A public research hub for learned communication: neural joint source-channel transceivers that keep delivering as the channel degrades, measured honestly against the best digital baselines on contested, fading and deep-space channels.
Try it: https://suncal.github.io/latent-comms/
- Latent Radio: the in-browser neural transceiver, +4.7 dB over a Shannon-separation baseline at low SNR
- Rosetta: three independently trained radios made interoperable through a shared reference standard and frozen-base adapters
- Contested Channel v1–v4: burst jamming, Rayleigh/Rician fading, adversarial follower-jammers, a hybrid digital-analog scheme, and a feedback-free regime where rate-adaptive digital delivers nothing
- Sonic / Sigil / Aether / LoRa: the same codec carried over sound, drawn as a fading glyph, invented by two agents, and stretched past LoRa's range cliff
- Honest framing throughout: where it wins, where it loses, and why
The hub is static; open index.html. Each experiment links to its own report.
If this is useful to you, a ⭐ on the repo helps other people find it. Issues and pull requests are welcome.
Built by Priyankar "Sunny" Chakraborty · everbuiltstudio.com