Snizhko Ye.M., Palamarchuk Yu.A., Botsva T.O.

Dnipropetrovsk National University named after Oles Honchar

ENERGY EFFICIENT ALGORITHMS BASED ON NODE STATE MANAGEMENT IN WIRELESS SENSOR NETWORKS

       Introduction

       Wireless sensor networks (WSNs) are ones of the most widely used and rapidly developed technologies. Typically, WSNs are distributed measurement systems consisting of a large number of measurement units deployed over a wide area; each unit is a low-power device that integrates processing, sensing and wireless communication functions. Units acquire information from the surrounding environment and, after a possible but generally not required local processing, send arrays of measurements to one or more collection points i.e. base stations for further data interpretation. As one of potential scenarios, environment monitoring application can particularly benefit from this technology as WSNs allow a long-term data collection at those scales and resolutions that are difficult to achieve with traditional data gathering techniques. The number of WSN applications has increased and expected to increase even more in future. But energy consumption still remains the major obstacle for further exploitation of this technology, even though batteries can be possibly recharged (e.g. solar energy batteries).

       Energy saving approaches for WSNs generally assume that data gathering and processing have an energy consumption significantly lower than that of communication (transferring data array to the base station) [1]. But numerous recent studies proved that this assumption does not work for a number of practical applications, where sensors may consume even more energy than the radio [2]. Thus we need an approach for power management on a node level that, however, still can support network-level requirements such as fault tolerance. Dynamic management of a node state can significantly decrease energy consumption in an idle mode, different configurations of a sleep mode and all parts of the node duty cycle that don't require maximum of performance.

       Theory

       A wireless node is battery operated and thus energy constrained. To maximize the sensor node’s lifetime after its deployment – the moment when the WSN becomes unattended, all it's life cycle parts including circuits, architecture, algorithms, and protocols— must be energy efficient. Energy savings can be reached by using dynamic power management (DMP) that generally means that the sensor node is shut down if no  events (in particular, for sensing) occur. Such event-driven power consumption is effective to maximize battery life. Also the node should have an energy scalability - a possibility to extend the lifetime at a cost of sensing accuracy by decreasing either sensing frequency or transmitting frequency (that lowers accuracy of the whole network). Different energy-scalable algorithms and protocols exist for these energy-constrained operating modes. [3]

       Sensing applications have a wide range of requirements that leads to an idea that protocols and algorithms are to be modified for each particular WSN application. In this article we propose an OS-directed power management technique to improve the energy efficiency of sensor nodes without significant loss in performance, data accuracy and fault tolerance.

       Fig. 1 represents a WSN node and its functional parts: embedded sensor, analog-digital converter, a processor (in our modeling experiment we consider StrongARM SA-1100 as a hardware implementation), memory and the radio (both transmitter and receiver). The node is running under a micro-operating system (μOS) that is responsible for power management. Also in our particular case we consider a network with cluster topology and nodes with a random uniform distribution.

       The core point of dynamic power management approach is to create an array of node modes for each phase of a life cycle and a bunch of policies for mode-to-mode transitions. We should not only carefully distinguish sleep and idle modes, but also apply different energy grades for active/working state, considering that maximum performance and CPU load may not even occur through the whole duty cycle. Dynamic voltage scaling can be used to reduce operational frequency as low as enough to process the data array between two data gathering/transmitting phases.


Figure 1. WSN node structural parts

         Modeling experiment

       In this particular experiment we focused on sleep mode usage for energy saving. We considered five both possible and making practical sense for this particular implementation distinguished sleep modes.

Table 1

Distinguished sleep modes

Mode

Processor

Memory

Sensor

Radio

SM1

Active

Active

On

Transmitter, Receiver

SM2

Idle

Sleep

On

Receiver

SM3

Sleep

Sleep

On

Receiver

SM4

Sleep

Sleep

On

Off

SM5

Sleep

Sleep

Off

Off

 

Also an event term is introduced as a moment when a sensor receives a signal above some predefined level and this event happens within the sensor visibility radius R.

       It's obvious, that SM5 mode has the lowest possible energy consumption but is also the only one with sensing function turned off. Thus this mode should be eliminated for all real systems where each and every event is critical (and should never be missed) and occurs only once and within only one node R. In other cases a probabilistic approach should be introduced for transitions to SM5 mode. The time for which the node K is switched to SM5 is proportional to the probability that no event happens on node K within this time frame. The probability equations are either field-specific, predefined and known or calculated on μOS level based on node K detection statistics.

       Using TinyOS embedded simulation framework a model for a WSN with N=1000 nodes uniformly distributed over area with L=100 m and W=100 m (see Fig.1), R=10 m has been built. The event behavior over the entire sensing region was as Gaussian spatial distribution centered around (25, 75). A probabilistic algorithm for transitions from SM1 to SM5 mode was applied.

Figure 2. (a) event distribution (b) energy consumption distribution over nodes

       Without dynamic power management for all nodes energy consumption would be equal.

       Now consider the fraction of events missed (while node in SM5 mode) for corresponding energy consumption (normalized energy per one node) (Fig.3)


 


       P stands for the probability to miss an event on node K calculated on μOS level for Gaussian event distribution. Those values for missed events are suitable for almost all environment monitoring WSN applications [3].

       Conclusion

       Energy efficiency was reached by using node-level dynamic mode management with probabilistic approach to mode transitions. The results showed that this approach allows to adjust power consumption to real sensing tasks without significant data accuracy decrease. Distinguishing sleep modes more accurately and applying the same mode transition policy we can reach minimal energy consumption for each particular WSN duty cycle.

         References

1. W. Heinzelman, A. Chandrakasan, H. Balakrishnan, “Energy-Efficient Communication Protocol for Wireless Microsensor Networks”, Proc. Hawaii International Conference on System Science (HICSS-33), January 2000

2. Rezaei. Zahra, Mobininejad Shima. “Energy Saving in Wireless Sensor Networks”, International Journal of Computer Science & Engineering Survey (IJCSES) Vol.3, No.1, February 2012.

3. C. Alippi, G. Anastasi, C. Galperti, F. Mancini, M. Roveri, “Adaptive Sampling for Energy Conservation in Wireless Sensor Networks for Monitoring Applications”, Proc. IEEE International Workshop on Mobile Ad hoc and Sensor Systems for Global and Homeland Security (MASS-GHS 2007), Pisa (Italy), October 8-12, 2007.