What Are The Challenges For Data Transfer In ADAS/Autonomous Vehicles?

In the automotive market, growing demand for advanced driver-assistance systems (ADAS), and eventually autonomous vehicles, means vehicles are being equipped with ever more data-generating sensors and cameras.

As the industry moves towards self-driving cars, these systems are steadily increasing in sophistication, resulting in a major increase in data requirements, with potential terabytes of information needed to be moved around the car. As such, the connectivity requirements available in connected vehicles are increasingly demanding.

While a traditional car with just a few interfaces might only need a few 1Gbps interfaces, an ADAS or an autonomous vehicle (AV) will need a significant number of network ports and much higher data throughput.

There are many problems with traditional automotive approaches to this data rate problem. There are many competing standards, all of which are wiring heavy and struggle to provide the required bandwidth cost-effectively and reliably.

An ideal solution for in-vehicle connectivity, therefore, is Ethernet.

Solving Data Transfer Problems

First, Ethernet is an excellent technology for high-bandwidth data applications. Why is this so important? Let’s look at the systems inside a self-driving car.

A radar module will typically require a 100Mbps Ethernet connection; a lidar module will typically require a 100Mbps or 1Gbps Ethernet connection, while high-resolution uncompressed camera data (typically used for ADAS) might generate 5Gbps.

Ethernet is an ideal solution for in-vehicle connectivity.

It’s not just bandwidth that’s the issue; an often-unappreciated problem with increasing data rates is the resultant increase in the number of packets. These packets need their header decoded and the addresses/IDs looked up before they can be queued for transmitting. Per packet, this is not a lot to deal with, but with many packets, this soon becomes significant work – and software solutions to handle this can’t cope. This normally means turning to CPUs. However, as data rates increase, the number of CPUs required can quickly become untenable, increasing cost and complexity and potentially creating thermal issues.

Alternatively, Imagination’s Ethernet Packet Processor (EPP) can handle this work, efficiently and robustly. The EPP is available as a Layer 2 or Layer 3 switch, or in router configurations, and can support up to 200 Gbps of traffic and/or 300Mpps of processing, in just about any combination of ports.

Keeping Time

It’s not just the amount of data that is a concern in a vehicle – it’s when the data is sent. In a vehicle, timing is of the essence. First, there is the priority issue – consider an ABS sensor that sends out a reading every 100ms, while at the same time the infotainment system is streaming audio and video to the back seats; how does the networking system decide which is more important?

Waiting for data is known as latency, but even more of an issue is jitter, which is when the latency is inconsistent. Consistent latency is usually much better for system performance than highly variable latency, particularly in automotive. For the example of our ABS sensor, if some packets have no delay and others have 5ms, then that is less optimal than a system that resends everything with a 4-6ms delay. Therefore, it is important to have systems in place that can deliver time-sensitive networking, such as Imagination’s EPP.

The other challenge regarding data transfer is how and where data from all the various sensors (cameras, lidar, etc.) is processed in the car. Having multiple electronic control units (ECUs) adds complexity, weight, and more potential points of failure.

To simplify this there is a trend towards “zone controllers”. Each zone in a car can act as a hub for all these various sensors and actuators, reducing cabling, saving money and reducing weight and assembly time.

As we factor in the connectivity requirements for AI, we soon find that we need new, smarter IP solutions for the automotive market.