LCD Display Inverter

Display Inverter / VGA Board / LCD Controller

4D imaging radar implements both software and hardware, and its performance is getting worse

Traditional millimeter-wave radars do not have the ability to measure altitude, so it is difficult to determine whether the stationary object in front is on the ground or in the air, and cannot refine braking scenarios, such as manhole covers, speed bumps, and other low-level obstacles on the ground that do not require braking; traffic signs, gantry, Overpasses and other air obstacles that do not require braking; as well as larger obstacles on the road such as vehicles that require braking and triangular cone barrels.

The “2020 Radar Industry Situation Report: Manufacturers, Applications and Technology Trends” released by yole pointed out that the automotive market is expected to grow at a CAGR of 11% by 2025, as automotive application radars have become standard equipment and become more common in test scenarios. After the rigor, two major trends are emerging: the first is to move towards imaging radars that can more accurately describe the scene in front of and behind the vehicle; the second is to increase the number of sensors around the vehicle to improve scene perception through coordination. The emergence of 4D imaging millimeter-wave radars in recent years reflects these trends.

1. 4D millimeter wave radar complements the shortcomings

Traditional millimeter-wave radars do not have the ability to measure altitude, so it is difficult to determine whether the stationary object in front is on the ground or in the air, and cannot refine braking scenarios, such as manhole covers, speed bumps, and other low-level obstacles on the ground that do not require braking; traffic signs, gantry, Overpasses and other air obstacles that do not require braking; as well as larger obstacles on the road such as vehicles that require braking and triangular cone barrels. For this reason, in order to avoid the frequent occurrence of false braking, the AEB algorithm decided to reduce the confidence weight of the millimeter-wave radar, focusing on the visual perception results. However, the challenge of visual perception is that the target obstacles must be trained in advance, and the model library cannot exhaust all types, so many static obstacles have become “slip through the net”. In addition, even if there is a model library, another challenge is that Whether the neural network can correctly identify the obstacles ahead. As a result, it often happens that there are obstacles ahead, but the self-driving car still hits it.

4D millimeter-wave radar is also known as imaging radar. “4D” refers to the addition of height-dimensional data analysis of the target on the basis of the original distance, azimuth, and speed, and can realize information perception in four dimensions of “3D + speed”.

The concept of “imaging” refers to its ultra-high resolution, which can effectively analyze the contour, category, and behavior of the target. This means that the 4D millimeter-wave radar system can adapt to more complex road conditions, including the identification of smaller objects, the detection of partially occluded objects, and the detection of stationary objects and laterally moving obstacles.

By upgrading to 4D millimeter-wave radar, the AEB algorithm can take more into account the perception results of the millimeter-wave radar, so as to identify static obstacles on the road with a higher probability. Combined with the advantages of its high resolution, it can more effectively analyze the target. profile, category, behavior, and then know when to brake (to avoid missing brakes).

2. Implement both software and hardware to achieve mass production

In the millimeter wave radar chip circle, leading manufacturers believe that the key to realizing 4D imaging lies in multiple antennas, and the technical threshold is not high. For many years, the market has been dominated by Infineon and NXP. At the beginning of 2020, Infineon announced to cooperate with Oculi, the first vehicle-mounted 4D high-definition point cloud imaging radar, to enter the automotive-grade imaging radar market, but no product has been seen so far.


Millimeter wave radar market structure and trend

In order to seize the market, Texas Instruments (TI) launched the AWR1642 series of highly integrated 77GHz millimeter-wave radar sensors based on CMOS technology at the end of 2016. It is a single-chip product that integrates DSP and MCU for short- and medium-range scenarios. The chipset market is still resigned.

Since it is impossible to compete head-on with the millimeter-wave radar chip giant, it can only take the road of saving the country through a curve. How can you see it? The above-mentioned Aoku pioneered the vehicle-mounted 4D imaging radar, and it is the chip of Texas Instruments that is used in practice.

Texas Instruments’ 4D imaging millimeter-wave radar concept was launched at the end of 2018, and it launched a complete design scheme of 4-chip cascaded 4D millimeter-wave radar based on AWR2243 FMCW (frequency modulated continuous wave) single-chip transceiver. The most difficult antenna is also integrated into it. Embedded four-element series-fed patch (4-element series-fed patch) antenna.


PCB Antenna for AWR2243

Based on TI’s low-power 45nm RFCMOS process, the AWR2243 FMCW transceiver implements 3Tx and 4Rx systems with built-in PLL and A2D converters in a single chip. Simple programming model changes enable multiple sensor implementations (short, medium, long) and can be dynamically reconfigured to enable multimodal sensors. The AWR2243 transceiver is offered as a complete platform solution, including reference hardware designs, software drivers, example configurations, API guides, and user documentation.


Vertically assembled EVM prototype

Algorithms include MATLAB MIMO and beamforming options, and the turnkey project has greatly lowered the threshold for 4D imaging millimeter-wave radar technology. It also allowed Aoku to be the first to try it.

The AWR2243 is a 76GHz to 81GHz automotive second-generation high-performance MMIC. At present, it has become the main cascade solution for 4D imaging millimeter-wave radar in China and even in the world.


2-chip cascade imaging radar

In March 2021, Aoku used TI chips to realize the Eagle, a forward-facing radar with ultra-high angular resolution 4D imaging, which is a product made of both software and hardware. Eagle 77GHz imaging radar uses the popular TI millimeter-wave radar chip in the market. Aoku’s unique software enables the millimeter-wave radar to achieve 4D high-definition imaging, with a detection distance of more than 350 meters.

Eagle realizes high angular resolution and elevation information in a wide field of view on a dual-chip platform. Through the virtual aperture imaging software driven by proprietary AI algorithms, the angular resolution is increased by 50-100 times; the multi-virtual antenna method completely solves the problem In order to solve the problem that has plagued the automotive millimeter-wave radar industry for decades, the angular resolution can only be improved by increasing the number of physical antennas, and the radar has been redefined with software. At the same time, the BOM cost of 4D imaging radar products is similar to that of ordinary millimeter wave radars, but the performance has been crushed.

Eagle provides 0.5° horizontal x 1° vertical angular resolution in a 120° horizontal/30° vertical wide field of view. Its long-range and high angular resolution enable it to be used in a variety of autonomous driving applications, including high-resolution radar mapping and localization, autonomous path planning and obstacle avoidance, object detection and tracking, indoor navigation, virtual fencing, and more.


Comparison between traditional commercial radar and Aokura radar

Traditional radar waveforms are single-frequency, repetitive, and non-adaptive, and the only way to generate multiple waveforms is to increase the number of receiving antennas. The virtual aperture imaging waveform is adaptive phase modulation. Each receive antenna produces a different phase response at different times, and the data is then interpolated and extrapolated to create a “virtual aperture.” Intelligent software that uses artificial intelligence to learn and adapt from its environment can continuously improve with exponentially growing data.


Traditional Radar Waveforms and Virtual Aperture Imaging Waveforms

The resolution of conventional radar depends on the number of antennas. This means that performance is fixed and more hardware is required — more antennas, more processing, larger size, and extra cost — to achieve higher angular resolution. Additional physical antennas can linearly increase performance, but cost, size, and power consumption increase exponentially, limiting the number of antennas that can be used in commercial radars.

Eagle uses a low-cost, low-power dual-chip (6T8R) hardware platform to provide angular resolution that can only be achieved by 8-chip cascade (24T32R) used in traditional radar. Higher angular resolution through software calculations and data can follow the exponential growth of Moore’s Law, enabling longer detection distances and lower costs.


Trends in the number and angle-resolved performance of “array” physical MIMO receivers

According to reports, Aoku’s high-resolution point cloud data can be fused with other sensors, such as cameras and lidars, at the raw data level to achieve deep sensor fusion capabilities, ensure operation under all-weather conditions, and obtain the best tracking results.

3. Sniper TI 4D radar processor

Although Infineon, the leader of millimeter-wave radar chips, has not moved yet, NXP has long been unable to sit still. In early June, NXP announced the mass production of two automotive chips based on TSMC’s 16nm process – the automotive network processor S32G2 and the radar signal processor S32R2949 (radar sensor chipset), and said it will pave the way for the future use of TSMC’s 5nm process.


S32R294 chip

The S32R294 can process 4D point cloud radar signals and will provide OEMs with the performance needed for scalable solutions, including advanced corner radar, long-range forward radar, and advanced multi-mode usage scenarios such as simultaneous blind spot detection, lane change assist and Elevation angle sensing, etc.


S32R294 Application Cascade Block Diagram

The size of S32R294 is the same as NXP’s previous generation chip S32R274. The size of the chip is 7.5mm × 7.5mm, but its performance is doubled. The S32R294 has two Power Architecture e200z7 32-bit cores for post-processing and task scheduling of radar signals, such as super-resolution algorithms, signal clustering, target tracking, etc. It also has a pair of lockstep z4 cores for running functional safety related software such as AUTOSAR OS, output decision instructions etc.


S32R294 block diagram

S32R294 has a built-in radar signal acceleration unit, referred to as SPT2.8, which performs hardware acceleration for the most resource-consuming operations such as FFT, modulo, peak detection, and histogram statistics of radar intermediate frequency signals. It is a signal processing acceleration unit specially designed for FMCW radar.

The software reuse rate between S32R294 and previous generation processors is as high as 80%. Thanks to the 16nm process, the power consumption of the S32R294 is less than half of the previous generation 55nm process processor, around 0.9W, which helps customers develop high-performance, low-power millimeter-wave radar products.

The S32R294 processor is ASIL D ISO26262 certified. Its dedicated hardware encryption engine CSE (Cyptographic Services Engine) supports high-level encryption algorithms such as security boot.

The various configurations of the S32R294 support the development of a full range of applications from entry-level to high-end, such as intermediate frequency signal processing such as one transmitter and three receivers, three transmitters and four receivers, and six transmitters and eight receivers. The intermediate frequency signal processing is mainly to obtain the distance, speed and angle information. The basic distance, speed and angle information are calculated by SPT2.8, and the subsequent data are sent to the Z7 and Z4 cores to realize super-resolution algorithm, signal clustering, target tracking and decision-making. and other functions. Target-level data and decision-making instructions are output to the back-end body control unit or ADAS domain controller through the CAN FD interface.

The six-transmitting and eight-receiving millimeter-wave radar supports up to two chips in cascade, and its two MMICs are NXP TF82 series microwave integrated circuits, which are connected by LO to achieve inter-chip synchronization. The waveform of the MMIC chip is controlled by the MCU through the SPI configuration channel to transmit its waveform. The intermediate frequency signal received by the receiving link is also transmitted back to the MCU through the MIPI-CSI interface for subsequent processing.

This shows that NXP’s radar chips are very flexible and can handle all applications from low-end to high-end. The more channels the radar receives, the wider its field of view, enabling advanced driver assistance functions such as lateral collision avoidance warning, valet parking, and 4D point cloud imaging.


S32R294 Target Radar Application

Like Texas Instruments, NXP also offers the S32R29 an MCU evaluation board for developing automotive radar applications. The evaluation board has access to communication interfaces such as Ethernet and CAN, MIPI-CSI2, and interfaces with radar transceivers and GPIO headers to access other MCU functions.


S32R29 EVB Automotive Radar Application Evaluation Board

The S32R29 EVB also facilitates software development and evaluation of signal processing toolkits to improve radar math speed, functional safety or information security for the S32R29 MCU. This is nothing more than to help radar manufacturers use this chip to make products and achieve mass production as soon as possible.

4. Make a 4D imaging radar that fits the national conditions

Everyone knows the complexity of China’s road conditions, or Tesla will not be acclimatized. To this end, Huawei, which does not make complete vehicles, also spares no effort to develop various sensing technologies, and high-resolution 4D imaging radar is one of them.

Huawei believes that one of the important reasons for the lack of perception in autonomous driving accidents in recent years is the inability to effectively identify stationary vehicles or accident vehicles, and inaccurate judgments on isolation piles or guardrails, which can lead to serious accidents. The high-resolution 4D imaging radar can detect the four dimensions of the target, including speed, distance, horizontal angle and vertical angle, and solve the problems of insufficient horizontal resolution capability of traditional radar, no support for vertical resolution, resulting in unclear and inaccurate vision. 4D imaging radar meets the requirements of full target, full coverage and multi-working conditions perception, and gradually approaches the ideal sensor target, and will form effective fusion and redundancy with cameras and lidars.

While greatly improving the resolution, confidence of target detection and detection range (such as distance and field of view (FOV)), 4D imaging radar has evolved a high-density point cloud like lidar, enabling rich perception enhancement applications. For example, environment characterization, radar composition, positioning, etc., it can also better achieve 360° detection around the vehicle through multi-radar point cloud-level fusion.


Huawei High Resolution 4D Imaging Radar

The improvement of Huawei’s high-resolution 4D imaging radar capabilities is reflected in three aspects:

Large array of high resolution. There are 12 transmit channels and 24 receive channels, which is 24 times higher than the conventional millimeter-wave 3-transmit and 4-receive antenna configuration, and 50% more receiving channels than typical imaging radars. It is said that this is the imaging radar with the most antenna configurations that can be mass-produced in the short term. |

Large field of view without blur. The horizontal field of view has been increased from 90° to 120°, the vertical field of view has been increased from 18° to 30° (equivalent to Aokura), and the vertical detection distance has been increased from 200 meters to more than 300 meters.

4D high density point cloud. The 4D point cloud includes speed, distance, horizontal angle and vertical angle. Compared with the lidar point cloud, there is one more speed dimension to analyze the target.

In the actual measurement, Huawei’s high-resolution 4D imaging radar showed some features:
When the front is congested, it can brake comfortably at 130km/h;
The cone detection distance is up to 110 meters, which can distinguish stationary vehicles beside the guardrail;

Obstructed or partially occluded target perception, you can see the car in front, the car in front and the car in front, and even the outline of the chassis of the car in front. By sensing the car in front, you can predict the sudden deceleration of the car in front in advance and other actions to reduce the risk of serial rear-end collisions.


Non-line-of-sight front vehicle perception

Chi Linchun, President of Huawei’s Smart Car Solutions BU Marketing and Sales Service Department, said when introducing the progress of the intelligent driving business: “Huawei’s antennas have achieved 128 transmissions and 128 receptions, which are very advanced. Huawei’s millimeter-wave radar is an advanced technology in communication technology. developed on the basis of

5. The battle for performance intensifies

TI’s AWR2243 is 3 transmitters and 4 receivers, which can be cascaded with 4 chips to achieve 12 transmitters and 16 receivers; Aoku Eagle using TI AWR2243 is a dual-chip 6 transmitters and 8 receivers. With the help of software, it can achieve 24 transmitters and 32 receivers; NXP S32R294 supports up to 6 transmissions and 8 receptions, and supports 2 cascades, that is, 12 transmissions and 16 receptions (equivalent to TI); Huawei currently has 12 transmissions and 24 receptions. It can be seen that if the radar manufacturer has its own unique software technology, the performance of the chip (hardware) can be improved again.

There is also a domestic millimeter-wave radar manufacturer using the TEF82xx MMIC launched by NXP in December 2020 to make a 4D imaging radar, which is said to have twice the performance of a foreign manufacturer’s product and double the performance of the domestic 2022H2 SOP product, and Mass production 18 months earlier. The technical details have not yet been obtained, and will be introduced later.

The Links:   PM100RRS060 LP121X04-B2 MODULESTOCK.COM