For automated driving at the L2 and L2+ levels, the system primarily consists of high-resolution optical cameras, as well as radar with a limited range and field of view. Short-range ultrasonic radar or radar sensors are also used for specific comfort and safety functions, such as automatic parking, intelligent cruise control and lane keeping. In order to reach the L3 level, designers have had to add lidar sensors, which has had many negative effects and limitations, not least the high cost. millimeter-wave radar
But a new generation of 4D imaging radar sensors is changing the situation dramatically. These affordable, high-resolution millimeter-wave radar sensors provide excellent resolution in 4 dimensions: distance, azimuth, pitch (for the first time), and accurate, directly measured velocity information. These new sensors also provide longer range and wider field of view and support an extended operational design domain. All data is transmitted in real time to the self-driving car’s fusion processor.millimeter-wave radar
The combination of these features greatly improves reliability and expands the operational domain. Unlike cameras or Lidar, millimeter-wave radar is naturally characterized by strong penetration in bad weather or poor lighting conditions. The combination of high spatial resolution and precise velocity measurements reduces ambiguity in object detection. As a result, more robust target detection and classification capabilities are realized over a wider range of environmental conditions.millimeter-wave radar
Radar Evolution
These new sensors are very different from most previous generations of automotive radars. Their strengths stem from years of building on proven technology in the military and aerospace sectors. Today, as semiconductor and antenna manufacturing integration and performance continue to improve, systems previously used in F-18 fighter jets are now able to be sunk into compact modules mounted in multiple locations in passenger cars, and these capabilities are trending downward in both size and cost.
The first major change is the introduction of low-cost millimeter-wave radar transmitter and receiver hardware that provides excellent transmitter power and high receiver sensitivity. This change has triggered two other changes: the use of dense multi-antenna arrays (referred to as multiple-input multiple-output, or MIMO, in 5G communications); and the use of complex waveforms, which greatly enhance the ability to make pitch, azimuth, and velocity measurements. Finally, advances in digital signal processing (DSP) intellectual property have enabled the converged digital processing capabilities that are necessary to handle the multichannel, high-speed data that these sensors generate in real time.millimeter-wave radar
Technical Challenges
How these incremental changes translate into valuable functionality is a matter of coupling hardware capabilities with firmware functionality. For example, to achieve highly accurate azimuth and pitch resolutions, sensors must employ large aperture and virtual array synthesis techniques. Advanced signal processing algorithms can combine multiple signals across the time, frequency, or code domains, or combine various types of signals across some combination of these domains, to construct a virtual antenna array that is larger than the physical array.millimeter-wave radar
This greatly improves resolution: current designs aim for an azimuth angle of less than one degree and an elevation angle of about one degree. This resolution generates a greater number of uncorrelated measurement points for each target object than a low-resolution radar, which in turn determines the position and profile of the target object more accurately. In this case, the subsequent processing will be smoother, allowing for easier differentiation of distant target objects and easier introduction of machine-learning based classification algorithms, which are similar to those used for processing camera or LiDAR data.millimeter-wave radar
In these systems, the design of the transmission waveform has a huge impact on the performance of the sensor and the overall cost of the entire solution. In order to ensure orthogonality between the transmitted waveform and the virtual array construction process, careful tradeoffs and thought need to be given to the selection of the underlying waveform structure (the sequence of transmitted chirps). For example, transmitting signals in parallel from multiple transmitting antennas will generate measurements for numerous virtual channels at the receiving end. Therefore, it is necessary to consider the challenge of separating these measurements with reasonable algorithms and processing power.millimeter-wave radar
Tradeoffs also need to be made when choosing a channel multiplexing method. Imposing restrictions on measurements may lead to artifacts. For example, such restrictions may cause coupling of angular and Doppler measurements, which can affect the range of accurate velocity measurements supported or cause other measurement ambiguity issues.
In addition, there are many system-level effects to be aware of. Here are just a few: as a first example, higher sample rates are required to achieve high analog bandwidth and short chirp durations. This in turn makes the design of the ADC converter more difficult and increases the cost of the ADC. Second example: Phase-based channel multiplexing schemes require analog phase shifters with high phase resolution. However, such phase shifters are challenging to manufacture and require fine and sensitive offline and online calibration. A third example: transmitting signals from multiple transmitting antenna elements at the same time requires a more complex thermal design at the system level to dissipate the additional heat generated by the transmitters.
In summary, fully utilizing high-resolution millimeter-wave radar in autonomous vehicle systems requires numerous design considerations, each of which should not be overlooked. However, the benefits of these sensors far outweigh the increase in system capability.
Real-World Advantages
These capabilities are not just in the form of improved numbers on a performance specification sheet. They allow for more fine-grained differentiation in functionality, which improves the safety, autonomy, and operational design domains of real-world vehicles when traveling in real-world scenarios.
For example, an increase in the amount and accuracy of 3D position and velocity data can significantly improve the ability of self-driving vehicles to recognize objects. The excellent data provided by the sensors enables the vehicle to make subtle, yet critical, distinctions, for example, between strong reflective signals from a large truck and weaker signals from a nearby child. Real-time Doppler velocity measurements mean, among other things, that the vehicle is able to immediately detect sudden changes in an object’s speed without the need for several field-of-view scans, as well as differentiate between close objects moving at different speeds.
All these advantages contribute to a better understanding of the situation around the vehicle. This means greater safety. Add to this the ability of high-resolution millimeter-wave radar to work in cluttered scenes with poor visibility and complex lighting, and you get a vehicle autonomy system capable of operating in a wider range of conditions with the increased safety and reliability that the industry strives for.
Flexible, Scalable Solutions
All of these challenges require a flexible and comprehensive radar SOC solution that is both “software-defined” and scalable to support advanced radar processing algorithms. Such a platform would include a high-performance DSP engine, optimized hardware gas pedals for multi-dimensional FFT operations, and a dedicated software development kit.
Developers can utilize the different products in the SensPro family to create different versions or iterations of their products. With its common architecture, DSP software code can be easily and smoothly migrated between cores, saving investment in previously developed software codebases and reducing time-to-market.
Of course, the underlying processing power must keep pace with emerging requirements, requiring a programmable architecture. As market needs evolve, CEVA continues to optimize its SensPro architecture and instruction set to support the following capabilities:
Reliable and robust target detection using advanced CFAR schemes (e.g., OS-CFAR);
Support for enhanced resolution beyond the “Fourier limit”, using advanced algorithms for super-resolution;
Supports inter-frame level processing by processing radar point clouds to enable advanced tracking schemes (e.g., segmentation and classification of objects from “post-tracker” point clouds using Kalman filters and the application of specialized AI models trained to do so).
Beyond Driver Assistance Systems
Today’s level of automated vehicle driving can significantly improve vehicle safety and smoothness of movement, when used judiciously. However, the ultimate goal is still to achieve fully automated driving, at least for certain vehicle classes. High-resolution 4D millimeter-wave radar will play a key role in meeting this challenge. The designers envision a sensor suite that includes a 4D radar in each corner of the vehicle and at least one LiDAR, all of which feed data into a sophisticated sensor fusion processor and artificial intelligence module. With sufficient quantity and quality of sensor data, sophisticated fusion and AI processing capabilities, and adequate training, the hope is that by reducing errors and extending the operational design domain, the goal of fully autonomous driving will eventually be realized. These vehicles will be able to operate in virtually any environment and become a widely accepted part of the transportation system.
However, there are still many challenges ahead. As the number of vehicles deploying millimeter-wave radar increases, so does the potential for interference. This will trigger innovation in the field of waveform processing for individual radar sensors and is expected to drive the development of vehicle radar standards. Accordingly, standards may facilitate cooperation between vehicles and between vehicles and infrastructure, which could mean a whole new role for radar sensors as part of a large-scale distributed intelligent network. The current lack of harmonized standards in the field of automotive radar transmission calls for a solution with a high degree of programmability to adapt to different situations. This may include, for example, the implementation of possible interference mitigation measures at the individual sensor level or the addition of coordination mechanisms between the sensors and the infrastructure so that the sensors can safely and efficiently utilize shared spectrum resources. Ongoing developments in the areas of underlying semiconductor technology, antenna design, and algorithm development will also keep pace with these emerging ideas.
Moreover, the applications for these new sensors extend far beyond self-driving cars. Clearly, there are many other types of vehicles that could benefit from self-driving technology or advanced driver assistance features as sensor size and cost are reduced in a wide range of operational design domains. But for stationary applications, such as traffic flow management and pedestrian safety systems in congested areas, these benefits are equally important. One can imagine a collaborative system in which trucks, cars, motorcycles, bicycles, and pedestrians are in constant communication with each other about their location, speed, and surroundings.
Overall, high-resolution millimeter-wave radar sensors will play an important role in any situation where there is a need to understand a dynamic environment and where visual camera data is insufficient. The technology of these radar sensors has enough room for development to meet new market needs. Therefore, the creativity of the system designer may be the only limitation in determining the scope of their applications.
English