--Collect the "Automotive Driving Automation Classification" (GB/T 40429-2021)--
LiDAR is a sensor that is currently changing the world, and it is widely used in self-driving cars, drones, autonomous robots, satellites, rockets, etc. By determining the propagation distance (TIME of Flight) between the sensor emitter and the target object (TOF) (as shown in Figure 1), the laser analyzes the reflected energy on the surface of the target object, the amplitude, frequency and phase of the reflected spectrum, and outputs a point cloud, thus presenting the accurate three-dimensional structural information of the target.
Figure 1 Lidar ranging and point clouds
Lidar is made up of a laser transmitting unit and a laser receiving unit, and the transmitting unit works by firing outward layers of the laser beam, the more layers, the higher the accuracy (as shown in Figure 2), but this also means that the sensor size is larger. After the transmitter unit emits the laser, when the laser encounters an obstacle, it will reflect and be received by the receiver, which creates a set of point clouds according to the time of each laser beam and return, high-quality lidar, which can emit up to 200 lasers per second.
Fig. 2 Laser point cloud formed by different laser beams
For the wavelength of the laser, laser emitters with wavelengths of 905 nm and 1550 nm are mainly used, and light with a wavelength of 1550 nm is not easily transmitted in the liquid of the human eye. Therefore, 1550nm can greatly increase the transmission power under the premise of ensuring safety. High power can get a longer detection distance, and long wavelengths can also improve the ability to resist interference. However, 1550nm lasers need to use InGaAs, and mass production is currently difficult. Therefore, more LiDAR with Si quality production of 905 nm is currently used. Safety is guaranteed by limiting power and pulse time.
Structure of lidar
The key components of the lidar signal processing signal chain include the control hardware DSP (digital signal processor), the laser drive, the laser emitting light emitting diode, the transmitting optical lens, the receiving optical lens, the APD (avalanche optical diode), the TIA (variable transconductance amplifier), and the detector, as shown in Figure 3. In addition to the transmitting and receiving optical lenses, they are all electronic components. With the rapid evolution of semiconductor technology, performance gradually improves while costs are rapidly reduced. But optics and rotating machinery account for most of the cost of lidar.
Figure 3 Key components of lidar
Types of lidar
At present, there are different types of lidar on the market, which can be divided into mechanical, MEMS, phased array, and flood surface array (FLASH) according to the driving mode.
Mechanical
Take, for example, velodyne's 64-line radar, which was introduced in 2007. It stacks 64 lasers vertically and rotates at 20 rpm. The simple understanding is to turn the laser point into a line by rotating, convert the line into a surface through a 64-line stack, and obtain point cloud data to obtain 3D environmental information.
Mechanical structures require complex mechanical structures, while the measurement of point clouds requires precise positioning of the installation. Considering the impact of the environment and aging, the average failure time is only 1000-3000 hours, which is difficult to meet the minimum requirement of 13000 hours for the depot. And because LiDAR is installed on the roof, the civil field needs to consider the problem of external maintenance, such as the impact of car washing. Therefore, the mechanical structure greatly limits the cost and application promotion.
MEMS
MEMS lidar uses the technology of microelectromechanical systems to drive a rotating mirror that reflects the laser beam in different directions.
The advantages of solid-state lidar include: fast data acquisition speed, high resolution, and strong adaptability to temperature and vibration; through beam control, detection points (point clouds) can be arbitrarily distributed, such as in the distance in front of the main scan of the highway, for sparse side scans but not completely ignored, and enhanced side scans at intersections. Mechanical lidar, which can only rotate at a uniform speed, cannot perform this delicate operation.
Typical applications are Valeo SCALA lidar. It is currently used in the Audi A8 (the first L3 level autonomous vehicle). Installed in the front bumper position, mems technology is used to obtain a scanning angle of 145° and a detection distance of 80m.
Figure 4 Lidar of the Audi A8
Phased Array (OPA)
Principle of light phased array radar: mainly uses the interference principle of light. The position of the central stripe (main lobe) after diffraction can be changed by changing the phase difference of the incident rays in different seams, as shown in the figure below.
Figure 5 Principles of phased array radar
Advantages and disadvantages of phased array (OPA).
merit:
Simple structure and small size: Since there is no need to rotate parts, the structure and size of the radar can be greatly compressed, improving service life and reducing costs.
Simple calibration: mechanical lidar due to the fixed optical structure, adapting to different vehicles often needs to precisely adjust its position and angle, solid-state lidar can be adjusted by software, greatly reducing the difficulty of calibration.
Fast scanning speed: Not subject to the speed and accuracy of mechanical rotation, the scanning speed of the optical phased array depends on the electronic characteristics of the material used, which can generally reach the order of MHz.
High scanning accuracy: The scanning accuracy of the optical phased array depends on the accuracy of the control electrical signal, which can reach more than one thousandth of a measure.
Good controllability: The beam pointing of the optical phased array is completely controlled by the electrical signal, and can be pointed arbitrarily in the allowable angle range, and high-density scanning can be carried out in the key area.
Multi-target monitoring: A phased control array can be divided into multiple small modules, and each module can be controlled separately to lock and monitor multiple targets at the same time.
shortcoming:
The scanning angle is limited: adjusting the phase can only change the central ming pattern by about ±60°, and it generally takes 6 to actually achieve 360° acquisition.
Sidelobe problem: Grating diffraction will form other bright stripes in addition to the central ming, which will cause the laser to form a side lobe outside the maximum power direction, dispersing the energy of the laser.
High processing difficulty: the optical phased array requires that the size of the array unit must not be greater than half a wavelength, and the current working wavelength of the current lidar is about 1 micron, so the size of the array unit must not be greater than 500nm. Moreover, the higher the density of the array, the more concentrated the energy, which increases the requirements for processing accuracy and requires certain technological breakthroughs.
Large receiver surface, poor signal-to-noise ratio: Traditional mechanical radar only needs a small reception window, but solid-state lidar requires a complete receiving surface, so it will introduce more ambient light noise, increasing the difficulty of scanning and parsing.
Flooded surface array (FLASH)
The principle of flood area array is similar to that of a TOF camera, that is, a flash mob, which is not like the MEMS or OPA scheme to scan, but directly emits a large laser covering the detection area in a short period of time, and then uses a highly sensitive receiver to complete the drawing of the image around the environment. It runs more like a camera. The laser beam diffuses directly in all directions, so a single flash can illuminate the entire scene. The system then uses an array of miniature sensors to capture laser beams reflected back in different directions.
Figure 6 Floodlit lidar
One of the great advantages of Flash LiDAR is that it can quickly record the entire scene, avoiding the hassle of moving targets or lidar during scanning. There are two current development directions, one is the single photon counting type of Geiger mode APD to directly generate digital images of photon counts, and the other is the traditional CMOS light intensity simulation to obtain an intensity map and convert the intensity map into distance information.
Data transmission of lidar
Due to the large amount of data in LiDAR, the current control architecture basically uses the raw data of each optical spot to be sent back to the central controller for processing, so it usually uses high-bandwidth networks such as FlexRay or Ethernet for communication. For example, Valeo's SCALA used Flexray in the first generation, and Ethernet in the second generation.
LiDAR usually supports timing at the hardware level, typically providing support for three time synchronization interfaces.
Take the most common rotary lidar data as an example, its data is 10hz, that is, LiDAR rotates a circle in 0.1s time, and cuts the data obtained by the hardware into different packets according to different angles, and each packet contains the data of all points in the current sector, including the timestamp of each point, the xyz data of each point, the emission intensity of each point, the id of the laser transmitter from each point, and so on. The latest Livox Horizon lidar also contains multi-echo information and noise information.
Reprinted from the automotive ECU development, the views in the text are only for sharing and exchange, does not represent the position of this public account, such as copyright and other issues, please inform, we will deal with it in a timely manner.
-- END --