Sarma Vrudhula at ASU‎ > ‎Research‎ > ‎

Energy Management of Wireless Sensor Networks

Support: My students and I gratelfully acknowledge the following agencies for their generous support: 

  • National Science Foundation (CNS-0905035):  Exploiting Battery­ Supply Non-Linearities in Optimal Resource Management and Protocol Design for Wireless Sensor Networks

Abstract: This project has been a collaborative effort between two research teams: at the University of Arizona (UA, PI: Marwan Krunz) and Arizona State University (ASU, PI: Sarma Vrudhula).  The overall objective has been to develop novel energy management techniques for wireless sensor networks (WSNs) that exploit battery characteristics for the purpose of maximizing the network lifetime.  Energy management must be performed at the node as well as the network levels. The UA team focused their efforts on network-level energy management, which included (1) clustered designs and application of virtual MIMO techniques to dense WSNs which involved the development of distributed protocols for clustering, intra-, and inter-cluster VMIMO-based communications, and (2) development of algorithms for coverage time maximization of a clustered WSN by optimal balancing of power consumption among cluster heads (CHs)

The main focus of the ASU team was on developing energy management techniques for wireless sensor networks (WSNs) that exploit battery characteristics for the purpose of maximizing the network lifetime.  This included (1) the design and construction of a solar-battery powered WSN for environment monitoring, implementing protocols and demonstrating proof-of-concept design, (2) development of protocols in which energy harvesting is employed in conjunction with batteries, and (3) development of optimal algorithms for controlling the sensors, and the charging (via solar power) and discharging of the battery to maximize network coverage.

Solar Powered Active Wireless Sensor Networks:  A WSN consists of many sensor nodes that are spatially distributed over some geographical area. In the case of active sensor networks, sensors must spend energy to monitor various targets. This makes the task of managing the network’s energy supply and consumption a vital part of ensuring high network performance.


Regardless of their energy efficiency, battery-powered WSNs will eventually cause the network to fail due to their limited power supply or their batteries will need to be replaced. This can be a costly procedure if the network is deployed in a harsh environment or in an inaccessible location. A promising solution to this problem is the use energy harvesting in conjunction with batteries.  However, with the time-varying and random nature of solar power maximizing the networks Quality of Service (QoS) becomes a very acute issue.  The QoS is defined as the minimum number of targets that can be covered by the network over a 24-hour period. Assuming a time-varying solar profile, the underlying problem is to optimally control the sensing range of each sensor so as to maximize the QoC. The problem is further constrained by requiring all active sensors to report any sensed data to a centralized base station, making connectivity a key factor in sensor management. 

The mathematical formualtion of the cover problem turns out to be a nonlinear optimal control problem of high complexity. Implicit in the solution is the allocation of solar energy during the day to sensing tasks and recharging of the battery so that a minimum coverage is guaranteed at all times. By exploiting the particular structure of the problem, we developed a novel method for determining near-optimal sensing radii and routing paths as a series of quasiconvex (unimodal) optimization problems. The runtime of the proposed solution is 60X less than the standard optimal control method based on dynamic programming, while the worst-case error is less than 8%. The proposed method is scalable to large networks consisting of hundreds of sensors and targets. The efficiency and the accuracy of the solution method permits exploration of the network design space. In one experiment, the trade-off between the cost of constructing a sensor node and the number of sensor nodes needed to ensure full target coverage is examined. The results suggest that there is a unique optimal number of nodes and corresponding average radii that minimize the network setup cost while guaranteeing a specified QoC. In two other experiments, we find that for certain deployment scenarios, increasing the sampling time of sensors can increase the minimum cover, and increasing the sensor beamwidth has diminishing returns on improving the QoC.

Building a WSN for Monitoring Water Conditions in a Lake

On the experimental side we focused on building and deploying an operational WSN that can be used as a test bed for experimenting with various energy management designs. To make it of practical value, the experimental WSN was targeted to a specific and novel application in the domain of environmental monitoring.  For this domain, it is extremely important that (1) the network not require any substantial infrastructure to work, (2) be of very low cost, and (3) achieve a very high degree of robustness and have practically unlimited lifetime.  To achieve these, the team decided to use energy harvesting (using solar cells) to power the sensor nodes.  This project idea arose after discussions with a professor of the department of Chemistry and Biochemistry at ASU, who routinely and often has to collect water samples for measuring a variety of characteristics. The manual measurements are limited to a few accessible locations along the shoreline. Our goal is to instrument a multi-hop WSN that would provide scientists such as Professor Hartnett and her colleagues, a much more advanced mechanism for monitoring the water in Tempe Town Lake from a remote location and also to greatly increase their dataset sample size.  A WSN will provide both better sampling area coverage as well as data at a rate exceeding manual measurement. Possible water characteristics for monitoring include temperature, pH, conductance, and dissolved oxygen levels that indicate the health of the water system.

The Platform and Power Profiling

The experimental platform we chose for a WSN node is the Texas Instrument’s eZ430-RF2500-SEH Solar Energy Harvesting Development kit.  This kit features TI’s MSP430 microcontroller unit, TI’s CC2500 low power RF IC, as well as a solar harvester that can convert solar and fluorescent light into a storable power source for the microcontroller and the RF IC. Sensors are connected to the pins on the board, which are accessible by the MCU.    

During the course of the work, the need became clear to understand the power consumption of our platform for use in energy calculations. A power consumption profile for programs written and executed on the MSP430 board allows us to predict energy consumption behavior over periods of activity/inactivity. A profile consists of the energy consumption metrics for each stage of processing or transmission during a device’s active state combined with the duration of said active state. Features on the selected platform contributing to the power profile include Active and Inactive Power States Dynamic Voltage Scaling.

Data on the power consumption for individual power states was provided by the device documentation. However, actual measurements were performed at varying voltages to understand real performance. DVS was confirmed by the scaling of power to voltage as well as the reduction in overall power between operational states. Simple test programs, usually flashing an LED, are approximated for power consumption using the measurement results. Using this simple profile, we calculated an average consumption bounded by high and low consumption values.

Automation of power profiles is an ongoing effort so that any program written can be quickly assessed for power efficiency. Using Code Composer, software recommended by Texas Instruments for our platform, we compiled our program and used the generated assembly file to mark the transitions in power states along with the number of instructions between said transitions. Using this calculated figure, knowledge of the instruction set, and our selected clock speed, we can approximate the duration of each state. Functionality to approximate the number of instructions for branches and loops has yet to be fully integrated.

Physical Infrastructure

A great benefit of WSN’s is their minimal infrastructure; however, since we are developing for a harsh environment (a lake) we needed to construct a weather/water-proof casing for the sensor node.  Our first objective is to build a waterproof enclosure to house the sensor board, and have a temperature sensor properly enclosed in a waterproof unit, protruding out of the sensor enclosure.  In addition, a waterproof solar panel must be attached to the sensor enclosure. The initial design the enclosure we settled upon was a Mason jar. This design was limited to a strict budget; however proved effective and cheap.  The pictures below show the details of the parts and manufacturing steps involved in building a waterproof WS node.  


The Software

Upon learning how to program the eZ430-RF2500 and the SimpliTI network protocol, basic shell programs were developed for monitoring the environment. As a result, we can establish an access point to manage the network addresses, store/forward information to other nodes, and log recorded information to a PC via the UART port of the MSP430; provide end devices for the network that will monitor temperature via the analog temperature sensor (TMP37 by Analog Devices).  The end devices periodically sample the external temperature sensor and send the data back to the access point which will confirm successful receipt of the message as well as log the information to a PC. This basic program is intended to act as a shell for this project.  Increasing functionality was added to the program to improve power management including additional sensors, sensor sampling rate optimization, network topology optimization, as well as micro controller sleep state optimization. 

The waterproof housing was constructed and the application was tested.  It was found that transmitter inside the glass jar must be above the water; otherwise the signal becomes highly attenuated. In addition, the temperature sensor had to be encased in a copper casing, and this required re-calibration.  Another major challenge was to prevent condensation from taking place inside the glass jar. Several low cost methods for doing this are currently under investigation.