Solutions for wireless sensor network localization

Abstract

Wireless sensor network localization opens the door to many location based applications. In

this thesis, some solutions obtained from localization algorithms are investigated. There are

two categories of problem on localization. Range-based methods are applied to the situation

in which information on the distances between each pair of nodes is available. Algorithms are

developed to estimate the location of each sensor in the network. Usually, the distance

between each pair of nodes is estimated by the signal strength received between them, and

this information is very noisy. Range-free methods, which are also called connectivity-based

methods, assume that the distances between any two nodes are unknown but the connectivity

information between them is known. If the distance between any two nodes in the network is

within a communication range, connectivity between these two nodes is said to be established.

In a range-based scenario, with the information of inter-sensor distance measurements as

well as the absolute locations of the anchors, the objective is to obtain the location of all the

unknown nodes. Two new localization methods based on gradient descent are shown in the

thesis. The gradient descent methods would minimize the difference between the measured

distances and the distances obtained from the estimated locations. From a comparison with

other well-known localization methods, the two newly developed gradient descent algorithms

can reach better accuracy at the expense of computational complexity. This is not surprising

as the proposed algorithms are iterative in nature.

For range-free scenario, a new model utilizing all the information derived from

connectivity-based sensor network localization is introduced. Unlike other algorithms which

only utilize the information on connections, this model makes use of both information on

connections and disconnections between any pair of nodes. The connectivity information

between any pair of nodes is modeled as convex and non-convex constraints. The localization

problem is solved by an optimization algorithm to obtain a solution that would satisfy all the

constraints established in the problem. The simulation has shown that better accuracy is

obtained when compared with algorithms developed by other researchers.

Another solution for the range-free scenario is obtained with the use of a two-objective

evolutionary algorithm called Pareto Archived Evolution Strategy (PAES). In an evolutionary

algorithm, the aim is to search for a solution that would satisfy all the convex and non-convex

constraints of the problem. The number of wrong connections and the summation of

corresponding distances are set as the two objectives. A starting point on the location of the

unknown nodes is obtained using a solution from the result of all convex constraints. The

final solution can reach the most suitable configuration of the unknown nodes as all the

information on the constraints (convex and non-convex) related to connectivity have been

used. From the simulation results, a relationship between the communication range and

accuracy is obtained.

In this thesis, another evolutionary algorithm has been examined to obtain a solution for

our problem. The solution is based on a modified differential evolution algorithm with

heuristic procedures peculiar to our domain of application. The characteristics of the sensor

network localization are thoroughly investigated and utilized to produce corresponding

treatment to search for the reasonable node locations. The modified differential evolution

algorithm uses a new crossover step that is based on the characteristics of the problem. With

the combination of some heuristics, the solution search can move the node to jump out of

local minimums more easily, and give better accuracy than current algorithms.

In the last part of the thesis, a novel two-level range connectivity-based sensor network

localization problem is proposed, which would enrich the connectivity information. In this

new problem, the information of the connectivity between any pair of nodes is either strong,

weak or zero. Again, a two-objective evolutionary algorithm is used to search for a solution

that would satisfy all the convex and non-convex constraints of the problem. Based on

simulations on a range of situations, a suitable range value for the second range is found.