Python implementation of VDXP (Visual Data eXchange Protocol) — unidirectional file transfer over HDMI.
vdxpy encodes files as visual cell patterns on screen, captures them via HDMI + USB capture device, and decodes them on the receiving side. No network connection. No USB drives. Every transfer is SHA256-verified.
[Sender PC] → Screen displays cell pattern → HDMI output →
→ [USB Capture Device] → USB →
[Receiver PC] → Decode → File restored (SHA256 match ✓)
- Python 3.10+
- USB HDMI capture device (tested: USB-CVHDUVC2, ~¥15,000)
- Two PCs (or one PC with HDMI loopback)
pip install -r requirements.txt# On sender PC (displays pattern on screen)
python src/sender.py --file secret.pdf --profile cvhduvc2
# On receiver PC (captures via USB device)
python src/receiver.py --profile cvhduvc2 --output received.pdfsha256sum secret.pdf received.pdf
# Both hashes match ✓vdxpy encodes data as visual cell patterns displayed on screen:
- Sender splits file into chunks, applies Reed-Solomon error correction, and renders each chunk as a grid of colored/grayscale cells on the HDMI output
- Capture device digitizes the HDMI signal via USB
- Receiver decodes cell values, corrects errors via RS, and reassembles the file
- BGR Cube Vertex 8-Color Palette — Maximizes chroma distance to survive MJPEG compression artifacts
- Y-only Grayscale (Y16/Y32/Y64) — Eliminates chroma subsampling interference entirely
- Fingerprint Skip — Identifies duplicate frames in ~0.1ms using 32-cell hash
- 3-Stage Decode Filter — Fingerprint → Header-only → Full decode, minimizing CPU load
- Profile System — Profiles optimized for different capture devices and environments
| Configuration | Throughput | Hardware |
|---|---|---|
| Free (MJPEG, 8-color, cs8) | ~100 KB/s | CVHDUVC2 or compatible USB2.0 MJPEG capture |
| With Accelerated Engine (YUY2, Y32, cs1) | ~15,000 KB/s | USB3.0 YUY2 device + Accel Engine |
All benchmarks SHA256-verified with random data payloads (5 MB -- 500 MB).
Have different hardware? Use the diagnostic tools to optimize:
# Measure color margins and ECC utilization on your device
python tools/diagnose.py --profile cvhduvc2 --duration 30
# Capture a single frame for visual inspection
python tools/snap_frame.py --profile cvhduvc2
# Probe actual color distribution from your capture device
python tools/probe_colors.py --profile cvhduvc2See the Tuning Guide (included with Accelerated Engine) for detailed optimization instructions.
| Solution | Throughput | Cost | Type |
|---|---|---|---|
| vdxpy + Accel | 15.1 MB/s | Contact us | Software + USB capture |
| vdxpy (Free) | 100 KB/s | Free + device | OSS + USB capture |
| libcimbar | 106 KB/s | Free | Camera-based OSS |
| TGXf | ~4 KB/s | Free | QR-code stream |
| Fiber optic data diode | 1--100 Gbps | $5,000--$100,000+ | Dedicated hardware |
- Defense / Government — Transfer data across isolated network segments
- Industrial Control (OT/ICS) — Export SCADA data from air-gapped networks
- Financial — Move data across regulatory network boundaries
The optional Accelerated Engine achieves up to ~150x higher throughput (with USB3.0 YUY2 hardware) through:
- Custom Cython Reed-Solomon decoder (Berlekamp-Massey + Chien + Forney)
- OpenMP parallel block decoding (up to 90.5x faster than standard RS)
- Optimized cell sampling with zero-copy memory access
The Accelerated Engine is available for purchase bundled with a verified USB capture device (Japan only). Contact us for details: contact@blackocean.tech
Bug reports and feature requests are welcome via Issues. Pull requests are not actively reviewed at this time.
MIT License. See LICENSE for details.
All benchmarks performed with SHA256-verified random data. No transfer is reported as successful unless the output hash matches the input hash exactly.