Perception-Aware Closed-Loop Scheduling for Capacity-Constrained Sensor Networks: A Cognitive Radar Case Study
Abstract
Capacity-constrained sensor networks – of which cognitive radar networks are a representative instance – must select, under a strict per-cycle bandwidth limit, which sensing nodes report updates to a fusion center. Age-of-Information (AoI) scheduling addresses this by prioritizing stale updates, but assumes that all sensor measurements are equally reliable. We propose weighting each node’s scheduling priority by a detection-confidence score: a logistic function of the local signal-to-noise ratio (SNR), calibrated against the node’s empirical detection probability, rather than output by a trained classifier. This weighting is applied to the existing AoI priority without altering the target motion model, Kalman/IMM (Interacting Multiple Model) filter, or AoI update rule. We formulate this as a centralized Top- $C$ selection rule and extend it, in principle, to a distributed Age of Incorrect Information (AoII)-based triggering policy, whose empirical evaluation is left for future work. Three scheduling conditions were compared under identical capacity ( $C=2$ , 50 nodes, ~12 targets): AoI-based (priority from AoI alone); PA Open-loop (AoI priority weighted by detection confidence, no feedback); and PA Closed-loop, where a node’s future confidence rises after it is selected, emulating real sensing improvements such as waveform refinement. PA Open-loop cuts the tracking root-mean-square error (RMSE) by 32.5% relative to AoI-based (7.45 to 5.03 m); the PA Closed-loop achieves a 57.3% reduction (3.18 m), improves the average detection confidence by 115.3%, and raises effective uplink-capacity utilization by 64%. Results use a calibrated SNR-based confidence proxy at a single capacity setting.