Many Tier-1s often already have established radar base software that they want to continue using on a new hardware platform.kaptn @ AdobeStock
Alongside the camera, radar is the fundamental sensor technology for ADAS and automated driving. However, radar offers many additional possibilities, and with Antenna in Package (AiP), there is a complete radar solution in a chip package.
The core element of every radar sensor is the combination of (analogue) high-frequency semiconductors and antennas, supplemented by digital computing units that process and evaluate the sensor signal. Therefore, the design of radar sensors brings together several disciplines in a very small space, namely high-frequency design, high computing power, and adequate algorithms for signal detection - and this is exactly where the challenges lie in the development of radar chips and radar sensors.
In doing so, the radar sensors of (the day after) tomorrow must be high-resolution, consume little energy, and have small dimensions - all under high cost pressure. At the same time, OEMs face the difficult task of scaling their (sensor) solutions across all vehicle platforms from small cars to the premium segment, while reusing as much hardware and software as possible.
Resolution in distance, angle, and speed
Resolution is a crucial factor for the performance of radar sensors, as it determines how precisely a radar system can detect, distinguish, and track the movements of objects - and thus how safely and reliably a vehicle reacts to its surroundings. In radar technology, resolution refers to the ability to distinguish multiple objects from each other spatially and temporally. It is considered in three dimensions: range resolution, angle resolution, and Doppler resolution.
The range resolution determines how well the radar can detect two objects at different distances from each other, while the angular resolution sets the accuracy with which the radar determines the angle from which an object is coming. The Doppler resolution, in turn, defines how precisely the radar can capture the relative speed of an object. In principle, the higher the resolution, the better the system can detect, for example, a bicycle next to a car or a pedestrian between parked vehicles.
Image 1: A pedestrian at a distance of 60 m can be clearly detected due to the high resolution and distinguished from the two vehicles.Calterah
High resolution improves object detection and classification - and these are fundamentally important in both the ADAS field and automated driving. Particularly at higher speeds or in heavy traffic, fine resolution is necessary to make quick and correct decisions. By using multiple channels, the resolution can be significantly increased.
Why are multi-channel radar systems more powerful?
In radar technology, a channel is understood to be a combination of a transmission and/or reception path, i.e., a physical or virtual connection through which radar waves are transmitted or received. Multiple channels mean multiple such paths - with multiple antennas and usually multiple signal processing strands. The more channels are used, the higher the resolution of the radar image that is composed of the signals from the individual channels can be. In combination with the appropriate antenna design, the radar can not only measure the distance and speed of an object but also determine its position in space via the azimuth and elevation angle. More channels enable higher spatial resolution, which is particularly important for recognising multiple objects close to each other, such as a pedestrian next to a car. This improves object detection and classification.
Image 2: Even when a pedestrian is almost obscured by a vehicle 40 m away, the radar can clearly recognise them.Calterah
When a sufficient number of channels are used, a proper radar image is created, which is why it is called imaging radar, as objects can be identified in a similar way to a photographic image. A 'simple' imaging radar system can, for example, use 8 x 8 channels, i.e., 8 transmit and 8 receive channels each. A setup with individual transceivers results in a technically highly complex construct. By using just two 4 x 4 radar SoCs (System on Chip) from the Andes family, a simple cascading already creates an 8 x 8 system, which is also significantly cheaper and smaller than a conventional solution based on individual transceivers.
Efficiency and compactness through integration
Because Calterah has developed many individual elements of its SoCs from scratch, the structures are highly optimised and small, but they also consume little power. An example: Before the company launched its UWB products on the market, Class-AB amplifiers in the sensor transmitters were the state of the art. However, Calterah has consistently relied on Class-D amplifiers from the start, which means the system requires only half as much energy to transmit the signal. Similarly efficient solutions are also used on the receiving side, and because the company is already using 22-nm CMOS technology for the radar and UWB chips, the ICs (Integrated Circuits) are small, enabling cost-effective energy-saving value-added systems. This means the SoCs are not only energy-efficient, but less waste heat needs to be dissipated, allowing the systems to become more compact.
What are the advantages of antenna-in-package (AiP)?
In the Alps and Kunlun and Lancang USRR (ultra short range radar) series, the antennas are already integrated into the housing of the SoCs, allowing for compact interior sensors to be realised with these antenna-in-package solutions. These can then implement, among other things, CPD (child presence detection; detection of whether a child is in the vehicle) and seat occupancy including classification of people in a confined space (for example, in the vehicle roof). But also in applications such as door radar, these SoCs enable highly compact system solutions with low power dissipation.
Image 3: In processing the radar signals on the chip, AI is also used in the form of a deep learning model. The CNNs used here are suitable for computer vision applications - here in the form of radar images.Calterah
High frequency and baseband
The available space in which the sensors can be accommodated is very limited, not only in the bumpers, so the sensor unit should have particularly small dimensions. By integrating baseband and high-frequency elements into a chip, the solutions already contribute here, but with an integration level of currently up to 4x4 or 6x6 channels per SoC and the possibility of simple cascading via a serial link to 8x8, new possibilities for space optimisation and scaling arise. With just two of these ICs, a true 8x8 radar transceiver system is created - without the need for major software adjustments. You only need to apply the single-chip transceiver multiple times and then scale the system with minimal development effort, which leads to very short turnaround times for the OEM.
A typical signal flow shows how complex the individual tasks are: first, an A/D converter converts the RF signal into a digital signal, which is then processed, filtered, processed, and transformed by FFT. Calterah uses a special radar signal processing architecture for this purpose. Additionally integrated parameterisable hardware accelerators ensure that even computationally intensive applications such as FFTs run very quickly, although the architecture is inherently very flexible due to its programmability. Further algorithms can also run efficiently in the RISC cores. The RISC-V instruction set has been expanded with commands that enable much simpler programming in the context of object recognition.
The output data is in the form of a pixel cloud. Its actual value is shown in the high accuracy of the individual pixels (detected echoes) as well as in their mutual precision, a property that significantly determines the quality and reliability of the subsequent signal processing.
A continuous object list could also be available at the output, indicating that a specific object is located at a specific position. In addition, the classification of objects is possible, whereby the classified objects can be tracked via an algorithm. Depending on the architecture and design of the vehicle, this can go so far that, for example, an emergency braking is initiated directly from the radar sensor because it has been recognised that an object is approaching too quickly.
What requirements do European OEMs have?
Currently, radar systems based on Calterah-SoC are still predominantly installed in vehicles from Chinese OEMs. To address the needs of European OEMs, there is a dedicated office with several FAEs (Field Application Engineers) located in Munich. For example, there is support for the demand that exists only in the German market, that an ACC radar (Adaptive Cruise Control) should operate at speeds of up to 210 km/h. One of the challenges here is that because the braking energy is proportional to the square of the current speed, the braking distances are correspondingly long at high speeds, so the radar must also detect far ahead.
Not only the FAEs but also the design-in aids provide the necessary support to bring adequate systems to market quickly. In addition to hardware, suitable Gerber files and application software are available to facilitate a quick start.
Many Tier-1s often already have established radar base software that they want to continue using on a new hardware platform. The spectrum of this algorithm ranges from Fast Fourier Transformation (FFT) to object recognition in the point cloud. When the originally analogue sensor signal is converted to a digital signal via an A/D converter, a typical in-cabin radar or a 77-GHz radar generates a raw data volume of well over 1 Gbit/s, which is transferred from the time to the frequency domain via FFT. This requires immensely high computing power. The radar SoCs have an exactly coordinated computing environment for this, which combines programmable elements such as DSPs or RISC-V cores with special hardware accelerators to recognise objects such as people, bicycles, cars, lorries, etc., much better and more specifically; this functionality is particularly important for automated driving.
Radar technology in the interior and for safety functions
Radar technology is suitable for monitoring the driver, including monitoring vital data. The 60 GHz radar chips have already proven themselves in this regard, as well as for general interior monitoring. This allows breathing and heart rate to be recorded with minimal effort. The detection of breathing and heartbeat is one of the most demanding tasks for in-cabin radar systems, as movements in the sub-millimetre range - such as chest movements when breathing or fine heart vibrations - must be precisely detected. Classical signal processing analyses the smallest phase changes over several seconds to reliably filter out periodic vital signs. This approach is combined with convolutional neural networks (CNN) that intelligently evaluate complex signal patterns and reliably distinguish real vital movements from interference sources such as fluttering clothing or vehicle vibrations.
Here too, an extremely precisely tuned board and antenna design is necessary. With antenna in package, for example, it is possible to integrate 6 x 6 antennas together with the radar SoC (HF + baseband) in a housing that measures approximately 15 mm x 15 mm. Directly under the housing label, you can see the antenna patches. When developers use the AiP design, no adaptation of the application is necessary despite the high-frequency specifics. The only adaptation that is still necessary is the adjustment to the bodywork and the paintwork. This coordination usually takes place between OEM and Tier-1. However, with antenna in package, only ultra-short-range radar sensors with a range of up to 50 m are possible, but for the interior, door sensors, etc., this is more than sufficient.
