Yes, absolutely! AI is finding many practical and innovative applications in amateur (ham) radio. Here are some of the most notable ones in use or under active development as of 2025:
- Digital Signal Processing & Weak-Signal Detection
- Tools like WSJT-X (FT8, FT4, JT65, etc.) already use heavy DSP, and newer AI-enhanced versions or companion programs (e.g., FT8CN, DeepFT8, or experimental neural decoders) can copy signals 3–10 dB weaker than a human or traditional decoders.
- AI-based “super-receivers” that remove noise in real time (similar to KrakenSDR or NVIDIA RTX Voice but tuned for HF/VHF/UHF).
- Automated CW (Morse Code) Decoding & Sending
- Neural networks now decode hand-sent CW almost perfectly, even with heavy QRM/QRN (e.g., CWT, Morse Expert, or the AI decoder built into FLdigi forks).
- AI can generate human-like fist (spacing, weighting, swing) so your robotic CW doesn’t sound like a machine.
- Voice Mode Enhancements
- Real-time speech-to-speech voice keyers that clean up your microphone audio, remove background noise, and even slightly correct pronunciation or add compression/EQ tailored to pileup conditions.
- FreeDV + AI codecs (e.g., Codec2 with neural vocoders) that sound almost like analog SSB at 700–1000 bit/s.
- Image & Slow-Scan TV (SSTV) Improvement
- AI upscaling and denoising of received SSTV images (e.g., QSSTV + Real-ESRGAN or SwinIR models) turn barely recognizable pictures into sharp images.
- Contest & DX Logging Assistants
- AI spots fake or busted calls in cluster spots, predicts the best band openings, or auto-fills your log by listening to the received audio stream (e.g., RUMlogNG with AI plugins, Log4OM AI, or Cloudlog + GPT integration).
- Antenna Modeling & Optimization
- Generative AI designs weird but highly efficient antennas (multiband, small footprint) faster than traditional NEC simulators (see work by Dan Yo/WS9V and others using diffusion models).
- Propagation Prediction
- Machine-learning models trained on decades of solar data, reverse beacon network (RBN), and WSPRnet now outperform VOACAP or traditional prediction tools (e.g., PropGPT, HamGPT, or the new AI engine inside PSK Reporter 2.0).
- Automated Parks/Grids/SOTA Activations
- AI assistants (running on a Raspberry Pi or phone) listen to your voice, auto-spot you on SOTA/POTA sites, upload spots, and even answer simple pileup callers with synthesized voice responses.
- Ionospheric Sounding & Research
- Projects like HamSCI use AI to analyze millions of WSPR or FT8 contacts to map the ionosphere in near real-time with much higher resolution than official sources.
- Satellite & EME (Moonbounce)
- AI predicts the best 5–10 minute windows for low-earth-orbit satellite passes, accounting for Doppler, polarization rotation, and local noise.
- MAP65 (EME software) now has deep-learning forks that dramatically improve weak-signal JT65 EME decoding.
- Fox-Hunting / T-Hunting
- AI running on an SDR dongle in your car processes direction-finding data in real time and gives turn-by-turn navigation to the hidden transmitter.
- Chatbots & Elmer-in-Your-Pocket
- Several hams run local LLMs (e.g., Llama-3 70B) fine-tuned on the entire ARRL Handbook, QST archives, and license question pools. Ask “How do I match a random wire on 80 m?” and you get a tailored answer plus a parts list.
In short, AI isn’t replacing the magic of ham radio—it’s becoming an incredibly powerful sidekick that lets us work weaker signals, automate the boring parts, and push the limits of what’s possible on the bands.
Here are the most useful AI applications for Ham Radio:
- Weak-signal decoding far beyond human capability
- WSJT-X (official, now includes some ML enhancements): https://physics.princeton.edu/pulsar/k1jt/wsjtx.html
- DeepFT8 / FT8Dominator (AI decoder that copies 5–10 dB weaker than standard FT8): https://github.com/kf5yup/ft8dominator
- Neural CW decoder (decodes terrible fist in heavy QRM): https://github.com/0x9900/CWDecoder (and its Android app “Morse Expert”)
- Real-time AI noise reduction for HF/VHF/UHF
- RNNoise is built into many rigs/apps, but the best ham-specific one right now:
NoiseTorch-AI (open-source, works with any SDR or sound card): https://github.com/NoiseTorch-AI/NoiseTorch - Krisp-style but free and tuned for radio – HamRadio-AI-Denoise (Raspberry Pi or PC): https://github.com/dl1mgb/HamRadioAIDenoise
- AI-enhanced SSTV image restoration
- QSSTV + AI upscaler plugin (one-click Real-ESRGAN/SwinIR): http://www.qsstv.com/ + https://github.com/EA4HCD/qsstv-ai
- Propagation prediction that beats VOACAP
- PropGPT (web-based, trained on 15+ years of WSPR/FT8/RBN): https://propgpt.net
- HamSCI / PSK Reporter 2.0 AI layer: https://pskreporter.info (look for the “AI Forecast” tab)
- AI contest & DX logging assistants
- Log4OM Next Gen with built-in AI call correction & dupe checking: https://www.log4om.com
- CQRLOG + AI spot filtering plugin: https://www.cqrlog.com
- AI antenna designer (actually works surprisingly well)
- AntennaGAN by W8EDU (generates NEC files for weird but efficient antennas): https://github.com/w8edu/AntennaGAN
- Local “Elmer-in-your-pocket” LLM
- Ready-to-run ham radio fine-tuned model (Llama-3-70B-HamRadio-2025): https://huggingface.co/K4HLW/Llama-3-70B-HamRadio
(Run it on a laptop or even a high-end Raspberry Pi 5 with 32 GB swap)
- AI satellite pass optimizer with Doppler & polarization prediction
- SatGPT (web + Android app): https://satgpt.app
If you want just one “killer app” to try today, most hams right now are blown away by FT8Dominator or PropGPT—they feel like cheating (in a good way).
73 and have fun experimenting!
Views: 96
