Welcome to Journal of Network Communications and Emerging Technologies (JNCET)

Volume 14, Issue 2, April (2026)                              

S.No	Title & Authors	Full Text
1	MARL-Q-Net: A Cooperative Multi-Agent Q-Learning Framework with LSTM-Based Traffic Scheduling for Energy-Balanced Routing in Large-Scale Wireless Sensor Networks Yogesh Juneja, Rajiv Dahiya Abstract - Ensuring energy longevity in large-scale Wireless Sensor Networks (WSNs) remains a critical challenge as node populations scale into the thousands under non-stationary traffic conditions. This paper presents MARL-Q-Net, a fully distributed cooperative framework combining independent Q-learning agents with an LSTM-based traffic load scheduler to achieve joint energy balance and routing efficiency in WSN deployments of 200 to 2,000 nodes. Unlike centralized deep-learning clustering approaches that predict per-node energy at the base station, MARL-Q-Net embeds routing intelligence directly within each sensor node. Every node maintains a Q-table over a five-dimensional state space - covering residual energy, sink distance, neighbor degree, queue depth, and an LSTM-predicted one-step-ahead traffic load - and selects routing and cluster-head actions through a cooperative epsilon-greedy policy reinforced by periodic Q-gradient exchange with one-hop neighbors. The LSTM scheduler is trained on the Intel Berkeley Research Laboratory real-sensor dataset. NS-3 experiments with IEEE 802.15.4z UWB MAC across 200 to 2,000 nodes over a 1,000x1,000 m2 terrain and 250 rounds demonstrate 31.4% more residual energy retention than LEACH-C, 81.2% node survivability at round 250, 97.3% Packet Delivery Ratio, 34.7% latency reduction, and 26.8% throughput improvement, with less than 9.2% scalability degradation as network size grows ten-fold. All results are validated over 15 independent runs using Welch's t-test (p < 0.01).	Download