We conduct research at the Technical University of Vienna (TU Wien) as part of the Institute of Telecommunications (ITC). Our work focuses on vehicular performance analysis. The United States averaged 90 deaths on the road per day in 2013. We expect applications based on vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communications to play a huge role in reducing this number greatly. By communicating dangers and construction sites ahead, a driver gains vital seconds to react to a perilous situation. Coupled with autonomous cars, we envision a large impact on overall road safety due to the ability of communications. Therefore, we must have a good grasp of the possibilities and limits of vehicular communication standards. This is a daunting task since the communication conditions differ strongly compared to prior applications.
The outdoor environment means that channel echoes arrive later, which results in a large delay spread. In addition, high vehicular speeds introduce challenges in terms of Doppler spectra not seen in relatively static mobile conditions. This makes it very important for us to thoroughly test the performance of a communication standard before employing it on the street. However, conducting such tests on real streets is impractical for many reasons. We can’t repeat measurements because too many factors change on the street. We can’t conduct measurements on a large scale since this requires blocking streets. Finally, for safety reasons, we cannot test certain scenarios in the real world.
Our basic solution to take the evaluation off the street is to emulate the vehicular channel. To this end, we had access to draft channel models defined by the European Telecommunications Standards Institute (ETSI) standard for intelligent transport systems at 5 GHz (ITS-G5), which are based on tapped delay lines. Our first approach used off-the-shelf modems and a channel emulator. However, it quickly became obvious that these solutions do not allow the necessary amount of configuration, control, and inspection we require. For example, the channel models defined a half-bathtub Doppler shape, which we could not implement on off-the-shelf channel emulators. Thus, we decided to replace all components with USRP-2953R devices. We could control the transmitter and receiver using the LabVIEW Communications 802.11 Application Framework. While this did not support the IEEE 802.11p extension of the IEEE 802.11 standard when we started working, we only had to make small modifications to make the framework 802.11p compliant. We programmed the channel using LabVIEW and allow a 10-tap tapped delay line in which we can update the position of the taps, as well as the small-scale fading coefficients, online. Our update rate for the parameters is given at 10 kHz. This relates to a movement of less than 5 cm for a vehicle going 161 km/h to ensure that we capture even small channel changes. The emulation bandwidth is given at 20 MHz, which is plenty for a communication standard working at a bandwidth of 10 MHz. Using this setup, we can now send arbitrary payload patterns with defined transmission times and packet lengths and we can log all the reception information we need all over channels that we can, within loose limits, set to arbitrary behaviors. Our first evaluation case was on point-to-point communication under six stationary channel models. We used compressed sensing methodologies to find optimal fits of channel sounder measurements. We took advantage of the fact that we can use NI host FPGA frameworks to stream data online and reproduced time-variant nonstationary channels in-lab, so we could evaluate for communication performance in a safe environment.
Finally, we introduced city maps and mobility simulators in the setup. We use vehicular traces generated for Linz, Austria, to analyze the worst-case scenario—a massive traffic jam. We use the traces and simulations of the network layer to generate communication graphs. We then simplify these graphs, so we can represent the street network with a small number of nodes. We use three USRP RIO devices and one off-the-shelf modem. Thus, we can measure the performance of ad hoc networks in urban areas in the lab without the requirement for a huge number of devices under test.
Benefits of Using NI Solutions
We are working with two different use cases that show completely different benefits. We developed our emulator that is based on LabVIEW from scratch. We leverage the full power of the FPGA by pushing it to its limits in terms of first-in-first-out capacity and FPGA usage. While we had to compromise in terms of computational complexity, we could choose which compromises to make. We are feeding arbitrary small-scale fading parameters at a rate of 10,000 per second for every tap. This means we can capture stationary models as well as nonstationary channel scenarios. We can completely shape the Doppler spectrum through this solution, which gives us even more flexibility. On the other hand, for the transceiver nodes, we benefit from the fact that the 802.11 application framework encompasses a fully developed 802.11 stack but still retains the flexibility of LabVIEW Communications. Hence, we could, with minimal invasiveness, extend the application framework to work at 5.9 GHz and with 10 MHz of bandwidth, which complies with the 802.11p standard. The framework gives us many out-of-the-box options for rapid prototyping of measurements, but it also provides more advanced features, such as UDP port forwarding, which we require for extensive measurements.
Using all these benefits, we can evaluate the performance of 802.11p for vehicular communications in a lab. The flexibility also means we can use channel emulators at highly nonstationary settings, change channel emulation mid-flight, and schedule complex packet transmission patterns that translate to whole clusters of cars. All these are necessary features that are vital to our scientific research, and they are all provided by NI solutions. Furthermore, given the setup, it would be simple to switch the emulation frequency and the application frameworks to LTE and conduct a similar analysis on a completely different protocol, which would be impossible using off-the-shelf hardware. We deliver thorough and reproducible throughput measurement results that are targeted at vehicular communications and conducted on real hardware, which has not been published in this form yet.
We needed to adhere to a standard-defined protocol stack, while requiring highly flexible test devices, by setting arbitrary packet transmission schedules, transmission powers, and modulation and coding schemes. We used NI hardware and software to address this challenge by taking all standard-compliant implementation off us. We easily hooked into the pre-existing application framework and focused on the emulator and the measurements. We also profit from the system’s resilience, often running the setup for weeks without rebooting, and the modularity allows us to extend our future targets. Among these, we are already evaluating large-scale ad hoc networks on multiple devices and multiple channel emulators. Furthermore, we plan to extend to other standards, considering LTE, and use the emulator in a 60 GHz frequency band. Our current research also addresses extending the achievable emulation bandwidth using advanced compressed sensing techniques.
Thomas Blazek - Institute of Telecommunications, TU Wien
Golsa Ghiaasi - NTNU Trondheim
Christopher F. Mecklenbräuker - TU Wien