Manufacturers might have self-driving cars ready to roll in the next five to seven years, but how far they’ll roll will, in large part, be determined by 5G mobile network deployments. To support fully autonomous driving, where no human driver is needed and passengers can just kick back and ignore the road, fast broadband connections will be necessary.
Nobody knows yet how fast, but minimum service levels will depend on three speed metrics: download throughput, upload throughput and latency. All three will have to be better than what’s available via today’s 4G networks.
The chart above was published by GSMA, which is a trade group that represents mobile carriers around the world. It reckons that up and down throughput will have to be in the 10 Mbps range, with latency – the round trip time – in the 1 millisecond range, in order to support autonomous driving.
Continental is one of the automotive technology companies that has to actually invent and manufacture the equipment, and design the supporting platforms for self driving cars. It takes the GSMA estimate as a starting point, and stretches those specs: latency might not have to be so good – they’re considering a range of 10 milliseconds to 100 milliseconds – but speeds might have to be faster, maybe as fast as 100 Mbps.
Despite the hype from carriers and the Federal Communications Commission, there’s little indication that ubiquitous 5G networks in the U.S. will be there when the automotive industry’s technology is ready to go to market. You might be able to buy a self driving car by the middle of the next decade, but opportunities to take your eyes off the road and your hands off the wheel (or whatever controls it might have) will be limited.
There will be a flood of bits swirling through self driving cars, and virtually all of that data will be processed by onboard computers, even where 5G networks are deployed.
“Autonomous vehicles are software defined”, said Deepu Talla, vice president of autonomous machines at Nvidia, a high end chip maker, speaking at CES. That software will run on onboard computers, and won’t be processed served from the cloud via mobile broadband networks, he said. There are four reasons for that:
Latency. If you’re in a moving car, the round trip for data takes too long.
Bandwidth. Cars will continually generate huge amounts of data, particularly from the many high definition video cameras they’ll use to monitor where they’re going and what’s around them.
Connectivity. It’s not always there, particularly in rural areas, but even in cities there are momentary holes and bottlenecks in network coverage. Not big enough, perhaps, for a human to perceive but enough to delay machine to machine communication for critical milliseconds.
Privacy. Although it’s not as big of a concern for cars as for, say, medical devices, it’s still a limiting factor.
5G won’t solve the problem, Talla said. Latency may decrease but it will still be there and 5G’s greater bandwidth will be eaten up by greater demand. “the amount of data will increase too”, he said.
Continental, a German automotive technology company, plans to scale up in-car local area networks to 10 Gbps to handle that load. Most of it will be video streams from high resolution cameras – 8 megapixels – that have to processed and analysed in real time. Each car will have at least four cameras, and possibly more. Plus radar and lidar, and video streams transmitted directly from cars up ahead.
Mobile broadband will still play a role. Live connections to the cloud are yet another source of data, particularly for error detection, debugging and instant repair. Connectivity will be required for cars to reach Level 5, the top level of autonomous operation, according to Continental staff who briefed industry analysts during CES. At that level, the car does everything, everywhere, without the need for human monitors. That’s the point where you can take a nap in the back seat while driving to work.
When fully autonomous vehicles arrive depends on where you want to be driven. If there’s a low latency, high speed mobile network available – likely, a 5G network – then you’ll be able to take a nap in the back seat. If connectivity isn’t there or isn’t up to the necessary standard, then you’ll have to sit behind the wheel and be ready to take control on short notice.
That’s my takeaway from a briefing at CES yesterday by [Continental, an automotive technology company](). Among other things, it makes data processing and communications systems for major carmakers. One of the challenges to get to the point where machines can be given full responsibility for driving people safely and efficiently is delivering sufficient information to them quickly enough.
On board sensors are the primary source of data. But they’re also the last line of defence. Their range is limited to somewhere between 100 meters and 300 meters – call it something like a city block. Any additional information will have to be gathered via wireless communications. And there’s a lot of it: weather, traffic signals, position and intent of other vehicles and more. With all of that information available quickly enough, cars can drive themselves, Continental believes.
The big hurdle is network latency. Continental reps said that current 4G networks typically have 100 milliseconds of latency, which is too much of a lag for safety related information. Experiments on 4G systems in Germany have brought that down to 20 milliseconds, which is fast enough, but as a practical matter it’ll take a general upgrade to 5G standards – whatever those turn out to be – to get that kind of performance on a generally available, commercial basis.
The timeline for 5G deployment and the development of cars that capable of fully autonomous operation under a given set of conditions is about the same: [the middle of the next decade](). As the technology develops, those conditions will encompass more and more places. But the ultimate limit will be mobile network availability and performance.