Control Signaling Constrained Topology Reconfiguration for Clustered Wireless Sensor Networks
In clustered wireless sensor networks (WSNs), battery-powered sensor nodes (SNs) deliver sensing packets to a base station (BS) through cluster heads (CHs), local aggregation, and multihop forwarding. Once a clustered topology is formed, residual energy, relay burden, and CH load diverge across nodes. Rebuilding the topology can rebalance this burden, but each rebuild also consumes control signaling energy for status reporting and configuration dissemination. This paper formulates clustered-WSN topology reconfiguration as a control-signaling-constrained online decision problem. The clustering and routing module serves as the topology-construction layer, while a semi-Markov decision process (SMDP) controller decides whether to keep the current topology or rebuild it with a selected cluster ratio. The controller is implemented using proximal policy optimization (PPO) with normalized energy descriptors, topology age, load-dependent imbalance, distance-dependent control overhead, and scalable cluster-ratio actions. Per-node energy conservation is audited so that data energy, control energy, idle energy, residual energy, and terminal clipping remain consistent with the initial network energy budget. In 30 energy-audited runs of 12,000 rounds, the proposed controller obtains first node death (FND) $1187.5\pm 168.3$ , half node death (HND) $1856.7\pm 25.9$ , packet delivery ratio (PDR) 0.9887, control energy 0.840 J, and 3566.7 delivered packets per Joule. Compared with a $T_{p}=10$ periodic baseline, it improves FND by 294.7 rounds and HND by 118.5 rounds, reduces control energy by 0.224 J, and increases delivered packets per Joule by 213.9. The trade-off is a 86.7-round reduction in last node death (LND) and higher delay under the PEGASIS-style chain-routing module. Calibrated threshold rules can achieve longer LND after deployment-specific tuning, whereas the proposed controller provides a threshold-independent online control layer with quantified reliability-energy-delay trade-offs.