Chris Dick
Chris Dick
Senior Distinguished Engineer, NVIDIA
Wednesday, 9 September 2026, 9:45 – 10:30
Title: Split Decision: ISAC in the O-RAN Stack
Abstract: A cellular network is already a coherent, wideband, phased-array sensor. Turning it into one is not the hard part. The difficulty is that the O-RAN 7.2x functional split draws a line through the receive chain, so a sensing function inherits a different signal depending on where it sits. This keynote examines the options that result, and how to choose among them. Three candidate tap points (the raw converter output, the post-low-PHY resource grid, and the compressed fronthaul payload) each offer a different balance of signal fidelity, how much of the network a function can see, and how much processing it can afford, on terms set by fronthaul economics rather than by software. Three constraints decide the choice. Dynamic range comes first: where the direct path, or a transmitter’s own coupling, exceeds the converter’s usable range, that excess has to be suppressed ahead of the converter. The deeper suppression that follows is digital, and can be performed blind, using the array as its own reference rather than a dedicated reference channel. Latency comes second: the achievable sense-to-act horizon depends on where the sensing sits relative to the loop it drives. The observable comes third: rotor micro-Doppler identifies the airframe, but its finest structure lives on faint spectral lines that compression thins first, which makes the demanding classification tasks the operating-point-sensitive ones. The conclusion is a map rather than a verdict. Detection, ranging and coarse Doppler are largely insensitive to the tap point. Fidelity-bound and reflex-latency functions do better near the converter. Multi-RU fusion, multistatic geometry, cross-cell tracking and learned classification belong where the aggregated data and the processing capacity are.
