From the perspective of optoelectronic technology, autonomous driving still has these problems to be solved!

LiDAR and competing sensor technologies (camera, radar, and ultrasonic) intensify the need for sensor fusion and the need for careful selection of photodetectors, light sources, and MEMS galvanometers.

Advances in sensor technology, imaging, radar, light detection and ranging technology (lidar), electronics, and artificial intelligence have enabled dozens of advanced driver assistance systems (ADAS), including collision avoidance, blind spot monitoring, lane departure warning, and parking assistance. Synchronizing these systems through sensor fusion allows fully autonomous vehicles to monitor their surroundings, warn the driver of potential road hazards, and even take evasive actions independent of the driver to prevent collisions.

Self-driving cars must also distinguish and identify objects ahead at high speeds. Using range-finding technology, these self-driving cars must quickly build a three-dimensional (3D) map at a distance of about 100m and create high-angular resolution images at a distance of up to 250m. If the driver is not present, the vehicle's artificial intelligence must make the best decision.

One of several basic ways to accomplish this is to measure the round-trip time of flight (ToF) of a pulse of energy from the autonomous vehicle to the target and back again. When you know the speed of the "pulse" through the air - the pulse can be ultrasound (sonar), radio waves (radar) or light (lidar) - the distance to the reflection point can be calculated.

The ToF method used in the demo made using APD

Of the three ToF technologies, LiDAR is the best choice for providing higher angular resolution images because it has smaller diffraction characteristics and beam divergence, which can better identify adjacent objects than microwave radar. This high angular resolution is especially important at high speeds, providing enough time to respond to potential hazards such as head-on collisions.

Selection of laser light source

In a ToF lidar, the laser emits a pulse of duration τ, which triggers an internal clock in the timing circuit at the moment of emission (illustrated below). The light pulse reflected from the target reaches the photodetector, which converts it into an electrical signal output that stops the clock. This way of measuring the round-trip ToF Δt time can calculate the distance R to the reflection point.

If the laser and photodetector are physically located at the same location, the distance is determined by the following formula:

where c is the speed of light in a vacuum, n is the refractive index of the propagation medium (approximately 1 for air), and there are two factors that affect the distance resolution ΔR: the uncertainty δΔt in measuring Δt and the spatial error w (w = cτ) caused by the pulse width.

The first factor represents the distance resolution ΔR = 1/2 cδΔτ, while the second represents the distance resolution ΔR = 1/2 w = 1/2 cτ. If the distance is measured with a resolution of 5 cm, the above relationship means that δΔt is about 300 ps and τ is about 300 ps respectively.

Time-of-flight lidar requires photodetectors and subsequent electronics with very small timing jitter (the main contributor to δΔτ) and pulsed lasers capable of emitting short pulse durations, such as relatively expensive picosecond lasers. Lasers in typical automotive lidar systems today produce pulses with a duration of about 4ns, so reducing beam divergence is necessary.

The divergence of the beam depends on the ratio of the wavelength and the size of the transmitting antenna (microwave radar) or the lens aperture size (lidar). Microwave radar has a larger ratio, so the divergence is greater and the angular resolution is lower. In the picture, the microwave radar (black) will not be able to distinguish between the two cars, but the lidar (red) can.

One of the most critical choices for automotive lidar system designers is the wavelength of light. Several factors govern this choice:

Safe for human vision

Propagation characteristics in the atmosphere

Availability of lasers and availability of photodetectors

The two most popular wavelengths are 905 and 1550 nm. The main advantage of 905 nm is that silicon absorbs photons at this wavelength, and silicon-based photodetectors are generally cheaper than the indium gallium arsenide (InGaAs) near-infrared detectors required to detect 1550 nm light.

Hamamatsu's near-infrared MPPC (silicon photomultiplier tube) that can be used for autonomous driving lidar has high detection efficiency at 905nm, fast response speed, and a wide operating temperature range. It is suitable for lidar applications in various occasions, especially long-distance measurement using TOF ranging method.

However, 1550nm is safer for human vision and can use lasers with greater radiation energy per pulse - an important factor in the choice of optical wavelength.

1550nm detector

Hamamatsu InGaAs APD G8931

Atmospheric attenuation (in all weather conditions), scattering by particles in the air, and reflectivity of target surfaces are all wavelength dependent. The choice of wavelength for automotive lidar in these conditions is a complex issue due to the wide variety of possible weather conditions and reflective surfaces. In most practical cases, the light loss at 905 nm is smaller because the absorption rate of water at 1550 nm is greater than that at 905 nm.

Choice of Photodetector

Only a small fraction of the photons emitted by the pulse can reach the active area of ​​the photodetector. If the atmospheric attenuation does not vary along the pulse path, the laser beam divergence is negligible, the spot size is smaller than the target, the incident angle is perpendicular to the detector and the reflector is a Lambertian body (reflection in all directions), then the peak light received power P(R) is:

P0 is the peak optical power of the transmitted laser pulse, ρ is the reflectivity of the target, A0 is the receiver aperture area, η0 is the transmittance of the optical system, and γ is the atmospheric extinction coefficient.

The equation shows that the received power decreases rapidly with increasing distance R. For a reasonable choice of parameters, R = 100 m, the number of returning photons on the active area of ​​the detector is about a few hundred to a few thousand, while the number of emitted photons is usually more than 1012. These echo photons are detected simultaneously with background photons, which do not contain any useful information.

The use of narrowband filters can reduce the background light reaching the detector, but it cannot be reduced to zero. The influence of background light reduces the detection dynamic range and increases the noise (background photon shooting noise). It is worth noting that under typical conditions, the ground solar irradiance at 1550 nm is less than 905 nm.

Schematic diagram of the basic principle of time-of-flight (ToF) lidar

Creating a complete 3D map in a 360°×20° area around a car requires scanning with a single laser beam after grating splitting, or scanning with multiple laser beams, or covering the entire required range with a beam and collecting the returned point cloud data. The former is called scanning lidar, and the latter is called flash lidar.

There are several ways to scan lidar. The first way, taking Velodyne (San Jose, CA) as an example, is to install a lidar platform on the top, which rotates at a speed of 300-900 rpm and emits 64 pulses of 905 nm laser. Each beam has a corresponding avalanche photodiode (APD) detector. Another similar method is to use a rotating multi-faceted mirror, each facet tilted at a slightly different angle, so as to guide the reflection of a single pulse beam at different azimuths and oblique angles. The mechanical moving parts in both designs are at risk of failure when the external driving environment is harsh.

Hamamatsu's new 100-meter-class autonomous driving lidar detector

16ch Silicon APD S14137-01CR

The second, more compact scanning LiDAR approach uses a tiny micro-electromechanical system (MEMS) galvanometer to electrically guide one or more beams of light in two dimensions. While there are still technically moving parts (the oscillating mirror), the amplitude of the oscillation is small and the frequency is high enough to prevent mechanical resonance between the MEMS galvanometer and the car. However, the geometric size of the galvanometer limits its oscillation amplitude, which makes the field of view limited - a disadvantage of the MEMS approach. Nevertheless, this method has attracted attention due to its low cost and high feasibility.

Hamamatsu's latest MEMS Mirror products

Just exhibited at the Munich Shanghai Expo

Optical phased array (OPA) technology, a third competing LiDAR technology, is becoming increasingly popular for its reliable "fixed part" design. It consists of an array of optical antennas illuminated by coherent light. Beam steering is achieved by independently controlling the phase and amplitude of each unit's light, so that interference produces the desired illumination direction in the far field, achieving a change from a single beam to multiple beams. Unfortunately, light losses limit the range of available OPA components.

Flash LiDAR floods the target scene with light, and the illuminated area matches the detector's field of view. The detector is an array of APDs in the optical focal plane of the detector. Each APD independently measures the ToF of the target features in its image. This is a true "no moving parts" approach, where the tangential (vertical, horizontal) resolution is limited by the 2D detector pixel size.

However, the main disadvantage of flash lidar is the number of return photons: once the distance exceeds tens of meters, the number of returned light is too small to make reliable detection. If instead of directly covering the entire detection environment with light, a structured light form (such as a dot matrix form) is used, and a certain tangent resolution is sacrificed, the return light intensity can be increased. In addition, vertical cavity surface emitting lasers (VCSELs) make it possible to emit thousands of beams simultaneously in different directions.

Comparison of Hamamatsu's optical semiconductor detectors that can be used for LiDAR

Report: Introduction to core semiconductor devices for autonomous driving Lidar

Getting rid of the limitations of ToF

ToF lidar is susceptible to noise due to its weak echo pulse and wide bandwidth of the detection electronics design, while threshold triggering will produce measurement errors of Δt. Therefore, frequency modulated continuous wave (FMCW) lidar is a very interesting alternative.

In FMCW radar or chirp modulation radar, the antenna continuously transmits a radio wave whose frequency is modulated. For example, it increases linearly from ƒ0 to ƒmax over time T, and then decreases linearly from ƒmax to ƒ0 over time T. If the wave is reflected back to the point of emission from a moving object at a certain distance, its instantaneous frequency will be different from the radio wave emitted at that moment. This difference is caused by two factors: the distance to the object and its relative radial velocity. The frequency difference can be measured electronically, and the distance and velocity of the object can be calculated at the same time (see figure below).

In a CHIRP radar, by electronically measuring fB1 and fB2, the range to a reflecting target and its radial velocity can be determined.

Inspired by chirp radar, FMCW lidar can be obtained in different ways. In the simplest design, one can chirp the intensity of the light illuminating the target. This frequency is affected by the same laws as the carrier frequency of the FMCW radar (such as the Doppler effect). The returning light is detected by the photodetector and the modulation frequency is recovered. The output is amplified and mixed with its own oscillation frequency to allow the frequency shift to be measured, and from this the distance of the target and its speed can be calculated.

However, FMCW lidar has certain limitations. Compared with ToF lidar, it requires more computing power and is therefore slower in generating a full 3D surround map. Moreover, the measurement accuracy is very sensitive to the degree of linearity of the chirp modulation.

While designing a fully functional lidar system is challenging, none of these challenges are insurmountable, and as research continues, we get closer to a time when most cars will be fully autonomous after production ends.