Introduction

For millimeter-wave radar, more precisely 4D millimeter-wave radar, 2021 is bound to be an unusual year. In the wave of “new four modernizations” in the automotive industry, various mainstream automakers’ blessings are pushing 4D imaging technology to the forefront, and its large-scale commercialization is getting closer and closer.

Millimeter-wave radar is not unfamiliar to people, but 4D imaging is, among the many sensors used in autonomous driving, what are its advantages? What role can it play? Who is leading the technological development? And how will it be commercially implemented?

At present, car radar has been used in ACC and AEB, but they are not enough to provide the best performance, nor can they meet the functional requirements of L2+/L3/L4 level autonomous driving applications. Arbe’s imaging technology provides the insight required for autonomous driving systems, making 4D imaging radar the backbone sensor suite. Arbe’s unique and disruptive radar technology enables the application of 4D imaging features in autonomous driving systems, filling the gap of visual sensors in challenging lighting and weather conditions. In addition, 4D imaging radar can provide depth information and instantaneous radial velocity data with almost zero delay.

Delicate Market Pattern of Millimeter-Wave Radar

For many years, Infineon and NXP have almost monopolized the market of millimeter-wave radar chipsets, and STMicroelectronics has been catching up from behind. In early 2020, Infineon announced its entry into the automotive 4D HD LiDAR market, but there have been no corresponding chip releases yet.

As early as the end of 2016, Texas Instruments (TI) introduced a highly integrated 77GHz millimeter-wave radar sensor based on the CMOS process, but this did not change the market pattern of millimeter-wave radar chipsets. Therefore, in late 2018, Texas Instruments launched the concept of 4D imaging millimeter-wave radar.

In fact, Arbe, an Israeli startup founded in 2015, has been working hard on researching and developing a new type of 4D imaging radar for automotive applications.

In January 2019, Arbe released a test version of the vehicle-grade 4D imaging radar product Phoenix, which uses patented chipset technology. Its imaging clarity is a hundred times clearer than similar radars in the industry, and it can provide excellent real-time dynamic and static object separation.

In September 2019, Arbe launched the first brand new high-density radar antenna, which has the highest number of channels, widest field of view, and highest resolution in the market, and can detect pedestrians with unprecedented excellent ability, and separate them from sidewalks, adding new safety guarantees for ADAS and autonomous driving.

In October 2020, Arbe launched the first 2K high-resolution imaging radar development platform, bringing revolutionary changes to customers’ imaging radar system designs.Arbe’s solution is small, lightweight, energy-saving, customizable, affordable, and can meet the autonomous driving needs of all levels of vehicles.

What is the difference between traditional radar and 4D imaging radar?

Radar systems estimate the distance and speed of targets by transmitting a signal and comparing the characteristics of the reflected signal and the transmitted signal. A single-transmitter, single-receiver radar system is called a single-input single-output (SISO) system. If the angle of the object (azimuth, elevation, or both) has to be estimated, mechanical or electronic scanning must be used, which results in inaccurate estimation of fast-moving object angles.

A radar with multiple transmitting and receiving antennas is called a multiple-input multiple-output (MIMO) system, which includes multiple transmitting antennas and receiving antennas.

Thus forming a virtual array, providing higher resolution. The radar performance is related to the number of Tx and Rx channels used in the radar, and using a 2×3 array will obtain 6 virtual channels, Arbe provides 48Tx channels and 48Rx channels, thereby obtaining a virtual array of 2304 virtual channels.

Each sensor has its own strengths and weaknesses

As vehicles become more electrified and electronic, the addition of ADAS functions requires the use of more sensors, increasing the complexity, cost, and weight of vehicles while improving safety. Each type of sensor has its own advantages and disadvantages, so the industry generally adopts a fusion approach to make up for shortcomings.

Cameras and other optical solutions can effectively detect targets, measure distances, provide precise imaging, and track multiple targets. However, their field of view is limited and they are ineffective in harsh weather, strong light, and shadow areas, and there are also privacy issues.

Ordinary radar performs well in harsh weather conditions, can detect existence, direction, distance, and speed, and is a powerful and scalable solution. However, its small antenna array results in low data output resolution, unable to generate rich images; its narrow field of view mainly focuses on a single axis, and its angular resolution is limited, unable to distinguish adjacent targets.

# LiDAR Can Accurately Detect Object Details

LiDAR can accurately detect object details, but it is also not suitable for harsh weather conditions and is expensive. As a result, the current sensors are not mature enough to support future autonomous driving. However, the breakthrough technology of 4D imaging radar can complement its shortcomings while maintaining all the advantages of existing sensors, allowing it to detect objects with ultra-high resolution in any environment and conditions and achieve the required level of safety.

Why Does 4D Imaging Radar Have Higher Resolution?

The inability to distinguish threats from false alarms is the main cause of accidents in autonomous vehicles. Due to the limitation of low spatial resolution, it is impossible to separate targets based on their arrival direction and maintain a low false positive rate, causing the radar to be downgraded to a supporting role in automotive sensors. Arbe’s 4D imaging radar is changing this embarrassing situation.

Real 4D Radar Images

The radar generated by Arbe is a real 4D radar image with ultra-high resolution in distance, azimuth, elevation, and Doppler dimensions. This is achieved through highly reliable target detection, low side-lobe level (SLL), and low false positive rates. It almost never generates virtual objects, thus eliminating false positive and false negative scenes.

In addition to higher accuracy in center and velocity, high-resolution imaging also provides more detailed information about the tracked objects, such as their direction and boundaries, making it more meaningful for fusion with other sensors, such as cameras.

Unparalleled Physical Resolution

Arbe’s high physical resolution is 2-10 times better than competing solutions and supports detection of over 100,000 times per frame, boasting the highest point cloud density on the market. This is due to its digital beamforming using 48 transmit and 48 receive antennas creating a 2,304 virtual channel array. Arbe avoids using unreliable synthetic or statistical resolution enhancement techniques such as super-resolution (SR) and instead leverages a wide aperture array that provides a physical 3dB beamwidth of 1.25° azimuth and 1.5° elevation with low SLL to improve the reliability and safety of the dynamic range. This enables the Arbe chipset to remain effective even in low signal-to-noise ratio (SNR) and multi-target scenarios. High physical resolution and dynamic range provide the ability to separate various objects such as motorcycles next to trucks, cars stuck under bridges, and pedestrians or cars changing tires next to fences.

The system can also track moving objects, draw maps of the environment and stationary obstacles, generate a map of free space, and provide precise positioning while also enabling path planning.

4D Imaging: How It Becomes the Key Player in Autonomous Driving?

From a physics perspective, time is the fourth dimension, as time elements are derived from Doppler. The imaging radar essentially creates an array that dramatically increases measurement density. Traditional 2D radar produces only one point for each object, while imaging radar can provide many points, generating vertical resolution, and enabling better understanding of the object being tracked.

In other words, the time factor has always been a critical part of radar functionality. The fourth element of 4D imaging sensors is “lateral resolution.” 4D imaging radar not only recognizes the horizontal plane but also the vertical plane. For example, a car can determine whether an object is “below” or “above” it by using 4D imaging radar.

For instance, a car is traveling at 80 km/h on a highway, and a motorcycle (a low-reflectivity, small object) is coming from behind at 120 km/h. Unlike cameras and lidars, 4D radar can detect the motorcycle, which was initially far away, and identify that the two objects are moving at different speeds. It can also determine whether an object is moving closer or farther away.

Arbe’s world’s first 2K ultra-high-resolution 4D imaging radar platform eliminates previous resolution limitations and reshapes radar. It can evaluate distance, height, depth, and speed with high resolution, providing a broad view while achieving low false positives. As this technology advances, radar is expected to upgrade from auxiliary accessories to core components of L2-L5 safety autonomous driving.

To summarize, 4D imaging radar has the following advantages:

• Real-time obstacle detection: Providing highly detailed environmental images with a broad view in all weather and lighting conditions, detecting various obstacles in real-time, including small targets like pedestrians or bicycles by determining their movement direction even if they are concealed by large objects like trees or trucks. It provides real-time situational data and alerts for vehicles.

• Long-range detection: Achieves the longest detection range among all sensors and is likely to be the first device to detect hazards. Then it can guide cameras and LiDAR to the area of interest, dramatically improving safety performance.

• Path planning: Provides real path planning, as it can create detailed images of the road within a range of over 300 meters while capturing the size, position, and speed of objects around the car.- Object height separation: Identify whether the object in front of the car facing it, such as a bridge, is stationary, whether it requires a stop, or can be safely driven through.

• Reduce processing and server requirements: By aligning the camera and LiDAR only to the area of ​​interest and using high-quality radar post-processing, the main problem of the current prototype-power consumption-will be solved.

• Significantly reduce production costs: Even at L3 or above, there is no need to equip each car with more than one LiDAR unit, or LiDAR may not be necessary at all, which helps manufacturers reduce costs. The mass production cost of the autonomous driving sensor suite should be less than $1,000, while some of the components and systems used by vehicles tested today cost 100 times that price.

This way, the ADAS system can trust the radar reading, achieve quick response, and prevent unnecessary stops. Therefore, 4D imaging radar provides the basis for navigation, path planning, and obstacle avoidance, supporting the sensing requirements of L4 and L5 autonomous vehicles in terms of safety and accuracy.

Technical breakthroughs of Arbe Radar Development Platform

The technical breakthroughs of the Arbe radar development platform include: high-resolution separation of objects according to azimuth and elevation; detection of the direction and boundary of the object; elimination of false alarms; ultra-high physical resolution; reduced mutual interference; and no Doppler blur. These breakthroughs are specifically reflected in the following six aspects:

• First is the adoption of the most advanced RF chipsets. Arbe’s proprietary millimeter-wave automotive-grade radar RFIC chipset includes a transmitter chip with 24 output channels and a receiver chip with 12 input channels. The AEC-Q100 certified RF chipset uses the new 22nm FD-SOI CMOS process (22FDX) and supports TD-MIMO, with the best performance in channel isolation, noise coefficient, and transmit power of the same kind. Using the latest RF processing technology, Arbe achieved the most advanced RF performance at the lowest cost per channel on the market.

• Second is radar processing technology. Arbe’s independently developed unique baseband processor (Everest) integrates patented radar processing unit (RPU) architecture and embedded proprietary radar signal processing algorithms to process and convert large amounts of raw data in real time while maintaining low silicon power consumption. RPU can process up to 48 Rx channels and 48 Tx channels in real time, generating 30 frames of complete 4D images per second, with an equivalent processing throughput of 3 Tb/s.- The third is mutual interference suppression. With the increasing use of radar sensors in vehicles, some vehicles have up to eight sensors emitting in the same frequency band. Therefore, as the number of radar installations continues to increase, the risk of radar mutual interference is increasing day by day, especially in densely populated urban environments where close-range (reverse and co-directional) interference is more pronounced. When radar interference occurs, detection leaks or false alarms can once again lead to accidents. Arbe’s patented FMCW2.0 system innovation effectively avoids and reduces the interference of other FMCW radar transmitters, minimizing performance degradation or even no degradation.

• The fourth is safety. Everest processors include a dedicated ASIL-D safety island to supervise the safety operation of the system. Arbe’s radar design takes safety into account, with particular emphasis on reducing false alarm rates. Arbe complies with the ISO 26262 standard and can implement continuous built-in radar system self-checking.

• The fifth is post-processing and SLAM algorithms. Arbe’s proprietary post-processing software stack includes a radar-based SLAM solution optimized for enhanced FMCW TD-MIMO imaging radar. SLAM algorithm realizes real-time clustering, tracking, self-positioning, false target filtering and target classification based on radar/radar-camera.

• The sixth is enhanced perception algorithm. As the foundation of advanced perception capability, Arbe’s 4D imaging radar platform includes the ability to accurately infer vehicle speed and lane positioning in real time. Radar data post-processing helps to track and classify targets in the entire field of view of the vehicle, determine their direction and motion vectors, and provide accurate and precise free space mapping to distinguish driveable and non-driveable environments under any weather or lighting conditions.

Accelerating Commercialization Process

Nowadays, radar plays a crucial role in safety systems such as adaptive cruise control (ACC), blind spot detection (BSD), and automatic emergency braking (AEB). However, the radar technology currently available in the market has to make a trade-off between medium resolution in a limited field of view and low-resolution in a wide field of view.

To achieve L4 and L5 vehicles, automakers must adopt next-generation sensing technology by using high-resolution imaging radars to perceive the environment in 4D high resolution within a 100-degree wide angle range at about 1-degree azimuth and 2-degree elevation.

Another important issue is the ability to filter out false alarms. To provide optimal sensitivity, radar typically uses the lowest detection threshold, which can report some noise that needs to be filtered out by post-processing and tracking. However, calibration schemes can only achieve very low side lobe levels.With the emergence of high-resolution imaging radar, many radar suppliers are eager to elevate radar to the only high-speed sensor that can work in harsh weather and lighting conditions.

To accelerate the commercial process, the Arbe 2K high-resolution imaging radar development platform provides Tier 1, OEMs, and new participants with the ability to completely change their imaging radar systems and enhance their perception algorithms. The development platform provides the following components:

• Complete Arbe imaging radar chipset, equipped with RF transmitters and receivers, as well as patented imaging radar processors;
• Radar antennas with the densest channel array in the industry, providing exterior dimensions that comply with current OEM size specifications;
• Software layer for abstracted hardware access and scheduling;
• Reference designs to guide Tier 1 and OEM system development.

Of course, optical sensors are also needed in the autonomous driving sensor suite, and 4D imaging radar enables vehicles to achieve the required safety performance by addressing the following issues:

• The highest reliability in all weather conditions, including fog, heavy rain, pitch-black night and air pollution;
• Detection of obstacles beyond 300 meters, exceeding the requirements of the automotive industry;
• Measurement of Doppler (radial velocity) for each frame;
• A vehicle-mounted security solution that prevents tampering or unauthorized access;
• A compact design that is easy to integrate into the vehicle’s front grille;
• Dynamic calibration support;
• The ability to move beyond the proof-of-concept (PoC) stage into mass production.

In December 2019, Arbe received investments from institutions such as Beijing Automotive Group and Hyundai Group; it is currently cooperating with 25 Tier 1 and OEMs from the United States, Europe, China and Japan to develop a new generation of radar systems based on the Arbe imaging radar development platform. It is expected that Arbe technology-based radar systems will be installed in mass-produced vehicles by 2022.

Becoming an industry standard is just around the corner due to the outstanding advantages of 4D imaging radar. The entire industry is making 4D imaging radar an indispensable and significant element in the autonomous driving sensor suite, providing more sensitive ears and eyes for autonomous vehicles, thus forming a safer automobile market. 4D imaging radar is also expected to become an industry standard and have a profound impact on autonomous vehicles.

This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email bd@42how.com.