Dynamic power management for QoS-aware wireless multimedia sensor networks

The integration of multi-hop capability into conventional wireless networks is perhaps the most promising architectural upgrade in next generation wireless networks with high throughput and coverage requirements. Multi-hop communication covers wireless sensor networks as an important component in the future heterogeneous wireless networks. In a distributed sensor network, each node acts as a relay node to forward data packets from other nodes. More recently, the production of cheap CMOS cameras and microphones, which can acquire rich media content from the environment, has created a new wave into the evolution of wireless sensor networks towards wireless multimedia sensor networks. Wireless multimedia sensor networks are gaining increasing popularity due to numerous potential applications such as video surveillance, environmental and habitat monitoring, etc. In a wireless multimedia sensor network, limited battery power at the wireless sensor nodes along with the transmission quality requirements for multimedia traffic make quality of service (QoS) provisioning a very challenging task.;In this work, the tradeoff between the QoS and the energy saving performance is first analyzed. In particular, a novel queueing analytical framework is presented for performance evaluation of a node in a multi-hop wireless network with distributed and energy-aware medium access control (MAC) protocol. The MAC protocol in the node acts as a server for different types of multimedia arrival traffic in the node with different priorities. A vacation queueing model is also used to model the sleep and wakeup mechanism of the server. Two different power saving mechanisms due to the standard exhaustive and the number-limited exhaustive vacation models (both in multiple vacation cases) are analyzed to study the tradeoff between the QoS performance of the relayed packets and the energy saving at a relay node. We use matrix-geometric method to obtain the stationary probability distribution for the system states from which the performance metrics are derived. Using phase-type distribution for both the service and the vacation processes and combining the priority queueing model with the vacation queueing model make the analysis very general and comprehensive.;A dynamic power management framework is developed based on a Markov decision process (MDP) for a wireless multimedia sensor node to improve the energy saving performance while satisfying the multimedia transmission quality requirements. Dynamic programming and reinforcement learning are used as two different approaches to solve the MDP problem. The dynamic programming framework considers the video traffic arrival process in the sensor node, the sleep and wakeup processes in both the camera and wireless transceiver electronics, the queue status, and the multi-rate wireless channel condition. Dynamic programming approach is used to find the optimum policy to achieve the desired performance measures in an energy-limited sensor node.;To overcome the curse of dimensionality in dynamic programming approach, a reinforcement learning-based distributed dynamic power management framework is presented. The queueing model in this case captures bursty nature of the multimedia (e.g., video/voice, data) traffic arrival process and prioritization between video/voice and data traffic. A model-based reinforcement learning approach is used to solve the MDP problem in which the structure of the transition probability matrix in the MDP formulation is exploited to reduce the convergence time in the learning process.