Design of a data aggregation circuit for Autonomous Driving LiDAR sensors

The automotive industry is facing an unprecedented era of change, revolutionizing the entire human driving experience, making it safer and more comfortable. The growth of Advanced Driver Assistance Systems (ADAS) is enabled by the data fusion collected from several types of sensors (camera, radar, ultrasonic). The gap between today’s ADAS and its higher levels can be filled with the introduction of LiDAR sensors. Light Detection and Ranging technology creates a 3D rich point cloud of the surrounding environment by using laser beams and time-of-flight distance measurement, allowing accurate real time objects detection, even at long distances and poor light or weather conditions when cameras lose their reliability. To cover the field of view required for ADAS functionalities from Level 3 and up, a LiDAR platform needs at least three optoelectronic modules (each equipped with a laser transmitter and a laser receiver) sending MIPI CSI-2 raw data streams to the CPU, which must process them to create a rich point cloud, decisive element for the autonomous driving strategy. However, Automotive CPUs usually only have up to two CSI-2 ports. MIPI Camera Serial Interface 2, with its high bandwidth D-PHY physical layer, is widely adopted in the automotive industry as interface for cameras, radar and LiDAR sensors, supporting a wide range of applications, raw data format, resolutions and frame rates. Starting from these premises, the application deepened in this Thesis work provides a cutting-edge solution for a high-performance LiDAR system developed by Marelli Automotive Lighting: the evaluation, the design and the implementation of a Proof Of Concept that aggregates data coming from two optoelectronic modules, to provide a unique output data stream to the CPU. Specifically, the main goal of this Thesis work is the development of a data aggregation circuit based on FPGA, exploiting one of the Long Packet fields named Virtual Channel, whose aim is to provide separate channels for different data flows interleaved in the same output data stream. The solution analyzed refers to an FPGA of the CrossLink Automotive family offered by Lattice Semiconductor. It supports a wide variety of protocols, pays special attention to energy efficiency and makes available all the features necessary for design purposes, as the entire blocks that implement MIPI protocol. This application requires flexibility, verification, and ability to iterate quickly: features offered by the reference software environment Lattice Diamond. It provides a complete set of tools for design entry, implementation, synthesis, analysis, programming and simulation activities. At the end of the implementation the project tested on the evaluation board provides positive results. The input MIPI CSI-2 raw data collected from the two modules are correctly merged in a single MIPI CSI-2 output data stream to be sent to the Automotive System on Chip, as demonstrated by the MIPI protocol decoding performed by the oscilloscope. Overall, this FPGA-based implementation proves to be extremely suitable and convenient for the project purposes. Indeed, comparing this technique with other hardware solutions previously investigated (ASIC, SerDes), many advantages in terms of flexibility, power consumption, costs, number of components involved (resulting in a reduced PCB area) make it an optimal choice for any possible future development.