22 Oct 2024
Radar is an essential sensor for ADAS and AD applications. Through multiple generations of automotive radar, the industry has continually improved radar performance such as sensitivity, resolution, and accuracy so as to increase comfort and safety in today’s cars. Today, as radar engineers continue to push the capabilities of this sensor, existing approaches are running out of steam.
The conventional design paradigm is to increase the number of antenna channels that can be integrated into a single sensor. Traditional medium range radars (MRRs) with 12-16 virtual channels evolved into long range radars (LRRs) with ~200 virtual channels which are just now entering production, and now a new generation of imaging radars is coming onto the market with over 1,000 channels.
A new technology in radar is distributed aperture radar (DAR), pioneered by Zendar. DAR uses multiple simple sensors, operated coherently, to realize a very large aperture with only a small number of channels (30-60). The distributed aperture is not limited by the size of a single sensor, therefore very large virtual apertures of 50-80 cm are possible.
The key radar performance indicators which determine real-world usefulness are angular accuracy and resolution. Radar physics shows that the sensor aperture, not the number of channels, is the strongest determinant of these KPIs. DAR is unique in its ability to attain large apertures with small sensors—opening a new realm of possibility for radar in industries like automotive and autonomous vehicles where sensor size is a design constraint.
Angular accuracy and resolution are essential specs for a radar system. Consider a frequently encountered road scenario. A car has pulled over next to a guardrail in the emergency lane of a motorway, and the driver of the car exits the vehicle.
The radar sensor needs to be able to:
All of this must be done at more than 150 m distance, so that the decision can be taken whether and how much to slow down or change lanes to safely pass the stopped car and driver.
Safely navigating this scenario requires radar specs of:
If any one of these specifications is not met, then the radar output cannot be trusted to implement the driving function. In legacy radar systems such as adaptive cruise control (ACC), only angular resolution was of concern. However accuracy becomes critical for L2 functions and beyond. For instance, if the angular accuracy is not sufficient, we cannot know if the stopped car is safely out of the driving path or partially obstructing it. Similarly, without sufficient accuracy, we cannot adequately detect and localize the driver of the vehicle as he gets out and walks around or stands nearby. In the above scenario, the human and car are separated by at most 0.1-0.4 deg (25 - 100 cm).
Selection of next-generation radars must focus on the accuracy specification in addition to angular resolution.
Coming to radar physics, all these important specs are determined by the radar system’s aperture.
The resolution of any sensing system is determined by the Rayleigh criterion, which states that the resolution is proportional to the wavelength (λ) divided by the sensor aperture (D).
Irrespective of which detection or angular processing algorithm is used, a radar system with larger aperture will always have better resolution than one with smaller aperture.
The accuracy of an angle estimate is determined by the Cramer-Rao bound (CRB) for angle estimation. Deriving this for a radar sensor, the CRB depends on the signal-to-noise ratio (SNR) and the aperture.
The angular accuracy depends very strongly on the aperture! Once again, irrespective of processing algorithms, a radar system with larger aperture will strongly outperform one with a smaller aperture.
An additional factor to consider in the above scenario is the difference in reflectivity between the human and the vehicle or guardrail. This requires the radar sensor to have adequate dynamic range to identify the weaker target (human) in the presence of the stronger target (vehicle or guardrail).
Crucially, because all these targets are real-world extended objects, the radar sensor separates them from each other in both range and azimuth simultaneously. Because modern radar sensors are able to utilize essentially the entire 76-77GHz band, they can achieve range resolutions of 20cm. As a result, in essentially all target configurations, range resolution is sufficient to separate these distinct objects. Moreover, dynamic range of 60dB in range is achievable in many radar sensors. In summary, the range dimension of the radar contributes much of the raw separation capability. It is important to add, however, that the angular accuracy also plays a significant role. Two objects may be separated in range but if their location is poorly estimated then they will nevertheless appear as a single object in the point cloud.
As such, the most relevant dynamic range specification is the combined range/azimuth separability. This is a function of multiple radar KPIs, including both range and azimuth resolutions and accuracies. Typically, this should be measured with real objects rather than with corner reflectors as this will better measure the actual behavior of interest. Some radar designers may advertise impressive pure-azimuth dynamic range which nevertheless are primarily relevant only when separating corner reflectors. In real-world scenarios, more radar specifications come into play in determining good object detection and localization.
The above discussion focused on the role of aperture in determining radar system performance. It is true that the SNR and the number of channels are also relevant. Higher SNRs improve target detectability and separability, and reduce errors in parameter estimation. One potential benefit of a higher channel count, depending on the multiplexing scheme, is increased SNR. A second benefit of adding channels is to simplify the angle processing, for instance by reducing the sidelobe level and therefore permitting simpler detection algorithms to be applied. In contrast the sparser antenna geometry of a DAR system requires more sophisticated detection and parameter estimation algorithms.
However, it is clear from the radar theory that aperture is the most powerful variable in determining resolution and accuracy. These specifications depend weakly on SNR and on the number of channels. Therefore, while it remains important to thoroughly design the radar for high sensitivity, it makes sense to choose an aperture-centric system architecture rather than a channel-centric architecture.
DAR is the only aperture-centric approach to designing radar systems. System designers choose the number of sensors and their placement to realize the desired aperture based on the target radar specs and vehicle requirements. Because the aperture is determined by the spacing between sensors rather than the sensor size, this is not limited by any hardware constraints. In fact, in the future apertures of 1 m or higher could be achieved with a DAR approach. In contrast, the aperture of an imaging radar is limited to the size of the module, and the size of the module must be traded off against cost and manufacturing considerations.
In addition to achieving the best radar performance, DAR systems are considerably more cost-effective than imaging radars.
By focusing on the number of antenna channels, imaging radars drive up cost. Each antenna channel requires silicon and circuit board area, consumes power, and generates large datarates that must be processed inside the sensor. Adding more channels drives up these sources of cost and complexity. Whereas each previous generation of radars consumed less power and used fewer chips than the preceding generation, imaging radars ironically are reversing those trends.
In contrast, DAR systems utilize fewer channels and instead place them cleverly to maximize the aperture. This maximizes the radar performance, without increasing cost. DAR makes it economically possible to increase radar performance without making these solutions unaffordable.
Finally, DAR is inherently software-defined. All the value and performance increase is delivered by the software which combines the radar sensors. This makes DAR very scalable - different trims and performance levels can be implemented purely in software without requiring new hardware designs. Similarly, the system and product development timeline can be decoupled from the time needed to develop and mature new hardware. DAR uses simple MRRs, which are already produced in very high volumes and at very high maturity levels, as the aperture building blocks. As new hardware technologies are introduced over time, those hardware development timelines are easier to decouple from the vehicle platform generations.
By moving to a software-defined radar, DAR helps OEMs unlock faster development cycles for ADAS solutions and quicker product updates. The advantage of software-defined vehicles will compound as time goes on. OEMs who choose hardware-defined radar solutions will limit their L2+ capabilities to more expensive models of cars, pricing out large swaths of potential buyers. Those who choose software-defined radars will be able to bring current luxury trim ADAS functionality to budget-friendly vehicles and capture more market share.
What does it take for OEMs to adopt DAR systems? There are some changes which must be adopted in vehicle architecture and integration to bring DAR technology into vehicles.
First, because DAR uses low-level sensor fusion, it relies on a satellite radar architecture. In satellite radar, processing is moved out of the radar sensor into a zonal or central processor. The vehicle E/E architecture needs to be adapted to support 1 Gb Ethernet traffic from each radar sensor to the zonal or central processing unit where DAR sensor fusion is performed.
Second, vehicle integration of the radars needs to be adapted to accommodate multiple radar sensors. DAR sensor fusion is sensitive to extrinsic errors and uncorrelated vibration between sensors. This sensitivity can be fully mitigated by using a common plastic bracket for mounting all DAR sensors along with a precise online extrinsic calibration algorithm. The multiple radars and the common bracket need to be integrated into the bumper design to accommodate the vehicle design as well as mechanical and safety considerations.
DAR represents a new paradigm in improving automotive radar. By focusing on what matters the most in radar – aperture – DAR systems can achieve the resolution and accuracy that is required for modern L2+ ADAS systems. As SDV architectures are being adopted by the industry, DAR brings a scalable, cost-effective, and software-defined paradigm to automotive radar. With low sensor cost and software defined performance, OEMs can roll out L2+ functionality on lower-priced vehicles—improving accessibility of these features and ultimately making roads safer.