Skip to content
@IONIS-AI

IONIS-AI

HF propagation prediction using physics-constrained neural networks trained on 13B+ amateur radio observations

IONIS-AI — Ionospheric Neural Inference System

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 13+ billion amateur radio spots (WSPR, RBN, Contest, PSK Reporter). It predicts signal-to-noise ratio for any HF path on Earth using real-time solar conditions.

What It Does

  • 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

Packages

Package Language Description
ionis-apps Go Data ingesters (WSPR, RBN, Contest, PSK Reporter, Solar) with idempotent watermark tracking
ionis-core SQL/Shell 34 ClickHouse DDL schemas, 13 population scripts, base configuration
ionis-cuda C++/CUDA GPU-accelerated signature embedding engine
ionis-training Python PyTorch model training, sensitivity analysis, version lifecycle
ionis-validate Python Model validation CLI (PyPI)
ionis-tools Go Prediction CLI and REST API
ionis-docs MkDocs Documentation site

Dataset

Source Volume Years
WSPR 10.8B spots 2008–2025
Reverse Beacon Network 2.18B spots 2009–2025
CQ Contest Logs 195M QSOs 2005–2025
PSK Reporter ~26M spots/day (live) 2026+

Model

IONIS V20 (Production) — 203K parameter PyTorch model with physics-constrained solar and geomagnetic sidecars.

Metric IONIS V20 VOACAP
Pearson correlation +0.49 +0.02
Live recall (PSK Reporter) 84%

Data Pipeline

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

Install

# Rocky Linux 9 / RHEL 9 (via COPR)
dnf copr enable ki7mt/ionis-ai
dnf install ionis-apps ionis-core

# Ubuntu 24.04+ (via PPA)
sudo add-apt-repository ppa:ki7mt/ionis-ai
sudo apt install ionis-core ionis-training

# Model validation (any platform, via PyPI)
pip install ionis-validate

License

GPLv3 — See individual repositories for details.

Popular repositories Loading

  1. ionis-docs ionis-docs Public

    Documentation site for the IONIS (Ionospheric Neural Inference System) project.

    HTML 1

  2. .github .github Public

    Advanced Ionospheric Neural Inference System (IONIS). Engineering the Digital Twin of the Ionosphere using 10.8B+ data points from Solar Cycle 25.

  3. ionis-apps ionis-apps Public

    Core ingestion and data-milling utilities for the IONIS (Ionospheric Neural Inference System) project.

    Go

  4. ionis-core ionis-core Public

    ClickHouse schemas and environment setup for the IONIS (Ionospheric Neural Inference System) project.

    Shell

  5. ionis-cuda ionis-cuda Public

    CUDA signature embedding engine for the IONIS (Ionospheric Neural Inference System) project.

    Cuda

  6. ionis-training ionis-training Public

    PyTorch training and validation for the IONIS (Ionospheric Neural Inference System) propagation model.

    Python

Repositories

Showing 9 of 9 repositories

People

This organization has no public members. You must be a member to see who’s a part of this organization.

Top languages

Loading…

Most used topics

Loading…