Orcas Airband streams, transcribes entire airband
Built by Mark Maunder, Orcas Airband pairs an Ettus B210 SDR with dual RTX 4090 GPUs to capture and channelize over 700 VHF aviation frequencies simultaneously. The platform demodulates active transmissions and runs custom speech models alongside downstream LLMs to generate real-time transcripts and airspace situational awareness.
Orcas Airband is a masterclass in edge DSP and applied AI, demonstrating how consumer GPUs can replace entire racks of dedicated radio receivers. By pairing wideband polyphase channelization with local speech models and LLMs, it turns raw RF spectrum into structured, real-time intelligence.
- –Custom CUDA polyphase channelizer splits ~19 MHz of raw VHF spectrum into 700+ parallel channels on-device, bypassing costly multi-receiver hardware arrays.
- –On-premise speech pipeline filters non-voice data transmissions before feeding audio through a dedicated aviation voice model tuned for noisy AM cockpit chatter.
- –Downstream LLM parsing extracts flight status, callsigns, and airspace anomalies to power an automated traffic board and emergency alerting system.
- –Real-world utility shone during a major regional ground stop, offering pilots instant ground-truth awareness when official communications lagged.
- –Creator plans to open-source the pipeline, laying the groundwork for a decentralized, community-hosted ATC monitoring network.
DISCOVERED
54m ago
2026-11-14
PUBLISHED
3h ago
2026-11-13
RELEVANCE
AUTHOR
mmaunder