Among sensors for ADAS and autonomous driving, radar has proven to be very reliable for ADAS applications such as ACC. The use of radar is expected to expand in the future, as radar is often used in conjunction with other sensing technologies such as cameras and LiDAR. The table below summarizes the pros and cons of the different types of sensors.

 

Fully autonomous driving requires the fusion of data obtained from different sensing technologies. With sensors distributed around the car, it will be possible to provide 360-degree coverage, forming a safety "cocoon" around the car.

 

Radar can be easily mounted behind common vehicle components, such as bumpers or vehicle logos, without compromising aesthetics. The 76-81GHz frequency band has been accepted by most countries as the frequency band for automotive radar. As radar antennas become smaller in size, it becomes easier to physically integrate high frequencies. However, as frequencies rise, new challenges arise due to power trade-offs, higher losses, and the impact of higher manufacturing tolerances.

 

Radar is also particularly suitable for applications in vehicles, since vehicles are good reflectors of radar waves. It can be used both for comfort functions, such as ACC, and for high-resolution sensing applications, which increase the vehicle's passive and active safety features. Examples include blind spot detection, lane change assist, rear cross traffic alert, and detection of pedestrians and cyclists near the vehicle.

 

Currently, radars are basically classified according to their detection range:

 

• Short-range radar: detection distance up to 50 meters, wide field of view, high resolution
• Medium-range radar: detection distance up to 100 meters, medium field of view
• Long-range radar: 250 meters or more distance, narrower field of view, lower resolution

 

With the development of new technologies, the detection range is expected to exceed these limitations, while increasing the vertical detection dimension to provide a complete 3D picture of the surrounding environment.

 

In the future, vehicles will be expected to add multiple radar modules, upgrading from a basic forward-facing radar configuration (which provides basic L1 functionality) to higher levels. Soon, more vehicles will be equipped with corner radar for L2+ functionality and NCAP levels 4-5, and higher levels L3-4 and NCAP levels 5.

 

However, how the radar data is processed will largely depend on the vehicle architecture. The current trend of improving the performance of the central computing unit has also promoted the transformation of the vehicle E/E architecture from a distributed architecture.

 

Although the migration to a fully distributed architecture will not be complete until 2030, partial implementations will appear on the market earlier.

 

First, some domain controllers will be used for specific functions.

 

In addition, the number of domain controllers will increase, and zonal controllers will also be introduced before a fully centralized E/E architecture is established, where the vehicle's central computer will be connected to the sensors through the zonal ECU. This development also requires increasing the capacity and reliability of the vehicle network, as well as the sophistication of the software. This can present significant challenges, including additional connections that may require more expensive wiring harnesses to meet the ever-increasing data rate demands.

 

With the introduction of the new E/E architecture, part of the radar processing can be offloaded from the radar module (edge computing) to the zonal or central ECU, enabling more efficient computing. Today, complete radar processing is performed at the edge, using "smart sensors".

 

This means that many independent radar modules are distributed around the vehicle, each with its own radar transceiving and processing capabilities. The processed data, usually a list of objects, is transferred to the ADAS ECU for further processing and possibly fusion with data from other sensors. By properly distributing sensors in the vehicle, it is possible to correctly perceive the environment of the vehicle and identify obstacles.

 

With the development of centralized computing architecture, data processing of some radar modules may be transferred from the radar to the remote processing unit to be directly handled by the zonal ECU or the central computer. The radar module itself is not that "intelligent" and only does limited processing of the received radar signal.

 

For example, the module will determine the distance to different objects and provide a distance profile to the remote processor. It will then receive preprocessed data from the different satellite radar modules and perform the remaining processing steps on each set of data, generating a list of targets with their respective characteristics (range, direction, and velocity) and creating a complete picture of the surrounding environment. The resulting results will be fused together or combined with results from other sensors.

 

In the initial implementation of this centralized architecture, preprocessed data from different radars can be transmitted to the zonal or central ECU via an Ethernet backbone. When higher resolution is required and the amount of data is prohibitive, such as forward-looking or imaging radar, radar processing may still occur on the sensor itself to reduce the amount of data to be transmitted.

 

Centralized processing of long-range radar data offers many advantages. First, the radar module itself becomes less complex, saving size and cost, and reducing thermal issues. Both hardware and software repairs and upgrades will be made easier.

 

Second, utilizing the car's existing network, the Ethernet backbone, also reduces the cost and weight of wiring. Additionally, data transmitted via Ethernet will be available in a format that is easier to store and process.

 

Finally, data processing in vehicle control units opens the door to greater efficiency and more complex operations. Sensing capabilities can be enhanced by enabling data fusion with other sensing technologies (cameras and LiDAR). Machine learning and AI can be considered for advanced detection and prediction, enabling higher levels of autonomous driving.

 

Edge processing and centralized computing are expected to co-exist for several years. Migrating to a centralized computing architecture will require access to high-speed links throughout the vehicle, which may result in the use of different standards for data exchange. Aside from cost and layout complexity, the jury is still out on which standard to use for data transfer. While CAN and Ethernet currently dominate, some manufacturers are pushing alternatives such as MIPI A-PHY.

 

In any case, additional security measures must be taken to guarantee the integrity and security of the transmitted data. For example, transferring data over an Ethernet link requires additional processing and memory, as media access control security (MACsec) and hardware security modules may be required.

 

Also, as the number of radar modules continues to increase, all of them are emitting and receiving radio waves, which can also cause interference problems. Interference reduces the detection performance of radar systems, thereby reducing the functionality and safety of ADAS and autonomous driving systems. Several mitigation strategies are currently being analyzed by the industry, which can be grouped into three groups: obstacle avoidance, detection and repair, and cooperation and communication-based mitigation.

 

Radar has become a key sensor for ADAS and autonomous driving applications. The imaging radar system consists of modules that require transceivers to cover the entire automotive radar frequency band; support short, medium, and long-range radar applications; and meet the needs of centralized processing. ECUs are also an important part of ADAS and autonomous driving solutions that require advanced SoCs that enable centralized processing and enable high-speed image recognition and processing of surrounding objects from cameras, radar and LiDAR.

Previous: High-performance Computing (HPC), the Foundation of Smart Cars

Next: Designing Computer Vision Systems for Self-Driving Cars