Hybrid simulation of ethernet based intra-vehicle networks

Tuohy, Shane
As Advanced Driver Assist Systems (ADAS) increase in complexity, it becomes essential to be able to rapidly prototype and test these systems in a reproducible and reliable way. This work explores the problem of simulating intra-vehicle automotive communication networks over which ADAS and other systems communicate through the development and use of new tools and methodologies to assist in the analysis of the performance and impacts of these networks. A comprehensive review of intra-vehicle communication is presented and related literature is explored which details the state of the art, as well as identifying potential solutions to the outstanding requirements of next generation intra-vehicle networks. A novel network simulation testbed based on a Linux platform which has been specifically designed and developed for the purpose of automotive network simulation is central to this work. This testbed platform uses a hybrid real time approach, leveraging lightweight virtualisation technologies to allow for more flexibility and real time extraction of simulation outputs. Thorough testing on double star and daisy chain network topologies demonstrates that the platform’s results are in line with those in the literature and shows that, under heavy load generated primarily by six video camera streams, automotive timing constraints may be violated in an Ethernet based in-vehicle network. The real time performance of the testbed is also evaluated and it illustrates that real time streaming up to a bandwidth of 50.7 Mbps can be maintained with a time dilation of less than 7.5 ms. Measurement of the overhead due to the use of virtualisation technologies demonstrates that the performance of the testbed is within 5% of the host system for all measured values. The methodology and results of a novel image quality experiment facilitated by this platform demonstrates one of many potential real world applications of the developed testbed. An experimental subjective test of 26 subjects is presented which uses a database of real world automotive video containing artefacts resulting from packet loss caused by a simulated network, which was generated as part of this work. The results of the experiment, including Mean Opinion Scores, were recorded from all participants. Image saliency data, which was captured through a custom application developed leveraging an infrared eye tracking device with sub-millimeter precision, demonstrates that packet losses do not significantly effect the visual attention of a viewer. However, the results demonstrate that packet losses in regions of high temporal activity are more salient. A novel image quality metric for use in automotive packet loss effected video is derived which achieves a correlation value of 0.82 with MOS.
Publisher DOI
Attribution-NonCommercial-NoDerivs 3.0 Ireland