Threshold-Aware Link-Graph Compression with Per-Sensor Observed-Packet Coverage Constraints and Temporal Robustness in Urban LoRaWAN Networks

In multi-gateway LoRaWAN deployments, a single uplink packet may be received through several sensor–gateway links, creating link-graph redundancy. We formulate compression as minimum-cardinality selection under aggregate and per-sensor observed-packet coverage constraints. The sensor-separable structure enables an exact dynamic program to certify optima and benchmark a reusable fairness-aware greedy ordering, a threshold-dependent multi-cover greedy, and a coverage-only ablation. Evaluation uses 1,232,813 uplinks observed by at least 1 gateway over 17 calendar months. On this deployment, proposed and multi-cover methods matched the minimum across 1000 threshold pairs. Coverage-only was optimal for 4%, with mean and maximum gaps of 5.84 and 8 links. These methods returned different equal-cardinality subsets for 506 pairs. The proposed subset improved worst-month coverage slack in 437 cases and met both requirements in more months in 53 cases, never in fewer. Retrospective monthly optimization required 9.24–11.59 links on average, versus 19 for an all-month-feasible subset. In a 14-month prospective evaluation, the expanding-window policy achieved pass rates of 78.6%, 57.1%, and 50.0% with three or four reconfigurations of at most two links each, and an unseen sensor caused a common cold-start failure. These results quantify trade-offs among link budget, future-month compliance, and subset stability.