4D Radar: Extraordinary sensors for the cars of tomorrow

Nowadays, any new car without navigate-on-autopilot (NOA) functions isn’t taken seriously. The increasing availability of high-speed NOA, highway NOA, and even urban NOA, confirms that Level 2+ (L2+) driving is an irreversible and unstoppable trend. 

The full realization of NOA requires three main sensing technologies – camera lens, millimeter wave (mmWave) radar, and light detection and ranging (LiDAR). Each of these technologies has its own strengths and limitations, which complement the others.

The high resolution of the well-known LiDAR is indisputably outstanding. However, as the technology of 4D imaging mmWave radar continues to evolve (4D represents the four dimensions of range, velocity, azimuth angle, and elevation angle), it is quickly becoming a viable alternative to LiDAR in many application scenarios. Many carmakers are replacing or aiming to replace LiDAR with less costly 4D imaging mmWave radar. In fact, it is estimated that L2+ models will account for 50% of the car market by 2030, indicating a bright future for 4D imaging mmWave radar in the future of the automotive industry.

What kind of 4D imaging mmWave radar meets the requirements of NOA? Let’s take a look at Avnet’s S32R41+2*TFE82 dual-chip cascade program and explore it from three different angles.

Excellent radio frequency (RF) front-end sensing

4D imaging mmWave radar is used for identifying humans or objects, such as pedestrians and bicycles. This requires the sensor to “see” things accurately at long distances and to distinguish small objects with reliability.

Imaging mmWave wave radar should be able to detect, distinguish, and track multiple static or moving targets at a maximum distance of 300 meters, even when the targets are very close to each other. RF performance is optimized with enhanced RF link budget, output power, noise figure, and phase noise, bolstering detection capabilities and resolution.

The TEF82 RFCMOS automotive radar transceiver is a high-performance, single-chip, low-power automotive frequency-modulated continuous wave (FMCW) radar transceiver that operates from 76GHz to 81GHz, covering the complete automotive radar frequency band. This type of fully-integrated RFCMOS chip contains three transmitters, four receivers, an analog-to-digital converter (ADC), a phase rotator, and a voltage-controlled oscillator (VCO) with low-phase-noise.

TEF82’s RF performance is enhanced with a power output of 13.5dB, low noise figure of 11.5dB, and low phase noise of 95dBc/Hz@1MHz, supporting cascaded high-resolution imaging radar.

TEF82 also supports frequency sweep with a slow frequency drift upon start, and the dedicated register is set to be simple and user-friendly. The starting frequency of each chirp is increased (or decreased, depending on the direction of the chirp) compared to that of the previous chirp. The waveform design shortens the time it takes to sweep a single chirp, which allows the blur-free upload speed to be higher when performing Doppler effect calculation. A higher sweep bandwidth can also be obtained through the chirp cascade method, further enabling a higher distance resolution. The different bandwidth of each chirp can also prevent radar interference to certain extent.

The RF front end’s excellent performance ensures that the chip meets the requirements of 4D mmWave radar.

Fully-integrated hardware processing capabilities

4D mmWave radar generates high-density point clouds and sub-degree resolution, which yield huge amounts of data. In order to process this high-density data, efficient processors and specialized radar accelerators are paired with advanced algorithms.

The powerful computing capacity of the processor allows the radar to process large amounts of data in real time and generate accurate three-dimensional environmental images. By employing multiple input, multiple output (MIMO) technology, the radar system can access more virtual channels, thereby improving the accuracy and range of target detection and tracking. The data is processed quickly right after being collected by the sensor, generating instant responses that fulfill the immediacy required for autonomous/assisted driving.

The S32R41 processor uses Arm Cortex-A53 and Cortex-M7 cores, combined with the dedicated radar processing accelerator SPT3.5 and internal 8MB SRAM. It integrates two Ethernet and two CANFD interfaces.

SPT 3.5 is a significant improvement upon the previous-generation SPT 2.8. The vector floating point unit (VFPU) on BBE32 DSP powers new radar post-processing functions. In addition, larger memory enables a significant increase in the number of radars, while supporting up to two cascaded transceivers for advanced high-resolution radar.

Signal Processing Toolbox (SPT) is the general term for various hardware accelerators used for mmWave radar digital signal processing. They include: fast fourier transform (FFT), max signal processing (MAXS; peak search), Copy, velocity mapping toolbox (VMT; mathematical operations for modulus and logarithm), HIST (histogram), and SORT (sorting). In other words, the basic radar signal processing in S32R41 is performed using hardware accelerators with higher speed and lower power consumption. Other digital signal processing in 4D mmWave radar, such as matrix operations, matrix inversion, and vector operations of matrix decomposition, are performed by digital signal processors (DSPs).

Friendly development environment

As new cars are being launched more frequently, the product life cycle continues to shorten. Due to the complexity of radar development, which involves RF antennas, high-speed hardware and complicated algorithms, more manufacturers now pay close attention to the tools, references, and standard software provided.

Avnet’s dual-chip cascade program provides antenna references and corresponding radar calibration algorithms. It also features Avnet’s aperture extrapolation technology, which uses time-constant signal processing algorithms, such as Burg and Marple, to linearly extrapolate the receiving aperture without having to expand the aperture by receiving and transmitting more. Avnet’s program also supports Python users with a friendly development environment in which RF engineers can evaluate the RF performance of TEF82 easily and algorithm engineers are able to develop and verify their own algorithms with efficiency.

Conclusion

Radar sensors have become essential components for Advanced Driver-Assisted Systems (ADAS) on modern vehicles. Compared to more established technologies such as LiDAR, 4D mmWave Radar promises a better RF front end, higher processing capacity, and a friendlier development environment. These advantages secure its vital role in the development of the new generation of extraordinary sensors in the ADAS domain.

The advent of cost-effective 4D imaging mmWave radar technology will have a significant impact on sensor combinations for ADAS on L2+ and higher-level cars. With the widespread application of L2+ safety and comfort functions, driving is becoming safer, smarter and more enjoyable.