Data fusion technology of wireless sensor network

Http://www.dzsc.com/data/html/2008-11-28/73975.html

Since most wireless sensor network applications are composed of a large number of sensor nodes, the tasks of information gathering, target monitoring and environment sensing are accomplished together. Therefore, in the process of information acquisition, it is obviously inappropriate to use each node to transmit data to the aggregation node separately. Because the network has a lot of redundant information, this will waste a lot of communication bandwidth and valuable energy resources. In addition, it will reduce the efficiency of information collection, and affect the timeliness of information collection.

To avoid these problems, a technique called data fusion (or data aggregation) has been adopted. The so-called data fusion refers to the process of processing multiple data or information, combining the data more efficiently and more in line with the user's needs. In most wireless sensor network applications, many times only care about the results of monitoring, do not need to receive a large amount of raw data, data fusion is an effective way to deal with such problems.

1. The background of data fusion technology is derived from several important functions of data fusion.

(1) Energy saving

Due to the deployment of wireless sensor networks, the reliability of the whole network and the accuracy of monitoring information (that is to guarantee a certain degree of accuracy) are considered and redundant configuration of nodes is required. In the case of this redundancy configuration, the data collected and reported by the nodes around the monitoring area are very close or similar, i.e. the data is highly redundant. If these data are distributed to the aggregation node, the aggregation node is not able to obtain more information, except that the network consumes more energy than it needs to satisfy the accuracy of the data. By using Data fusion technology, we can guarantee the processing of large amount of redundant data information before sending the data to the aggregation node, thus saving the energy resources of the node in the network.

(2) To obtain more accurate information

Due to the environmental impact, data from the sensor node has a high degree of unreliability. By combining the data collected by the sensor nodes in the same region, the accuracy and reliability of information acquisition are improved effectively.

(3) Improve data collection efficiency

Network data fusion, reduce network data transmission, reduce transmission congestion, reduce data transmission delay, reduce transmission data collision, to a certain extent, improve the efficiency of data collection. Data fusion technology can be classified from different angles, the main basis is three kinds: data information content before and after fusion, the relationship between data fusion and application layer data semantics, and the level of fusion operation.

2, according to the data information content before and after fusion divided into lossless fusion and lossy fusion

The former in the process of data fusion, all the details are preserved, only the redundant parts of the information is removed. The latter usually omits some detail information or reduces the quality of the data.

3. According to the relationship between data fusion and application layer data semantics, it is divided into application-dependent data fusion, application-independent data fusion and two combined fusion technologies.

The data fusion which relies on the application can obtain the large data compression, but the cross-layer semantics understanding to the protocol stack implementation brings the great difficulty. Application-independent data fusion preserves the independence of the protocol stack, but data fusion is less efficient. The combination of the above two technologies can be more consistent with the actual application needs of the fusion effect.

4. Data-level convergence, feature-level convergence, and decision-level convergence based on the level of fusion operations

Data-level fusion refers to the fusion of data collected by sensors, which is the lowest level of fusion, and usually depends only on the type of sensor. Feature-level fusion refers to the fusion of object-oriented objects by means of feature extraction, the representation of data as a series of eigenvectors, reflecting the attributes of things. Decision-level convergence is the highest level of integration, based on the needs of the application to make more advanced decisions.

5. The data fusion technology of wireless sensor network can be combined with each protocol layer of the network.

In the application layer, through the distributed database technology, the collected data can be screened initially to achieve the fusion effect; in the network layer, the data transmission capacity is reduced by combining the routing protocol. At the data link layer, you can combine the Mac to reduce the sending conflict and head overhead of the MAC layer, and to save energy, The integrity of the information is not lost. The data fusion technology of wireless sensor network can be applied only when it is designed for application demand.

(1) Data fusion of application layer and network layer

Wireless sensor networks usually have data-centric characteristics, so the application layer of data fusion needs to consider the following factors: wireless sensor networks can achieve multi-tasking requests, the application layer should provide convenient and flexible query submission means; The application layer should provide users with a user interface that masks the underlying operation. Users do not need to change the original operating habits, and do not have to care about how the data collected, because the node communication cost is higher than the cost of node local computation, the application Layer data form should be advantageous in the network computation processing, reduces the communication data quantity and reduces the energy consumption.

From the network layer, the data fusion is usually related to the way of routing, such as address-centric routing (shortest path forwarding route), routing does not need to consider data fusion. However, in the data-centric routing, the source node does not find the shortest path routing data, but it needs to do the data fusion in the intermediate node, and then continue to forward the data. 1, here is a comparison of two different routing methods. The key of data fusion of network layer is the structure of data fusion tree (Aggregatton trees). In the wireless sensor network, the base station or aggregation node collects data from the scattered sensor nodes in the form of a reverse multicast tree. When each sensor node detects an emergency, the transmission

Figure 1 The difference between an address-centric route and a data-centric route

The path to the data forms a reverse multicast tree, which becomes the data fusion tree. As shown in 2, wireless sensor networks report monitored events through a fusion tree.

Figure 2 Using a data fusion tree to report detection events

The structure of data fusion tree can be transformed into a minimum Steiner tree to solve, it is a NP com-plete complete each problem. In this paper, three different non-optimal fusion algorithms are presented.

① is centered on the nearest source node (center at nearest source,cns): The source node closest to the base station or aggregation node acts as the Fusion hub node, and all other data sources send data to that node, which then sends the fused data to the base station or aggregation node. Once the Fusion center node is identified, the fusion tree is basically determined.

② Shortest path tree (shortest paths tree,spt): Each source node transmits data along the shortest path to the base station or aggregation node, and these shortest paths from different source nodes may cross and converge to form a fusion tree. Data fusion is performed at the intersection of the intermediate nodes. When the shortest paths of all source nodes are established, the fusion tree is basically formed.

③ greedy growth tree (greedy incremental tree,git): The Fusion tree in this algorithm is established sequentially. First determine the trunk of the tree, and then gradually add foliage. Initially, the greedy growth tree only has a shortest path to the base station or aggregation node and the node closest to it. Then each time a node from the remaining source node in front of the greedy growth tree is selected to connect to the tree until all nodes are connected to the tree.

The above three algorithms are more suitable for the application of event-driven wireless sensor networks, which can be used to process data fusion before remote data transmission, thus reducing the amount of redundant data. The data can be fused to a certain extent, the above three algorithms of energy efficiency is usually: GIT) SPT) CNS. When the distance between the base station or the aggregation node and the sensor coverage monitoring area is not the same, it may cause some differences in the above algorithm energy saving.

(2) Independent Data Fusion protocol layer

There are some shortcomings in data fusion technology, whether it is combined with the application layer or the network layer: In order to realize the cross-protocol layer understanding and interaction data, the data must be named. Using the naming mechanism can result in the fusion of data from different data types of the same source node, breaking the independent integrity of the traditional network protocol layer, the upper and lower layer protocols can not be completely transparent, the use of intra-network fusion processing, may have a high degree of data fusion, but will result in too much information loss.

He has proposed a data fusion mechanism independent of application (application independent data Ag-gregatlon,aida),

The core idea is to combine multiple data units according to the next hop address, which can save energy by reducing the overhead of the data encapsulation head and reducing the sending conflict of the MAC layer. Aida is not concerned about the content of the data, the background is mainly to avoid relying on the application of data fusion (application dependent Aggregatton,adda) of the shortcomings, but also to enhance the data fusion of network load adaptability. When the load is light, no fusion or low degree of fusion, high load or MAC layer conflict is heavier, the higher degree of data fusion, 3, Aida's basic functional components are divided into two major parts: one is the aggregation of network grouping and convergence of the convergence and elimination of aggregation function unit, and the other is the aggregation of fusion control unit. The former is mainly responsible for the fusion of data packets and fusion operations, the latter is responsible for the link of the busy state control fusion operation, adjust the degree of convergence (combined maximum number of groups).

Figure 3 Basic components of AIDA

Before introducing the workflow of Aida, several structural designs of different methods of data fusion are compared. The traditional Adda has a cross-layer design between the network layer and the application layer, while the AIDA is an independent interface between the MAC layer and the network layer of the Data fusion protocol layer. There are pros and cons to the layering and cross-layer data fusion mentioned earlier. Of course, Aida and Adda can also be combined to apply, as shown in 4. Aida's proposed is to adapt to the network load changes, can be independent of the other protocol layer data fusion, can guarantee not to reduce the integrity of the information and not reduce the network end-to-end delay, reduce the congestion of the MAC layer, reduce the energy consumption.

Figure 4 Several structural designs of data fusion in different ways

Aida's workflow mainly includes operations in the following two directions: send and receive.

Send mainly refers to the operation from the network layer to the MAC layer, the data packet sent by the network layer into the convergence pool, Aida function unit according to the degree of fusion required, the next hop address of the same network unit (data) merged into a AIDA unit, and sent to the MAC layer for transmission. When the fusion unit is called and the degree of fusion is determined by the Fusion control unit.

The receiving operation is mainly from the MAC layer to the network layer, the AIDA unit sent up by the MAC layer is divided into the original network layer grouping unit and sent to the network layer. This guarantees the modularity of the Protocol and allows the network layer to reroute each data packet.

Data fusion technology of wireless sensor network