In-network aggregation with size reduction for wireless sensor networks--quantitative analysis.

Vasar, Cristian ; Biriescu, Marius ; Mihet Popa, Lucian 等

1. INTRODUCTION

A WSN is set of spatially distributed autonomous unwired sensors (devices) that monitor at different site locations, various physical or environmental conditions, such as: physical properties (temperature, pressure), motion properties (velocity, position), contact properties (force, torque), presence (on), biochemical etc. The development of wireless sensor networks was initially motivated by military applications such as the case of battlefield surveillance. However, in many non-military applications, related to environment and habitat monitoring, healthcare applications, home automation or traffic control, the usage of wireless sensor networks become very broaden. (Cook & Das 2004)

Wired connections have important disadvantages: limit units mobility, does not allow wired devices to be closed to the monitored phenomenon, and not in the last considering the large number of sensors or actuators from a real environment, the wires implicate important maintenance problems, with high costs (Rabaey et al. 2000). These facts make wireless communication between such devices, an inevitable necessity in many applications. However, the wireless communication can be extremely costly in comparison with local processing of data, in the specialized literature it is state that sending a single bit over radio is at least three orders of magnitude more expensive than executing a single instruction locally (Shrivastava et al.2004).

2. WIRELESS SENSOR NETWORK

In wireless sensor networks, each sensor collects information from the surrounding environment, perform basic processing and transmit the data to user using the network infrastructure. A simple approach is based on periodically communication between sensors and a base station, where data is stored and processed. The main inconvenience of this basic method is the significant amount of transmitted data. The communication optimization becomes a demand, considering the "energy efficiency vs. data accuracy" trade-off. Usual means to improve these issues are based on multi-hop routing protocols applied to WSN. Data from multiple sensor nodes is merged together and then rerouted to base station to reduce the number of packets that propagates through the network. The base station becomes the root of the routing tree and the sensor nodes are organized in branches. Data flow from the sensors toward base station along this routing tree. The number of messages propagated through the network is reduced at the cost of rising the size of the messages (Heinzelman et al. 2000). Also the benefit of in-network data aggregation over significant energy savings within WSNs has been confirmed theoretically (Krishnamachari et al. 2002) by reducing the in-network communication, even if this is achieved by increasing the local processing at mote level.

In the simulations performed in the paper the WSN topology is according to Fig. 1 (but different WSN sizes were considered). Each mote has sensors, limited processing and radio capabilities. All the measurements performed inside WSN are required at the base station level, which has limited radio capability but can be integrated into a powerful computational system for complex processing of data provided by WSN.

Due to the power limitation of radio communication results a limitation on the feasible distance between the sender and the receiver. Because of this restriction, the simple and direct communication between source and destination is not always possible, specially in WSNs, that are designed to cover large areas (environmental applications) or that operate in difficult radio environments with strong attenuation (in buildings). In order to solve the issue of limited distances, an usual approach is the usage of relay units, where the packets take multi hops from source to destination (Karl & Willig 2005).

If the data is communicated using minimized routes this energetic costs are optimized. This can be achieved by defining routing trees, which will route all communication to motes located on the first diagonal and then along it.

In our case, the distances between sensor nodes were maximized, but obviously they cannot exceed the radio-communication range. Each node can communicate directly only with nearing neighbours. For example, the covered radio-communication area for node 22 is marked with grey colour. In this circumstances, the transmission from other node (except node 36) to base station will involve other nodes that will be used as repeaters, and the communication will require more receptions and transmissions that will increase the energetic consume as the distance between the node and the base station grows.

[FIGURE 1 OMITTED]

The main objective of the wireless sensor network is to extract the global maximum and the secondary objectives are to find local maximums from the monitored area.

3. SIMULATION STUDIES

To determine the quantitative impact of in-network data aggregation on the total amount of network and implicit over the energy consume of a wireless sensor network, a number of simulations had been performed, considering networks of different size and different distributions of measured values.

For example, in Fig.2 is depicted the case of a WSN containing 625 sensors, and the measured values are uniform distributed among the network. In this case, using a WSN with no aggregation and extracting all the measured values from it requires a total number of 10725 communications at each sample time.

In order to reduce this amount of communications but not loose information regarding global and local maximum values, since each sensor receive directly the communicated values from its nearing neighbours (maximum four neighbours), and if its own value is smaller than the maximum value of its neighbours it will not communicate its value at the present sample time. The communication start from the mote situated at the longest distance from base station and is performed row by row, and at one sample time all measured values are evaluated and/or communicated.

Taking into all above considerations, for uniform distributed values of measurements, the number of total communications is decreased at half (an average of 5650 instead of 10725)--extending with about 90% the average battery life. In Fig. 3 is presented the evolution of total communications number during an interval of 40 sample time. The simulated networks had 625 motes (placed in a 25 X 25 matrix). The number can be decreased even more by aggregate the maximums over clusters defined over the routing tree, but in that case the local maximums are not detected.

[FIGURE 2 OMITTED]

[FIGURE 3 OMITTED]

[FIGURE 4 OMITTED]

To study the influence of data aggregation on different network size, another simulation study has been performed, considered wireless sensor networks having 9 to 1024 sensors and a number of 20 iterations for each network case. The results are depicted in Fig. 4, with solid--the network with data aggregation and with dotted--the case of network without aggregation. The network size n is on the abscise, considering that all network have n elements, placed in a n X n matrix.

4. CONCLUSION

The presented aggregation method decreases at half the number of communication at each sample time, but this can be further lowered combining this method with other techniques (using clusters for example). Performing in-network data aggregation in case of wireless sensor networks with limited resources (as bandwidth or energy) can optimize the communication process, without loosing sensitive information. This improves the trade-off between battery life and data accuracy. One disadvantage of the presented method is the relative reduced tolerance to node malfunctions, in case if one node fails, the entire communication suffers. This can be overcome by using dynamical routing trees.

Future research directions will consider the possibility to use a priori known information about the monitored process to optimize the energy consumption of the entire WSN (for example defining the routing tree according with the probability to[degrees]Ccur the maximum for different monitored area, or using multiple base stations).

5. REFERENCES

Cook D.J. & Das S.K. (2004). "Smart Environments: Technologies, Protocols, Applications", Chapter 2 "Wireless Sensor Networks-F.Lewis", Wiley-Interscience, New York.

Heinzelman W., Chandrakasan A., & Balakrishnan H. (2000). "Energy-efficient Communication Protocols for Wireless Microsensor Networks", In Proc. Hawaaian International Conf. on Systems Science, January 2000.

Karl H. & Willig A. (2005). "Protocols and Architectures for Wireless Sensor Networks", John Wiley & Sons, Ltd. ISBN: 0-470-09510-5, pp 60-64.

Krishnamachari B., Estrin D., & Wicker S. (2002). Impact of data aggregation in WSN. International Workshop on Distributed Event-Based Sytems, Vienna, Austria, July 2002.

Rabaey J. M., Ammer M. J., da Silva J. L., Patel D., & Roundy S. (2000). "PicoRadio Supports Ad Hoc Ultra-Low Power Wireless Networking" IEEE Computer, 33(7): 42-48.

Shrivastava N., Buragohain C., Agrawal D. & Suri S. (2004). "Medians and Beyond: New Aggregation Techniques for Sensor Networks," in Proceedings of the Second ACM Conference on Embedded Networked Sensor Systems (SenSys 2004), August 2004.