Architecture

Pipeline Overview

1. The Encoder

1. Input Stream: Reads the input file in 1MB chunks.
2. Reed-Solomon FEC: Splits data into 10 data shards and calculates 2 parity shards (configurable). This allows recovering the file even if 20% of the data in a frame is lost.
3. Block Scaling: Each logical bit is expanded into a block of pixels (e.g., 4x4). This makes the signal robust against video compression algorithms (H.264/VP9) which blur high-frequency noise.
4. FFmpeg Pipe: The raw pixel frames are piped into ffmpeg via stdin to generate the video container (mkv or mp4).

2. The Decoder

1. FFmpeg Pipe: Spawns ffmpeg to read the video file and output raw RGB frames.
2. Bit Extraction: Scans the center of each pixel block to determine if it is a 0 or 1, effectively downsampling the image.
3. Header Parsing: The first frame(s) contain a JSON header with file metadata (Filename, Size, Hash).
4. Reconstruction: Uses the Reed-Solomon engine to rebuild missing or corrupted shards.
5. Output: Writes the reconstructed bytes to the output file.

Zero-Copy Design

The tool is designed to handle files larger than available RAM. It uses streaming iterators and buffers only a few frames at a time.

Additional Resources

• Getting Started Guide
• Recommended Platforms