A new wireless design lets a single radio signal both talk and see.
Researchers built SemISAC, a system that merges semantic communication with semantic sensing in one dual-function waveform, rather than treating them as separate jobs. They tested it in a vehicle scenario: cars convert road-scene camera images into compressed "semantic symbols" using a deep learning encoder, then transmit them over an OFDM grid alongside pilot signals used for channel estimation and sensing reflections. The pilot layout adjusts itself based on current channel conditions, shifting resources between communication and sensing as needed. On the receiving end, deep learning models rebuild the road segmentation and use reflected signals to classify nearby objects and estimate how far away they are.
This targets a real tension in 6G research: cramming both rich sensing and rich communication into the same limited spectrum usually means compromising one for the other. SemISAC's simulation results hold segmentation accuracy close to a communication-only system while beating conventional and prior semantic baselines at object recognition and range estimation, suggesting the tradeoff is smaller than expected.
It's still simulation results, not a chip on a bench, so treat "outperforms conventional and semantic baselines" as a promising benchmark rather than a working car doing 100 mph.