KR100948837B1 - Method and appratus for distributed position recognition in wireless sensor network - Google Patents

Abstract

Disclosed is an invention relating to distributed location recognition in a wireless sensor network.

The present invention searches for nodes located in an area where a location is known and can directly transmit a signal wirelessly, selects a probability distribution model indicating a probability of distributing nodes according to a distance, and selects a probability based on each location of the found nodes. It is characterized by calculating a probability that nodes are distributed in an area where a signal can be directly transmitted wirelessly by applying a distribution model, and determining a position among areas in which nodes are most likely to be distributed.

Wireless sensor network, location awareness

Description

Method and device for distributed position recognition in wireless sensor network

The present invention relates to a large scale wireless sensor network, and more particularly to location recognition of a wireless sensor terminal.

The present invention is derived from the research conducted as part of the IT growth engine technology development of the Ministry of Information and Communication and the Ministry of Information and Communication Research and Development. [Task Management Number: 2005-S-038-03, Title: UHF RF-ID and Ubiquitous Networking Technology] Development].

In the wireless sensor network, there are largely a centralized method and a distributed method for calculating node positions. The centralized method assumes a central server, transfers all the data needed for location calculations from each node to the central server, and then calculates the location for all nodes on that central server. After all calculations, the location information of each node is sent back to the node. Such centralized methods include SDP (Semidefinite Programming) and MDS (Multidimensional Scaling).

SDP expresses the geometric conditions of nodes in linear matrix inequalities (LMI). Once all the conditions are expressed in these LMIs, they can be combined into a semidefinite program. This creates a figure representing each region of each node, which has the following disadvantages.

First, not all geometric conditions can be represented in LMIs. Usually, the condition is easy to express in one LMI only when the condition is convex, that is, all faces are convex. For example, in the case of AoA (Angle of Arrival) data, a triangle is represented, and in the case of a signal strength such as RSS (Received Signal Strength), a circle is represented. On the contrary, in the case of a ring shape, since it is not a convex form, there is no way to properly express data in a more accurate range. These points are important disadvantages in the accurate location recognition technique.

In addition, in the execution time of the SDP, the AoA data has a characteristic that is proportional to the square of the number of geometric conditions, and the RSS data is proportional to the cube of the number. Therefore, even if all the conditions expressing the network are shown, the execution time is long.

MDS is a technique for indicating inconsistencies of elements within a set. When applied to position recognition, the following method is used.

First, measure the distance between each node in the network. We put this into a matrix with the number of nodes and create a relative coordinate scale using a technique called Singular Value Decomposition (SVD). Applying this value to each node based on the values created in this way can calculate the relative position information of all nodes, although it is not absolute position information. After that, with the help of nodes with absolute location information, the location information of all nodes is calculated.

This method shows satisfactory performance when the total number of nodes is above a certain level. However, as the number of nodes increases, a lot of effort for SVD is required.

The distributed location recognition technique is proposed to improve the disadvantages of the centralized method described above. This means that when there are a large number of nodes or nodes that need to do calculations are not sufficient, the calculations are distributed so that one node does all the calculations, and all nodes divide their calculations so that the overall system load It can be seen as a way of dropping.

For example, the Multilateration and distributed MDS methods are examples. The Multilateration method determines the location with the help of neighboring nodes that know their location. Finally, the position information is combined to calculate its own position. The advantage of this method is a simple calculation method, but the fact that the location calculation cannot be performed if there are not enough nodes in the entire network knowing their location information, and the error in the distance prediction The disadvantage is that it propagates to neighboring nodes using location information.

Distributed MDS is divided into three main steps. First, in any node, it forms a group with neighboring nodes in its communication area. Secondly, the distance information between all nodes in a group is calculated through communication between the nodes and neighboring nodes, and then the centralized MDS described above is performed. Finally, the total location information is calculated by combining the location information created in each group from the nodes that belong to several groups in common. It has the advantage of applying MDS that calculates location information effectively, but similarly to Multilateration just described, when there is an error in the location information calculated in one group, when the location information of each group is combined, the error is There is a disadvantage that it increases the error of the global location information by spreading to the entire group.

The problem to be solved by the present invention is to provide a distributed position recognition technique with improved accuracy while reducing the complexity of the calculation.

In the wireless sensor network according to the present invention, a distributed location recognition method includes (a) location of a plurality of reference nodes whose location is known, network information of the reference nodes, and information of neighboring nodes located in an area capable of transmitting a direct signal wirelessly. Determining an initial position based on the; (b) selecting a probability distribution model indicative of the probability that the nodes will be distributed over distance; (c) calculating a probability that a node is distributed in the region by applying the probability distribution model around the positions of the reference nodes; And (d) correcting the initial position to one of the regions where the node is most likely to be distributed.

In the wireless sensor network according to the present invention, the distributed location recognizing apparatus is initially based on the location of a plurality of known reference nodes, the network information of the reference nodes, and the information of neighboring nodes located in an area capable of transmitting a direct signal wirelessly. An initial position determiner for determining a position; A probability distribution model selection unit for selecting a probability distribution model indicating a probability that nodes are distributed according to a distance; A probability distribution calculator configured to calculate a probability that a node is distributed in the region by applying the probability distribution model around the positions of the reference nodes; And a position corrector configured to correct the initial position to one of regions where the node is most likely to be distributed.

According to the present invention, highly accurate and efficient location recognition can be performed while reducing complexity in the wireless sensor network.

Hereinafter, with reference to the accompanying drawings will be described in detail the present invention.

The present invention includes a reference node search step, a neighbor node search step, an initial position determination step, a position correction step through node distribution modeling, and a position determination step through a cyclic algorithm. Hereinafter, each step will be examined in detail.

1.Retrieve reference node step

In order to calculate, correct, and determine the positions of general nodes that do not know their positions, they need to know the position information of the reference nodes and their network relationships. Here, the network relationship may include the hop count calculated by the short path method. When a network is first formed, general nodes that do not know their location receive location information and network information of a reference node. In a practical example, these reference nodes may be devices such as cluster headers that are equipped with a self-locating device such as GPS and perform better than a normal node.

The formation of clusters is as follows.

In order to efficiently manage large-scale sensor networks, the formation of clusters should be preceded. The cluster is formed by selecting a cluster header and forming a cluster around the cluster header.

1 is a structural diagram illustrating one embodiment of the distribution of nodes in a network, in accordance with the present invention.

Referring to FIG. 1, each circle represents nodes belonging to a network, among which circles 110-150, which are shown in black, represent cluster headers. The determination of the cluster header is generally determined to be the highest among neighboring candidate nodes, a measure of connectivity determined by the number of neighboring nodes that can communicate among candidate nodes that may be cluster headers. Communication in the present invention means that the terminal receives a signal level of a certain level or more, or has a signal-to-noise ratio value of a certain level, or has an environment in which packet transmission can be received without errors. In other words, being able to communicate may specify that all types of information that can be exchanged between nodes within the area are trustworthy.

2 is a conceptual diagram illustrating an embodiment of a cluster formation process according to the present invention.

Referring to FIG. 2, nodes coming into the communication ranges 215, 225, 235, 245, and 255 of the cluster header may receive header recognition information from the cluster headers 210, 220, 230, 240, and 250. Therefore, these nodes may belong to the cluster by responding to the recognition information or through a series of processes corresponding to the recognition information. At this time, if one node can accommodate recognition information of several cluster headers, that is, nodes in overlapping sections (gray areas in FIG. 2) of two or more cluster areas may be overlapped in the corresponding clusters.

2. Neighbor Discovery

For efficient location awareness, each node in a cluster forms a set of neighboring nodes through communication with other nodes. All nodes in the cluster are recognized in the cluster header, which means that all nodes in the cluster know their existence by sending official information in the cluster header.

3 is a conceptual diagram illustrating an embodiment of a neighbor node search process according to the present invention.

Referring to FIG. 3, the circles 310 and 320 represent nodes, and the large circle 330 including the circles represents an area (wireless transmission range) in which the wireless communication of the node 310 corresponding to the white circle is directly possible. Indicates. Each node uses only nodes in the same cluster within its radio transmission range to create its neighbor node set. Thus, nodes 320 are each neighboring nodes of node 310.

3. Initial Positioning Step

In a general node that does not know its location, the initial location value calculation using the location information of the reference node, network information, and information of its neighbor nodes may be performed in various ways.

As a representative example, using the multi-line method (mulilateration), the node first determines the distance value of how far from the neighboring nodes on average by measuring the received signal strength from the neighboring nodes. Next, the node calculates how far it is from the reference node. In this case, hop number and average distance information using the short distance method may be used. Next, the initial value is calculated using the position value of the reference node and the estimated calculated distance value from the reference node to the corresponding node. At this time, since the initial value has a high probability of including an error in the calculation characteristic, a correction step must be performed.

In addition, there is a relative initial position coordinate calculation using only the connection diagram. It calculates the relative coordinates, not absolute coordinates, in the initial position calculation, so that the cluster header of the cluster to which it belongs is set to (0,0) and calculated, and then sends its relative position coordinates to the cluster header. The header can also send the server the relative coordinates of normal nodes with their location information. Like the previous method, this initial position calculation involves a lot of errors, so a correction step must be made.

4. Location Prediction Based on Node Distribution Modeling

A typical large scale wireless sensor network is characterized by its uniform distribution mostly for efficient sensing in the practical purpose of the network. This is a distribution model that provides optimal performance in wireless sensor networks with low power and low capacity, which has been proven in many papers.

Lattice models to represent a uniform distribution of neighbor nodes are as follows.

4 is a structural diagram of an embodiment of a rectangular grid model for showing the distribution of neighboring nodes.

Referring to FIG. 4, the white circle 410 is a node whose current position is to be determined and the surrounding black circles are neighboring nodes of the node 410. 4 is suitable when the number of neighboring nodes is large.

5 is a structural diagram of an embodiment of a triangular grid model for showing the distribution of neighboring nodes.

Referring to FIG. 5, the white circles 510 and the black circles are the same as in FIG. 4. The triangular lattice model of FIG. 5 may be applied when the number of neighboring nodes is relatively smaller than that of the rectangular lattice model of FIG. 4.

6 is a structural diagram of other embodiments of a triangular grid model to represent the distribution of neighboring nodes.

Referring to FIG. 6, white circles 610, 620, and 630 and black circles are the same as in FIG. 4. The triangular lattice models of FIG. 6 are preferably applied when the number of neighboring nodes is smaller than that of FIGS. 4 and 5.

If the number of neighbor nodes is less than three, the grid model cannot be formed. Such a case is not considered in the present invention. In addition, in the present invention, it is assumed that the model of the formed cluster has a convex shape in all parts and can have a shape similar to the shape of a polygon or a circle because it deals with it in a large sensor network in forming a cluster.

5. Positioning using a cyclic algorithm

In the initial operation, we figured out the number of nodes that need to be located and formed a set of neighbor nodes for each node. Now, we select the geometric model of the distribution model according to the number of neighboring nodes and choose the probability distribution model accordingly.

7 is a graph illustrating an embodiment of a distribution probability of a node over a distance according to the present invention.

Referring to FIG. 7, as a gamma distribution model, r represents a wireless transmission range of a specific node, x-axis represents a distance from a specific node, and y-axis represents a distribution probability of a neighbor node. This gamma distribution model is preferably applied when the distribution of neighbor nodes is a rectangular grid model.

8 is a graph illustrating another embodiment of a distribution probability of a node according to a distance according to the present invention.

Referring to FIG. 8, as the Gaussian distribution model, the r, x, and y axes are the same as in the case of FIG. 7. This Gaussian distribution model is preferably applied when the distribution of neighboring nodes is a triangular lattice model.

Finally, the number of algorithm cycles is determined, which is based on the computational power of the node, the amount of memory available, and the amount of power available.

After the above initialization, the probability of nodes is calculated by applying a probability distribution model around each position of neighboring nodes. Estimate the position of the node among the highest probability. Also, by returning this location information to neighboring nodes, it is possible to know the relative location information between all nodes in one cluster later, and when the neighboring nodes become the corresponding nodes in the location calculation, the location information thereof can be accurately calculated. Make sure After the position estimation is completed for the nodes requiring position determination in one cluster, it is desirable to repeat this process by the number of algorithm cycles defined above for more accurate position estimation.

In relation to the application of the present invention, an absolute coordinate position recognition method for correcting by using a reference node having absolute position coordinates from the beginning in initial coordinate calculation and a local map in each group by dividing several nodes into one local group In other words, all local maps are calculated by calculating relative position coordinates, and then later combined to make a global map, and then recalculate absolute position coordinates.

Since the position recognition method using the relative position coordinates described in the latter calculates the position information about one cluster, the cluster header which is the basis of the position estimation will first perform such position estimation calculation. Therefore, the position of the cluster header is preferably (0.0) because it is centered.

Subsequently, a total position information calculation for absolute position information acquisition, which is the final process in position recognition through relative position coordinates, is not shown but conceptually described as follows.

In the case of nodes belonging to two or more clusters, its location information is one for each cluster to which it belongs. By using this information, the positional information between clusters can be rearranged by comparing relative positional information of adjacent or overlapping neighboring clusters with each other. After arranging the location information of the entire network in this way, the absolute location information of the entire network can be determined for all nodes using some reference nodes that know the absolute location.

The above invention is summarized as follows.

9 is a flowchart illustrating an embodiment of a distributed location recognition method according to the present invention.

9, the initialization is first performed (S910). The initialization operation includes searching for neighboring nodes of a node to which the current position is to be determined and determining a probability distribution model indicating a distribution probability of nodes according to a distance. At this time, it is desirable to determine the probability distribution model for the neighbor nodes according to the number of neighbor nodes or the distribution environment such as density, topographical characteristics, and equipment configuration. It also includes determining the number of cycles of the algorithm to increase the accuracy of the position.

Next, the probability of existence of a node is calculated by applying the probability distribution model determined in the initialization step around the positions of each neighboring node. The position of the node is estimated in a place where the node is likely to exist (S930). This location is sent to another node to estimate the location of the other node.

This position estimation is performed for nodes in a certain group, for example nodes in a cluster or at least neighboring nodes of each node, and then repeated as many times as the algorithm iterates. This iterative operation may be implemented by decreasing C by one after the initialization step S910 (S920) and performing the node position estimation step S930 until C = 0 (S940). The algorithm is repeated as many times as the number of cycles and then terminated (S950).

10 is a functional diagram of an embodiment of a distributed position recognition apparatus according to the present invention.

Referring to FIG. 10, the neighbor node search unit 1010 searches for nodes within a radio transmission range of a node for which a current position is to be determined. At this time, it is preferable to search for nodes within a wireless transmission range belonging to the same cluster formed by the cluster forming unit 1020. The probability distribution model selector 1030 selects a distribution probability of the node according to the distance. In this case, it is preferable to select a node distribution probability according to the number of nodes searched by the neighbor node search unit 1010. The probability distribution calculator 1040 calculates a probability distribution by applying a probability distribution model selected based on the positions of the nodes searched by the neighbor node searcher 1010. The position determiner 1050 estimates the region having the highest value among the calculated probability distributions as the position of the node to determine the current position.

The transmitter 1060 transmits the estimated position to the other nodes for use in estimating the positions of the other nodes (1060). As a result, the neighbor node searcher 1010 recognizes that the positions of the neighbor nodes have changed. Estimate the position of the node to determine the position again. This is preferably repeated a predetermined number of algorithm cycles.

Although preferred embodiments of the present invention have been described, the technical spirit of the present invention is not limited to the embodiments described herein and may be modified in other forms. Embodiment of the present invention is provided to those skilled in the art to more fully understand the present invention.

1 is a structural diagram illustrating one embodiment of the distribution of nodes in a network, in accordance with the present invention.

2 is a conceptual diagram illustrating an embodiment of a cluster formation process according to the present invention.

3 is a conceptual diagram illustrating an embodiment of a neighbor node search process according to the present invention.

4 is a structural diagram of an embodiment of a rectangular grid model for showing the distribution of neighboring nodes.

5 is a structural diagram of an embodiment of a triangular grid model for showing the distribution of neighboring nodes.

6 is a structural diagram of other embodiments of a triangular grid model to represent the distribution of neighboring nodes.

7 is a graph illustrating an embodiment of a distribution probability of a node over a distance according to the present invention.

8 is a graph illustrating another embodiment of a distribution probability of a node according to a distance according to the present invention.

9 is a flowchart illustrating an embodiment of a distributed location recognition method according to the present invention.

10 is a functional diagram of an embodiment of a distributed position recognition apparatus according to the present invention.

Claims (10)

In the location recognition method in a specific node of the wireless sensor network, (a) determining an initial position based on a location of a plurality of reference nodes whose location is known, network information of the reference nodes, and information of neighboring nodes located in a region capable of transmitting a signal directly wirelessly; (b) selecting a probability distribution model indicative of the probability that the nodes will be distributed over distance; (c) calculating a probability that a node is distributed in the region by applying the probability distribution model around the positions of the reference nodes; And and (d) correcting the initial position to one of the areas where the node is most likely to be distributed. 2. According to claim 1, wherein the step (a), (a1) determining an average distance from the neighboring nodes from the strength of the signal received from the neighboring nodes; (a2) determining the number of hops from each of the reference nodes and the distance from the reference nodes from the average distance; And (a3) determining the initial position by using the position of the reference nodes and the distance from the reference nodes; and a distributed position recognition method in a wireless sensor network. The method of claim 1, In the step (b), the probability distribution model is selected based on the number of neighboring nodes. The method of claim 1, Repeating steps (b), (c) and (d) a predetermined number of times; And Finally, determining the corrected position as the final position; distributed position recognition method in a wireless sensor network comprising a. The method of claim 1, And transmitting the corrected initial position. 17. The method of claim 1, further comprising transmitting the corrected initial position. An initial location determiner configured to determine an initial location based on locations of a plurality of reference nodes whose locations are known, network information of the reference nodes, and information of neighboring nodes located in an area where a direct signal can be transmitted wirelessly; A probability distribution model selection unit for selecting a probability distribution model indicating a probability that nodes are distributed according to a distance; A probability distribution calculator configured to calculate a probability that a node is distributed in the region by applying the probability distribution model around the positions of the reference nodes; And And a position corrector configured to correct the initial position to one of the regions where the node is most likely to be distributed. 2. The method of claim 6, wherein the initial positioning unit, A first measuring unit measuring an average distance from the neighboring nodes from the strength of the signal received from the neighboring nodes; A second measuring unit measuring a distance from the reference nodes from the number of hops from the respective reference nodes and the average distance; And And a determiner configured to determine the initial position by using the positions of the reference nodes and the distances from the reference nodes. The method of claim 6, The probability distribution model selecting unit selects the probability distribution model based on the number of neighboring nodes. The method of claim 6, wherein the position correction unit, A cycle count counter for incrementing the count by one each time the initial position is corrected; And And a final position determiner configured to determine the last corrected position as the final position when the count reaches a constant value. The method of claim 6, And a transmitter configured to transmit the corrected position.

Images