Prioritizing Vehicle Data Traffic

It’s no surprise to hear that data complexity is on the rise inside vehicles today, but the scale just might cause you to gag on your coffee. It’s expected that a fully connected car will upload as much as 30 gigabytes of data to the cloud every hour.

Given the scale of data, it is imperative to make sure every bit of it is classified and tagged properly so that the subsystems, and infrastructure may process it correctly — making sure each piece of information goes only where it is supposed to go, for the right reason. The challenges are numerous.

“You’ve got a platform that’s spatially large so distance actually matters,” said Rob Knoth, product management director for the DSG group at Cadence. “Data prioritization is being handled in a couple of different ways today as far as the system design goes. On a phone or a drone or other platforms like that, they are so small that distance can actually be factored out of the equation. Not so with a car. You’re integrating a huge heterogeneous suite of technologies—powertrain (hard core hundreds of volts, big macho electronics), ADAS, cameras, parking sensors, LiDAR, and many other new technologies that are helping navigate the car, along with the real safety-critical applications. Then there is a huge amount of consumer electronics and infotainment data being integrated into it. It’s like a parfait of different technologies. And those are just the brains.”

If challenges are good for the engineering community, then this area is loaded with problems to solve. “Within the networks, this is where it gets extra exciting because we all focus so much of our time on what’s going on inside the chips,” Knoth said. “But now you have to expand it outward. It’s not just the chips. It’s not just the software running on the chips. It’s the network that’s interconnecting all of those chips, the software that’s running, the wires, the cables, the brains — and that is a fascinating problem to solve.”

There is debate about exactly how to tackle the data prioritization problem. In one camp are engineers who believe separate networks are the answer. “If you talk about prioritization, one of the easiest ways to prioritize is to build a road that is dedicated to just one vehicle, or build a network that’s dedicated for just one type of traffic,” he said. “That’s the easiest way to make sure I don’t have to deal with priorities. I’ve got my HOV lane. I can do whatever I want. This works to a certain extent, but it isn’t a get-out-of-jail-free card.”

One example of where individual networks start to fall apart is easily seen in an airplane. “Airplanes had to start dealing with integrating infotainment and electronics in the backs of seats,” Knoth noted. “The amount of weight that that adds to a commercial airliner is tons, so what that does to fuel costs and ranges of flying is astronomical. In that kind of industry where they are scraping by on thin margins, they can’t handle that. While that is taking the problem to the extreme, automobiles also have to consider some of this because they can’t afford an individual network for every single kind of traffic. It’s one option. It works for safety-critical or noise sensitive traffic, so you can do that to a certain extent.”

In the same way the airline industry solved this problem with wireless technology, Knoth expects the automotive industry to do the same. “For an airplane, going wireless — while it is injecting a safety concern because now suddenly you’ve got this broadcast going on that’s going to interfere with navigation that they have to be afraid of — the amount of weight that it saves by yanking all of the cabling out, yanking the screens out, was such a huge carrot that driving more and more down from the wireless aspect you can see that they’ve clearly chosen that area. To dial the contrast between automobiles and airplanes up to 11, you will start to see more and more that wireless will be an option in cars. And while the problem is slightly different — there isn’t as big an area to cover, not as many people — there will be traffic that doesn’t need to be prioritized. It’s not safety critical.”

Intel views data traffic prioritization today in terms of a new platform, namely a software-defined cockpit that has certain capabilities with speeds and feeds, according to Ken Caviasca, vice president of Intel’s IoT group and general manager of platform engineering and development. “There is also a parallel compute system that’s handling the really highly automated or fully automated driving capabilities, so you’re trying to divide up the ingestion of the sensor fusion and information largely coming into those two platforms.”

He explained that the outbound communication from either of those systems is shared. For instance, in an advanced driver-assist feature such as lane departure, that could be workload-partitioned into the software-defined cockpit. The service is handled there, and queued there, but it’s using the same electro-mechanical mechanism as the ADAS system to do autonomous driving in a more aggressive way. As such, the outbound will be a lot more shared in the basic infrastructure of the vehicle, which then has a strong requirement around security on that bus and the agents that are on it, because it’s sharing a common workload through multiple compute actuators that are driving it.

Where to begin
The starting point, according to Sherif Ali, application engineer for automotive channels at Mentor Graphics, is determining the type of vehicle because communication between connected cars won’t only be between the vehicles. It also will involve the infrastructure (V2I or V2x).

The first point to take into consideration is whether the data is either safety critical or non-safety critical data, which includes navigation updates and infotainment. Then, as part of the prioritization determination, it must also be decided if the data is vehicle-to-vehicle communication or vehicle-to-roadside assistance. For example, if the data is coming from the roadside unit, like a road information update to vehicles, that is typically considered safety-critical data. However, if it is a navigation update or pings from roadside systems, this is non-safety-critical data, he said.

Once this is clearly defined, the prioritization can be done based on the class of the data. Another consideration is the type of vehicle. There may be different tags for the data to differentiate between emergency vehicles like police cars and ambulances to give them higher priority in terms of their data over other vehicles.

The details of how and when are still being worked out among automotive OEMs, along with the automotive ecosystem, but Ali pointed out that one of the ideas is that the priority scheme will be based on the car position. “The main idea behind this proposal is that, usually, traffic accidents are concentrated in certain areas like intersections and highway entrance ramps. As a rule of thumb, we know this is usually where traffic accidents will be concentrated, so somehow it is justified in this case to put some kind of roadside unit — either permanently or temporarily — that will do the data prioritization based on the position of the car. As such, it is reasonable to expect there will be two to four zones around this unit, and based on the position of the car within a certain zone, the data will take higher priority than other zones.”

So while the automotive ecosystem has not yet agreed on the tagging method for data, given that it will wait until the prioritization schema is determined, it has come to agreement on the communication protocol of IEEE 802.11p, an amendment to the WiFi protocol, and more specifically is the Wireless Access in Vehicular Environments (WAVE) version for V2V and V2X.

As is typical in the world of standards, it’s still not smooth sailing here. Some believe 802.11p is not responding well to the real-time data, particularly the safety-critical data. “They claim the technology it is based on Carrier Sense Multiple Access (CSMA) with collision avoidance, which doesn’t have an upper bound for the delay on the network,” Ali said. “If this upper bound is not in place, we can’t rely on this for real-time data or safety-critical data. The solution to that is to add another layer to 802.11 to serve the safety-critical data, specifically in this case.”

The data glut challenge
Other challenges to prioritizing vehicle data include how to manage connected car information, predicted to be in the realm of 20 to 30GB to the cloud, every hour. “It is a huge amount of data,” he said. “This is per hour, and this data will be about everything. It could be some root data about the speed of the car. Sometimes it will be the status of a component in the vehicle that will be reported to the infrastructure or to the OEM or dealer. Maybe on the other side it will be entertainment data or map updates. One could imagine a traffic jam with multiple cars stopped, bumper to bumper, waiting for the road to be cleared out, and all these cars are doing some updates or uploads to the network, or downloads from the network. You can imagine the problem that will happen. There will be spikes in cellular data demand that will require special bandwidth.”

This will require such scenarios to be taken into consideration when designing the network for both V2V and V2I. Just dealing with the volume of data will be a challenge. And then there is the issue of data classification and prioritization, along with the race toward autonomy and connectivity. It’s a lot to troubleshoot.

Related Stories
Sensors Enable ADAS
Advanced driver assistance requires a careful and complex balance of hardware, software and security.
Tech Talk: ADAS
Why the Advanced Driver Assistance Systems standard is so important and where the potential pitfalls are.
Preparing For The IoT Data Tsunami
As the number of interconnected devices increases, technology challenges grow.
Grappling With Manufacturing Data
Questions persist about how to deal with an explosion in data, and who has access to it, but changes are on the horizon.
Big Data Meets Chip Design
As the volume of data grows, companies are looking at what else they can do with that data.