Predictable Messaging in Wireless Automotive CPS

Today’s vehicles are much more than a mechanical device, and complex systems of sensing, computing, communication, and control are ubiquitously deployed to se rve as the intelligent nerve systems of vehicles. For instance, the number of electronic control units (ECUs) is well over 70 in today’s high-end vehicles, and these ECUs process up to 2,500 signals (i.e., elementary i fo mation such as vehicle speed) and support up to 500 features such as brake-by-wire and active safety [4 ]. The increasing number of ECUs and control systems deployed in vehicles pose significant challeng es to the scalability of vehicular communication system, which is a basic element of automotive CPSes, and it h as become a common practice to deploy multiple communication networks (such as CAN networks) wit hin a single vehicle. These many vehicular networks are starting to add significant weight to vehicles a nd reduce gas efficiency. For instance, it has been shown that wiring harness is the heaviest, most complex , bulky, and expensive electrical component in a vehicle and it can contribute up to 50 kg to the vehicle mas s [2]. Therefore, wireless networks such as wireless, embedded sensor networks have been envisioned to b a basic element of future automotive CPSes [5, 6]. Besides reduced weight and thus improved gas efficien cy, wireless networks also enable communication and coordination among vehicles on the road for purpo ses such as active safety. It is thus expected that wireless networks will be ubiquitously deployed and se rve as a basic element of both intra-vehicle and inter-vehicle CPSes. In supporting mission-critical task s, automotive CPSes pose stringent requirements on the predictability and reliability of wireless messagin g. Nonetheless, wireless messaging is subject to the impacts of inherent uncertainties and dynamics within t e system itself and from the environment in automotive CPSes. Within a system, wireless communication assumes complex sp atial and temporal dynamics due to unpredictable channel fading, network topology constantly c hanges in inter-vehicle networks due to vehicle mobility, network traffic pattern can be dynamic due to event -triggered data traffic and varying applications (e.g., adaptive control logic), and application requireme nts on messaging quality (e.g., throughput, latency, and/or reliability) may also vary over time and across diffe rent applications. Moreover, different dynamics may well interact with one another to yield complex behavior s. For instance, dynamics in network traffic pattern introduce dynamics in co-channel interference and thus dynamics in wireless link properties (e.g., reliability), which in turn affect link estimation and rout ing in wireless networks [7, 9]. For instance, Figure 1 shows the network conditions in the presence of differe nt t affic conditions, where network condition is represented by the unicast ETX (i.e., expected number of t ransmissions required to successfully deliver a unicast packet) for links associated with a randomly selec t d node in the Kansei testbed [3]. We see that unicast ETX changes significantly (e.g., up to 32.44) as traffic pattern and thus co-channel interference varies. From the environment, a wide variety of factors can affect th e behaviors of wireless messaging. Environmental factors such as temperature and humidity can affe ct wireless communication, electromechanical