Cross-Layer Scheduling for Power Efficiency in Wireless Sensor Networks Mihail L. Sichitiu Department of Electrical and Computer Engineering North Carolina State University IEEE Infocom 2004 Speaker ： jenchi

Outline Introduction Proposed approach The steady state phase The setup and reconfiguration phase Simulation results Conclusion

Introduction The characteristic of sensor nodes Very small form Be inexpensive and deployed in very large numbers Once deployed, the sensor networks are usually unattended The life of the sensor network is determined by the life of its batteries Such a network is typically expected to work for extended periods of time

Introduction The power consumption in sensor networks Idle listening Retransmissions resulting from collisions Control packet overhead Unnecessarily high transmitting power Sub-optimal utilization of the available resources

Introduction Two modes of sensor networks Event driven sensor network Sensor nodes do not send data until a certain event occurs  A forest fire monitoring application difficulty ： the sensor network is able to wake up the entire network when the event occurs Continuous monitoring sensor network Data is sampled and transmitted at regular intervals  A temperature monitoring station Indeed, in order to detect an event, a continuous monitoring scheme is necessary

Introduction This paper presents an approach tailored specifically to the needs of sensor networks with continuous monitoring capabilities The proposed scheme derives its power efficiency from eliminating idle listening and collisions in the sensor network The need for such a scheme is highlighted and prompted by recent habitat monitoring applications

Proposed approach To develop a framework for deterministic optimal energy conservation while maintaining the network real-time characteristics Sensor nodes dynamically create on-off schedules in such a way that the nodes will be awake only when needed and asleep the rest of the time

Proposed approach The proposed scheme can be decoupled into two distinct phases for each flow in the network The setup and reconfiguration phase To set up the schedules that will be used during the steady state phase The steady state phase It utilizes the schedule established in the setup and reconfiguration phase to forward the data to the base station

Proposed approach Assumptions The traffic is periodic, with the same period in the entire network Each node originates only one packet in each period To ensure that the control packets necessary to set up schedules do not collide with the data packets forwarded, a two-level priority scheme MAC layer has to be used

Proposed approach State diagram for each data flow in the network

The steady state phase Assume that the network is perfectly synchronized The nodes on the path forward DATA packets at the appropriate times Each node on the path stores a schedule table that specifies when various actions have to take place

The steady state phase The three different actions Sample The source node taking a data sample Be forwarded along the path to the base station Transmit The action of transmitting a packet of the flow to the next node on the path to the base station Receive The reception of a packet The packet will be further transmitted to the next node

The steady state phase The actions are stored in a schedule table that has two columns the first column ： what type of action that has to be performed the section column ： when a certain action has to take place

The steady state phase □

The setup and reconfiguration phase The schedule setup algorithm for any flow proceeds in two steps Route select A route to the base station is selected Route setup Schedules are being setup along the chosen route If it fails, a new route is selected

The setup and reconfiguration phase The route select step The setup and reconfiguration algorithm is independent of the underlying routing algorithm Power aware routing algorithms may be preferable

The setup and reconfiguration phase The route setup step A route setup (RSETUP) packet will be sent on the selected route from the source of the flow to the base station To find a time when a DATA packet can be scheduled without colliding with other nearby transmissions The RSETUP packet contain the node source and the duration of the packets on that flow

The setup and reconfiguration phase A lower priority (lower than data on established paths) RTS/CTS handshake ensures that only non-interfering transmissions are scheduled at overlapping times The length of the route setup packet is larger than the length of a data packet Proposed priority scheme similar to (SIFS/DIFS)

The setup and reconfiguration phase

Simulation results Custom simulator (existing simulators cannot simulate hundreds of nodes for periods of months and years). Network lifetime = when 50% of the nodes cannot report to the base station (either the batteries are depleted or no available routes) Will compare three technologies: Always On (sensors in receive mode when not transmitting) power savings mode Proposed scheduling approach.

Simulation results Will use a base case and vary one parameter at a time. Base case parameters: Nodes: 100 Transition from “off” to “on”:3ms Area: 100x100m Transmission Radius: 25m Current in TX mode:17mA One sample sent every: 60s Synchronization precision: 1ms PSM beacon interval: 500ms Current in RX mode:10mA Current in idle mode:10µA

Simulation results Simulation results for base case Network Lifetime MeanStd. deviation No Power Savings 8.3 days4 minutes PSM3.2 months7.5 days Power scheduling 24.2 months5 months

Simulation results — Dependency of the network lifetimes on the number of nodes for a constant deployment area Network lifetime [s] Initial number of nodes

Simulation results — Dependency of the network lifetimes on the number of nodes for a constant density Network lifetime [s] Initial number of nodes

Simulation results — Dependency of the network lifetimes on the measurement period of the network Network lifetime [s] Measurement period of the sensor network(s)

Simulation results — Dependency of the network lifetimes on the power consumption in idle mode Network lifetime [s] Current drawn in idle mode (mA)

Simulation results — Dependency of the network lifetimes on the precision of the synchronization algorithm Network lifetime [s] Synchronization precision Δ [ms]

Conclusion Presented a new distributed scheduling algorithm for stationary continuous monitoring sensor networks Is fully distributed and works in tight cooperation with popular sensor networks routing and MAC families of protocols For the right type of networks, it is shown via simulations that the network lifetime can be increased by orders of magnitude