Lattice LF-EVDK1-EVN

This document describes the design and setup of the Lattice Embedded Vision Development Kit (EVDK), used to demonstrate dual CSI-2 camera to High-Definition Multimedia Interface (HDMI) bridging functionality using a CrossLink FPGA, ECP5 FPGA, and SiI1136 transmitter. The Lattice EVDK is designed as a stackable modular architecture measuring 80mm × 80mm, consisting of three boards: the CrossLink Video Interface Platform (VIP) input bridge board, the ECP5 VIP processor board, and the HDMI VIP output bridge board. The dual-camera Mobile Industry Processor Interface (MIPI) CSI-2 to HDMI demonstration uses Sony IMX214 cameras, outputting 1080p video over four MIPI data lanes, each running at 371.25Mbps. The CrossLink VIP input bridge board receives MIPI video streams from the onboard camera sensors and extracts video pixels, merging the two cameras' video pixels side by side; the combined image data is then transmitted via board-to-board connectors as a parallel CMOS interface to the ECP5 on the ECP5 video processor board. The ECP5 FPGA processes the combined sensor image and sends the processed parallel image data through board-to-board connectors to the SiI1136 HDMI transmitter on the HDMI VIP output bridge board, which transmits the video data to a 1080p display via HDMI. The dual-camera to parallel design receives serial, source-synchronous MIPI data from two MIPI CSI-2 cameras, re-serializes the serial data and extracts control signals; the byte data is sent to a byte-to-pixel module for conversion to RAW10 data, the two RAW data streams are fed into image merging logic, and the combined data is sent to the ECP5 board. The onboard CSI-2 cameras are configured via the I²C master interface on the ECP5 VIP processor board. The ECP5 FPGA receives RAW10 data from the CrossLink, performs basic image processing, and sends it to the HDMI board. Lattice programmable image processing modules improve sensor image quality in the following ways: the auto-brightness module adjusts the intensity of incoming sensor data; the de-Bayer module converts RAW10 Bayer data into separate red, green, and blue pixels per clock cycle; the color space converter matrix corrects mismatches between image sensor colors and real-world colors, with gain and offset controls for each color as well as cross-color influence adjustments; and gamma correction applies pre-distortion correction to the video.