Man or Machine?: Developing a "Turing Test" for Radio Intelligence
summary
Builds a neural‑network PHY called deepmod and runs it on real hardware (GNU Radio with a B210), first over a speaker-microphone link then powerline and 900 MHz RF, showing the model invents constellations (e.g. 16QAM → an 8×2 layout) and relearns through jamming and mobility after a few hundred training epochs, cooperates through a learned relay/cooperative-MIMO behavior, imposes spectral envelopes and trades samples for lower BER — concrete demos and measurable training times, not just theory.
AI-written summary · about
chapters (21)
- 0:00Intro
- 1:23Compliance
- 2:11What is Radio Intelligence
- 3:18Maxwells Demon
- 4:46Radio Intelligence
- 5:45Deep Mod
- 6:51Power Line
- 7:30RF Channel
- 8:08Deepmod
- 10:52The Turing Test
- 12:18The Captcha Test
- 12:46The Dr Anderson Test
- 16:02Observations
- 18:32The Machine
- 21:12Deep Mod Test
- 23:06Relay Test
- 24:09Cooperation Test
- 26:12Mobility
- 27:16Envelope
- 28:33Spectral Efficiency
- 30:37Conclusion
related talks
24mAn Open Channel Identifier using GNU Radio
13mOne Protocol to Rule Them All: Wireless Network-on-Chip using Deep Reinforcement Learning
29mEngineering Machine Learning Technologies
15mWelcome to the IoT Casino
14mBrain-on-Switch: Towards Advanced Intelligent Network Data Plane via NN-Driven Traffic...
23mDeep learning inference in GNU Radio with ONNX
58mVoice Control Made Simple – Panel Discussion
49mFunctional Electromagnetism