For autonomous vehicles, LiDAR (Light Detection and Ranging) systems mounted on the roof and around the vehicle have become the "eyes" for environmental perception. LiDAR is widely adopted because it can quickly and accurately capture surrounding objects and generate 3D point clouds, providing crucial data for vehicle route planning. The first function of any LiDAR system is to emit and receive light signals. It continuously scans its surroundings through a laser emission module, emitting thousands of laser pulses or continuously modulated light beams. These beams are reflected back when they encounter obstacles and captured by the receiving module. The time difference between these round trips is the core metric for LiDAR ranging—Time of Flight (ToF). Some advanced devices calculate distance and speed by comparing the frequency difference between the emitted and echoed light (FMCW technology). To achieve coverage of the entire surrounding environment, traditional solutions often mount the laser on a high-speed rotating bracket, continuously scanning 360 degrees. Emerging solid-state phased array solutions utilize microelectromechanical systems (MEMS) or optical phased array elements, enabling large-angle scanning without mechanical movement.
The hardware of a lidar system is mainly divided into four parts: transmission, reception, scanning, and processing. The transmission unit requires a laser capable of outputting stable pulses or a linearly modulated light source; the receiver is equipped with a high-sensitivity photodetector to capture the echo signal; lenses and mirrors in the optical system are responsible for focusing and adjusting the direction of the beam; finally, the signal processing unit converts the sampled electrical signals into digital data, which, after filtering, peak detection, and algorithm calculations, is pieced together to form a series of detailed 3D point cloud maps. For autonomous vehicles, these point cloud maps are the foundation for recognizing pedestrians, bicycles, vehicles, curbs, and traffic signs.
LiDAR Composition
When traffic is light, each LiDAR operates in a relatively "clean" environment, and most of the light pulses it emits will hit real obstacles and return, resulting in high-quality data. However, if multiple LiDARs are traveling close together and emitting lasers with similar wavelengths and modulation methods, sometimes a light pulse emitted by one device might happen to fall within the sampling window of another receiver and be mistaken for its own echo, producing false ranging results. To put it more simply, if a pulse emitted by car A happens to fall within the "listening" period of car B, car B will identify the signal emitted by A as a return signal from the road surface or obstacle. A similar situation occurs in FMCW-type LiDARs. When the frequency modulation bandwidth or starting frequency of two devices is close, multiple frequency difference peaks will be generated after mixing, making it difficult for the receiver to distinguish which one is truly targeting its own object.
In scenarios involving urban congestion or waiting in line at traffic lights, this "crosstalk" effect becomes particularly pronounced. For example, when two autonomous vehicles equipped with LiDAR are driving side-by-side, if their scanning angles overlap, the echoes generated by tree trunks in front of vehicle A are likely to be received by vehicle B. This would cause vehicle B to see extra "trees" in its point cloud, affecting the identification and judgment of actual pedestrians or vehicles. Or, while waiting at an intersection, the echoes from a vehicle next to it might mistakenly enter the point cloud of the vehicle itself, causing the system to think someone is crossing the road ahead, thus triggering unnecessary emergency braking. More seriously, in congested tunnels or multi-level parking lots, multi-path echoes can also be superimposed, posing a greater challenge to algorithmic filtering.
Interference between lidar devices can be attributed to overlaps in frequency, time, and space. Laser wavelengths are typically concentrated in the 905nm or 1550nm band. Even with slight variations in factory calibration, the wide bandwidth of the receiver allows for the capture of photons at adjacent wavelengths. If different devices lack precise synchronization in their transmission timing, pulses or frequency-modulated signals can easily collide. Furthermore, if the scanning directions are not aligned, the emitted beams will overlap in space, meaning the beam from one device will be received within the field of view of another.
If a car equipped with LiDAR experiences the aforementioned issues, it could lead to serious consequences. First, the false alarm rate increases, meaning the system might mistake non-existent objects for real ones, resulting in too many "false obstacles." This reduces the autonomous driving system's response speed to real obstacles and could even lead to unnecessary emergency stops or detours. Second, there's the risk of missed detections. Real pedestrians, vehicles, or obstacles can easily be lost in the cluttered point cloud noise, making it difficult for the algorithm to accurately extract them and thus hindering timely avoidance decisions. A deeper concern is that LiDAR data is often fused with multiple sensors, such as cameras and millimeter-wave radar. If the quality of the LiDAR data remains consistently unstable, it will affect the reliability of the entire perception chain.
To address this issue, the industry has launched a multi-pronged approach, addressing both hardware and software challenges. Some manufacturers are attempting frequency domain isolation between different wavelengths or modulation bandwidths to prevent signals from adjacent vehicles from overlapping in frequency. Others are promoting time synchronization solutions based on GNSS or vehicle-to-vehicle (V2V) communication, strictly staggering the transmission time slots of different vehicles to ensure that only a few vehicles in the same area are transmitting at any given time. The emergence of solid-state phased array LiDAR has provided more possibilities for spatial isolation, electronically altering the beam direction to attenuate "intruding" signals from other directions while maintaining high resolution in critical directions.
At the software level, some technologies propose using encoding and matched filtering techniques to add a unique "identity tag" to the light pulses or FM signals of each LiDAR unit. The receiver only decodes signals with its own tag, while other signals with different tags are treated as noise and discarded. This approach is similar to CDMA (Code Division Multiple Access) in the communications field, but achieving real-time encoding and decoding of high-frequency signals in the optical domain poses a significant challenge to processor performance. Furthermore, point cloud post-processing algorithms are becoming more intelligent, using machine learning models to identify potential crosstalk point clouds online, classifying suspicious point clouds as "interference," and correcting them after fusing data from other sensors.
To solve this problem, in addition to improvements to individual vehicles or systems, vehicle-to-vehicle coordination is also crucial. Utilizing high-speed, low-latency C-V2X (Cellular Vehicle-to-Everything) or DSRC (Dedicated Short Range Communication), different vehicles can exchange radar status and time slot arrangements in real time. Once a potential transmission conflict is detected, they can immediately adjust transmission power, change transmission time slots, or adjust scanning angles via network commands. This centralized or distributed resource scheduling allows each vehicle to maintain high-precision awareness of its surroundings while avoiding "laser interference" with other vehicles.
In the future, LiDAR anti-jamming technology may be deeply integrated with the intelligent and integrated design of vehicles. Continuous advancements in photonic integrated chips will make miniaturization and low power consumption at the LiDAR chip level possible, significantly reducing the cost of large-scale vehicle installations. The onboard central processing unit (domain controller) will integrate more powerful AI computing power, capable of fusing, judging, and correcting multi-source data on a millisecond-level timescale, distinguishing its own laser signal from others in real time, and dynamically adjusting transmission parameters. The rise of cloud-based big data platforms will enable the aggregation and analysis of radar "real-world" data from various roads and road conditions, providing continuous feedback for algorithm updates and configuration optimization.
When multiple autonomous vehicles use LiDAR simultaneously, mutual interference is indeed possible, leading to problems such as false alarms, missed detections, and perception distortion. Fortunately, various countermeasures are constantly maturing and being implemented, including frequency and timing isolation, anti-interference methods for encoding and decoding, collaborative scheduling in vehicle-to-everything (V2X) networks, and backend intelligent algorithms. With the continuous improvement of technology and standards, the anti-interference capability of LiDAR will be significantly enhanced, providing autonomous vehicles with a more reliable "third eye" and helping smart transportation truly achieve large-scale commercialization in the future.
