Q Learning Based Adaptive Protocol Parameters for WSNs

Abstract

Wireless sensor networks (WSN) are widely used for multi-disciplinary applications. According to the requirements and the goal of the application, the network is designed and the protocol is tuned to obtain the best performance of the WSN. In real world applications, all nodes in the network have a common protocol parameter set, irrespective of their position in the network. In several experiments with multihop sensor networks, we observed that individual nodes perform differently depending on the protocol parameter values. With the observation the question was raised whether the performance of the network can be improved by using tuned parameter sets for each individual node in the network. Tuning protocol parameters for each node manually is tedious and may not be practical for large number of nodes. As a solution, adaptive protocol parameters are introduced using reinforcement learning. The learning algorithm gradually approaches an optimal set of protocol parameter values for each and every node during the runtime resulting in average improved network performance with 13.44% and 29.41% compared to networks with static common parameter sets in a network of 20 and 30 nodes respectively in simulation environment. The performance of the adaptive protocol is validated using real testbed with 10 nodes and the performance improvement is 16.21%. With the simulation results it was observed that networks with higher number of nodes obtain more performance gain using the adaptive protocol algorithm compared to networks with lower of nodes.