Changsha Mozhibi Chen Haowen: The future will inevitably move towards large bandwidth and higher frequency 60G and 79GHz radars [Expert Headlines]

Introduction

Chen Haowen, Chairman of Changsha Mozhibi Intelligent Technology Co., Ltd., gave a keynote speech on the theme of "Making millimeter-wave radar affordable for everyone" at the "2019 Automotive Radar and Sensor Fusion Forward-looking Technology Exhibition and Exchange Conference (Suzhou Station)". From the perspective of Mozhibi's product positioning at 60G and 79GHz, he systematically introduced several key issues and technical solutions for automotive millimeter-wave radar, 3D target detection, parameter estimation and target recognition, resolution of targets at the same distance and speed under limited resources, and interference and confrontation of automotive millimeter-wave radar.

▲ The future will inevitably move towards 60G and 79GHz radars with larger bandwidth and higher frequencies

Mr. Chen believes that the competition in the traditional 24G market is fierce, and 60G and 79GHz frequency band radars will inevitably become the priority products in the future due to their technical advantages such as miniaturization, lightweight, high resolution, high integration (SOC), and low cost. In terms of the market, international manufacturers are on the same starting line, and have first-mover advantages in China.

Medium-range wide-bandwidth radar 77-81GHz band 4G bandwidth pedestrian detection

▲ Characteristics of pedestrian detection

It introduces that the current characteristics of pedestrian detection include small RCS and slow movement speed, which is a typical slow and small target, and the influence of ground clutter cannot be ignored; the movement direction includes straight and cross-sectional forms; the human body movement characteristics of pedestrians are obvious, which are different from vehicles, fixed targets, clutter, etc.; the importance of the human body is highlighted, and the accuracy and timeliness of detection are high; there are high requirements for the identification of target types such as pedestrians and vehicles to provide accurate attribute information.

The advantages of large bandwidth are:

◎High range resolution means more precise target resolution in the range-Doppler grid, and the probability of different targets being in the same resolution unit is smaller;

◎The higher the resolution, the more likely it is that the target will occupy multiple distance units and become an extended target;

◎The higher the resolution, the finer the division of clutter units, which may improve the signal-to-clutter ratio;

◎The higher the distance resolution, the richer the features of the target at the radial distance, and the more conducive it is to target identification based on these features.

Therefore, the movement characteristics and features of pedestrians are more obvious, and a large bandwidth is conducive to the detection and recognition of pedestrian targets.

▲ The relationship between pedestrian echo and clutter in a typical scene

According to the vehicle lane driving rules and pedestrian movement characteristics, the typical relationship between vehicle-mounted radar and pedestrians is divided into three situations: same direction , opposite direction , and cross-border .

Reverse: The pedestrian's Doppler frequency is greater than the maximum frequency shift of the sidelobe clutter, and the target appears in the clutter-free area;

Same direction: The vehicle speed is much greater than the pedestrian speed. The pedestrian target echo falls into the main lobe clutter area or falls between the main clutter and the height line clutter, which is related to the pedestrian speed and pitch angle.

Crossing: If the direction of pedestrian speed is exactly the same as the direction of vehicle-mounted radar illumination, the pedestrian echo will fall into the main lobe clutter area; if it is not completely vertical, it may fall into the main lobe clutter area, which is related to the direction angle. The relationship between pedestrian echo and clutter in range-Doppler is closely related to pedestrian speed, travel direction, distance between vehicle and pedestrian, pitch angle, etc., and may appear or alternate in the non-clutter area and the main clutter area .

▲ Pedestrian detection processing method

Therefore, there are two types of pedestrian detection methods:

A - Pedestrian echo is not in the main lobe clutter area : When the echo of pedestrian targets does not fall into the main lobe clutter area, the clutter does not block the target. At this time, the key to pedestrian detection is to accurately estimate the background , control false alarms , and increase the probability of correct detection of the target.

B - Pedestrian echo is in the main lobe clutter: including the main lobe clutter shielding the pedestrian target echo, resulting in a low signal-to-clutter ratio. At this time, the focus of pedestrian target detection is to ensure the detection probability of the target, to ensure that no missed detection, no lost target. Processing method: By lowering the detection threshold, it can ensure that the target information of the pedestrian is not lost, but at the same time it will increase the probability of false alarms, resulting in more points passing the threshold. The main task at this time is to eliminate the false alarms caused by clutter through signal processing means to achieve stable detection of pedestrian targets.

▲ Vehicle-mounted radar recognizes pedestrians

After the vehicle-mounted radar correctly detects pedestrian targets from ground clutter through various signal processing methods , it is also necessary to judge the type and attributes of the detected targets , that is, to identify the targets and classify the pedestrians, vehicles, animals, small fixed targets, etc. that may appear in the field of view, so as to provide the driver with accurate target type information to assist decision-making and ensure driving safety.

Based on the characteristics of pedestrians and the difference between them and other types of targets, in addition to the conventional method of using the distance, speed and characteristics of pedestrians for rough judgment and classification , there are two typical features that can be used for target recognition.

1) Pedestrian recognition based on micro-motion features

2) Pedestrian recognition based on one-dimensional range profile features

Traditional vehicle-mounted radars mainly detect target environments in two dimensions, namely, distance and azimuth, while the actual target environment is three-dimensional (distance, azimuth, and elevation), which is a lack of dimensionality. The meaning of 3D detection is that on the basis of existing processing, through array formation design and corresponding signal processing, the vehicle-mounted radar has the ability to measure angles in the elevation dimension, thus achieving a three-dimensional description of the distance, azimuth, and elevation of the target environment.

▲ 3D object detection, parameter estimation and object recognition

Typical application scenarios of 3D detection: detection of viaducts, height-restricted roadblocks, height-restricted culverts and tunnel entrances; detection of driving signs; detection and early warning of falling objects from high altitudes; and identification of small, medium and large vehicles by combining pitch angle measurement. The basic formation requirement of 3D detection is based on azimuth measurement and resolution. The pitch dimension mainly considers that the vehicle-mounted radar can measure the pitch angle of the target, and there is no need to consider target resolution in the pitch dimension.

Detection, estimation, and recognition characteristics of 3D detection

From the perspective of radar detection processing flow, compared with the existing 2D detection, the 3D detection of vehicle-mounted radar mainly adds the angle measurement of the pitch dimension in signal processing. The target detection is still a two-dimensional constant false alarm detection process in the range-Doppler plane;

In terms of parameter estimation, the target distance, speed, azimuth and pitch angle are mainly measured using phase comparison. In terms of target recognition, the target's pitch angle information is added, which can roughly estimate the target's height, thereby adding one-dimensional scale information compared to two-dimensional detection, which is more conducive to the determination and identification of target types.

When introducing the resolution of targets at the same distance and speed under limited resources, it is introduced that the same distance and speed means that two targets are in adjacent lanes, and the radial distance difference between the two targets and the vehicle-mounted radar is less than the distance resolution , and the Doppler difference between the two targets for the vehicle-mounted radar is less than the speed resolution , making it impossible for the radar to distinguish the two targets in the distance-Doppler plane.

▲ Two targets are in adjacent lanes with the same distance and speed

The solution is:

Solution from the perspective of array

The essence is to improve the angular resolution performance of the array, so that when two targets are inseparable in terms of distance and speed, the target resolution can be achieved by simply relying on the angular resolution capability of the array. Requirements: 100 meters, 2° -- à 3dB beam width 20°, number of virtual array elements greater than 10.

Solution from the perspective of super-resolution

The essence is to obtain the DOA values ​​corresponding to the two targets respectively without changing the existing array formation, thereby realizing the angle estimation of the two targets. However, the conventional super-resolution method has high computational complexity and requires matrix eigenvalue or singular value decomposition or spectrum peak search. The computational cost is too high and the system cannot be implemented. The key point is: equivalent substitution fast calculation .

▲ In-vehicle forward-looking MIMO 2D/3D imaging

If multiple radar transceiver components are installed and arranged on the front/rear bumper of the vehicle to construct a one-dimensional linear array or a two-dimensional array , instantaneous two-dimensional/three-dimensional forward-viewing area imaging (also known as MIMO-SAR) can be achieved based on MIMO signal processing to achieve two-dimensional/three-dimensional mapping of targets in the field of view directly in front of/behind the vehicle. Compared with conventional forward/rear-view non-imaging radar processing, more detailed scene description , more accurate target detection and type judgment can be achieved , providing vehicle drivers and intelligent driving systems with more three-dimensional, rich and comprehensive information to assist safe driving.

The problem lies in the form of interference: when two/more vehicles are driving in opposite directions , the transmission signals of other vehicle-mounted radars of the same frequency enter the radar receiver of the vehicle, causing interference; when two/more vehicles are driving in the same direction , the reflection echo of the transmission signals of other vehicle-mounted radars of the same frequency irradiating other targets enters the radar receiver of the vehicle, causing interference; other non-vehicle radar signals of the same frequency band enter the receiver, causing interference; How to solve the existing problem of anti-radar same-frequency interference? Mr. Chen put forward several suggestions:

▲ Existing methods to resist radar co-frequency interference

Mozhibi Radar adheres to the determination to serve the industry and build an industrialized, standardized process, automated production, and replicable process system. In solving problems that require good use, it integrates the technology of a professional team and accelerates the iteration of technology to solve industry pain points. Its goal this quarter is to build an automated production line with an annual output of 300,000 units.