The day in conversation for 2026-11-08

discussion

Transcript

The discussion: Rosa: The angle estimation work seems promising because maximizing Fisher information under a rate constraint leads to waterfilling, which solves the convex problem theoretically, but I wonder how much of that theoretical benefit translates when we have real-time constraints in true-time-delay systems. Taro: That’s where the complexity spikes for me; if the system relies on precise time synchronization for those angle estimates, does the inherent noise in those delay measurements negate the convexity advantage you're mentioning? I’ve been thinking about how core scheduling, like that LDPC decoder work, tackles latency by essentially managing resource contention at a much lower level than solving continuous optimization problems. Dev: I see that as a different kind of trade-off. In the ML receiver papers, combining machine learning with CFAR for preamble detection means you are trading off computational load for false alarm rates; what specific metric did they use to define an acceptable false alarm threshold when integrating that learned model into a classical detector?

More episodes

← Home