Positioning and Tracking in Asynchronous Wireless Sensor Networks
Licentiate thesis, 2005
This thesis deals with the problem of locating mobile nodes in an
asynchronous wireless communication network, i.e., a wireless network of mobile
terminals where some or all nodes do not have access to a global time reference.
A strong focus lies on reducing the complexity associated with straight forward
classical algorithms of node coordinate estimation. The thesis is based on a
number of previously published papers, listed in the introductory chapter.
Two complexity reducing data preprocessing methods are presented. Both
preprocessors achieve a complexity reduction through a cancellation of unknown
clock-offsets from the estimation problem. Based on a concept of invariant
preprocessors, we show how the individual unknown clock-offsets at some or all
nodes in the network may be discarded from the estimation problem, without any
degradation of the asymptotic performance bounds of the positioning problem.
We further present two, fully distributed, sub-optimal positioning algorithms
that operate on a set of asynchronous delay measurements. The first, called the
kernel algorithm, reduces complexity by a divide and conquer approach. The
second algorithm is based on a mechanical analogy of the positioning problem. We
evaluate the performance of both algorithms, in terms of the mean-squared
positioning error, by computer simulation. The performance of the kernel
algorithm is found to lie on the order of the delay measurement accuracy, while
the second algorithm is shown to attain the Cramér-Rao lower bound under a set
of reasonable assumptions.
In the last part of the thesis, a novel tracking filter is proposed to reduce
the complexity associated with tracking maneuvering objects in a wireless
network. The tracking filter is based on a classical Kalman filter, but uses
additional information, supplied by the tracked node, to aid in the tracking
process. One drawback associated with this type of approach to tracking is the
possibility of an unstable filter. We argue that the implementation can be made
robust using very simple alterations. Further, we argue that the classical
mean-squared-error performance measure is not fully appropriate for delay
sensitive applications, and introduce a novel performance measure called the
time margin measure, suitable for evaluation of tracking algorithms that operate
under latency constraints. We discuss the merits of our proposed tracking
filter, with respect to this new performance measure, as compared to a classical
Kalman implementation.