The logs were speaking for decades, but nobody was listening. Now we're listening.
IONIS is an open-source HF radio propagation prediction system trained on 14+ billion amateur radio observations (WSPR, RBN, Contest, PSK Reporter). It predicts signal-to-noise ratio for any HF path on Earth using real-time solar conditions.
- Predicts HF propagation (SNR) between any two points, any band, any time
- Trained on the largest curated amateur radio propagation dataset that we know of
- Uses real-time solar flux (SFI) and geomagnetic (Kp) indices
- Validated against VOACAP: Pearson +0.49 vs +0.02, 84% recall on independent live data
| Package | Language | Description |
|---|---|---|
| ionis-apps | Go | Data ingesters (WSPR, RBN, Contest, PSK Reporter, Solar) with idempotent watermark tracking |
| ionis-core | SQL/Shell | 36 ClickHouse DDL schemas, 15 population scripts, base configuration |
| ionis-cuda | C++/CUDA | GPU-accelerated signature embedding engine |
| ionis-training | Python | PyTorch model training, sensitivity analysis, version lifecycle |
| ionis-mcp | Python | MCP server for AI-assisted propagation analytics (PyPI) |
| ionis-jupyter | Python/Jupyter | Interactive notebooks for propagation research — 10 starter notebooks |
| ionis-validate | Python | Model validation CLI (PyPI) |
| ionis-hamstats | Python/MkDocs | ham-stats.com — live HF propagation reports |
| ionis-docs | MkDocs | Documentation site |
175M+ aggregated propagation signatures derived from 14 billion observations, distributed as SQLite files on SourceForge.
| Source | Signatures | Raw Observations | Years |
|---|---|---|---|
| WSPR | 93.6M | 10.9B beacon spots | 2008–2026 |
| Reverse Beacon Network | 67.3M | 2.26B CW/RTTY spots | 2009–2026 |
| CQ Contest Logs | 5.7M | 234M SSB/RTTY QSOs | 2005–2025 |
| DXpeditions | 260K | 3.9M rare-grid paths | 2009–2025 |
| PSK Reporter | 8.4M | 514M+ FT8/WSPR spots | 2026+ |
| Solar Indices | — | 77K daily/3-hour records | 2000–2026 |
IONIS V22-gamma + PhysicsOverrideLayer (Production) — 205K parameter PyTorch model with physics-constrained solar and geomagnetic sidecars, plus a deterministic post-inference clamp for high-band night closure.
| Metric | IONIS V22-gamma | VOACAP |
|---|---|---|
| Pearson correlation | +0.49 | +0.02 |
| RMSE | 0.821σ | — |
| KI7MT operator tests | 17/17 | — |
| TST-900 band x time | 9/11 | — |
Every data source follows a paired download → ingest pattern with idempotent
watermark tracking. Downloaders write to disk, ingesters load into ClickHouse.
All ingesters support --full, --prime, and --dry-run modes.
| Source | Downloader | Ingester | Schedule |
|---|---|---|---|
| WSPR | wspr-download | wspr-turbo | Daily |
| RBN | rbn-download | rbn-ingest | Daily |
| Contest | contest-download | contest-ingest | Weekly |
| PSK Reporter | pskr-collector | pskr-ingest | Hourly |
| Solar | solar-download | solar-ingest | 15 min / 6 hr |
| DSCOVR L1 | dscovr-ingest | dscovr-ingest | 15 min |
# AI-assisted propagation analytics (any platform, via PyPI)
pip install ionis-mcp
ionis-download --bundle minimal # ~430 MB — contest + solar + grids# Model validation (any platform, via PyPI)
pip install ionis-validate
# Rocky Linux 9 / RHEL 9 (via COPR)
dnf copr enable ki7mt/ionis-ai
dnf install ionis-apps ionis-coreGPLv3 — See individual repositories for details.