An Ethernet-Integrated Accelerator for On-the-Fly Spherical-to-Cartesian LiDAR Coordinates Conversion

This paper presents an open-source hardware accelerator that performs on-the-fly spherical-to-cartesian conversion of VLP-16 LiDAR packets directly inside the front-end of a bufferless Ethernet IP in the Carfield SoC, using INT16 precision. By processing the incoming packet stream without storing raw spherical data, the proposed design reduces front-end latency and memory overhead necessary for the coordinate conversion in software, while preserving accuracy. The accelerator achieves up to a 5.6 × speed-up and 82 % latency reduction over software, and keeps MAE/RMSE below 6 mm across the tested RPM ranges. A bare-metal INT16 RANSAC validation on the converted data further shows sphere-fitting relative errors below 3 %, demonstrating suitability for point-cloud processing.