-based Control of Multi-agent Systems over Wireless Channels" (Raisch, Stanczak)) a benchmark for CPN was developed [Zoppi, 2020] [Music, 2019] [Gallenmüller, 2018]. The framework, including the two-way inverted pendulum robot, was replicated across multiple sites to cross-validate the results. Despite different environments, a replication of results was possible. Besides the successful collaboration on the benchmarking framework, a collaboration with other projects could be established. The results of this collaboration led to a joint paper [Gallenmüller, 2019]. There, a concept for gain scheduled control was investigated, where the control algorithm can be adapted to the current network conditions. The PRRT protocol was used to collect and evaluate the network conditions during runtime. The shielded testbed environment was used as a reference platform to perform measurements on the investigated system.