You are on page 1of 3

Problem Statement

The requirements of a target tracking system continuous monitoring and trajectory estimation|
provide us space for enhancing energy efficiency. The typical sequence of operations of a target
tracking system is as follows: the tracking application first detects the target at anode (or a set of
nodes) and makes local measurements of the target's movement status.
Target properties, such as its latest position and velocity etc., are then estimated. These
properties are then reported to sink nodes in a timely manner. In this problem space, we
identify and focus on three directions for improving energy efficiency. These constitute the
dissertation problems and are discussed as follows.

1.1.Particle Filtering For trajectory estimation, target tracking is usually considered as a
dynamic state estimation problem, for which, particle filtering is one of the most widely used
Bayesian estimation methods that approximate the optimal solution . Particle filters are
sequential Monte Carlo methods that estimate nonlinear and/or non-Gaussian dynamic
processes. As a numerical method, the posterior probability density function (pdf) of
Bayesian estimation is represented with discrete samples (or particles) with associated
weights. Then, in each iteration, PFs draw particles from a proposal distribution (or
importance density), assign them with corresponding weights, normalize the weights,
possibly resample, and finally make the estimation as the weighted average of these particles.
Despite their attractive tracking performance and exibility , the application of PFs in WSNs
is very challenging, due to the conflict between the significant computational/communication
complexity of PFs and constrained resources of WSNs. The complexity of existing particle
lters is embodied in many aspects, among which three main problems are identied as
follows:
1) Number of particles. The complexity of particle filters is linear to the number of particles.
Though it is often necessary to draw a significant number of particles to achieve a suboptimal
posterior density, it will result in substantial computational and communication costs.


1.2. Sleep Scheduling: For target detection and measurement, idle listening is a major source of
energy waste . To reduce the energy consumption during idle listening so as to extend the
network lifetime, duty cycling is one of the most commonly used approaches. The idea of
duty cycling is to put nodes in the sleep state for most of the time, and only wake them up
periodically. In certain cases, the sleep pattern of nodes may also be explicitly scheduled, i.e.,
forced to sleep or awakened on demand. This is usually called sleep scheduling . Despite the
energy conservation, forcing nodes to sleep will probably result in target missing, thereby
impairing the tracking performance. As one of the sleep scheduling approaches, proactive
wake-up has been studied for awakening nodes proactively to prepare for the approaching
target . However, most existing efforts about proactive wake-up simply awaken all the
neighbor nodes in the area that the target is expected to arrive without any direrentiation . In
fact, it is sometimes unnecessary to awaken all the neighbor nodes. Based on target
prediction, it is possible to sleep-schedule nodes precisely, so as to reduce the energy
consumption for proactive wake-up. For example, if nodes know the exact route of a target, it
will be sufficient to awaken those nodes that cover the route during the time when the target
is expected to traverse their sensing areas.

2) Sample impoverishment problem. This problem is introduced by resampling, when particles
with high weights are selected many times and the diversity is lost . But existing solutions to this
problem are usually costly .
3) Centralization. For a centralized particle filter, weight normalization and resampling
require collection of data from multiple nodes to a single computational center (either a
cluster head [25, 26] or a global transceiver/sink node [27, 2). This will lead to significant
communication cost, yielding low energy efficiency. Besides, centralized particle lters also
introduce the following problems: a) the convergecast communication introduces a long
delay, as the computational center has to receive messages in a sequential order; and b) a
centralized implementation is vulnerable as a single point of failure [29].
The second dissertation problem that we study is to decrease the computational and com-
munication cost of particle filters, and improve their energy efficiency accordingly. The
dissertation addresses all the three aforementioned problems associated with PFs:
1) Based on vector space theory, we reduce the number of particles, i.e., reduce the problem
scale;
2) We present an efficient solution to the sample impoverishment problem; and
3) We design a completely distributed particle filter to simplify the solution process and
reduce the communication cost.

1.3. Constrained Flooding: For real-time data propagation, multipath forwarding is a commonly
used redundant copy approach [30], which bypasses network congestion and node/link
failures with multiple data copies transmitted along multiple paths. Many previous research
efforts [31{36] have studied the efficiency of multipath forwarding protocols toward
enhancing real-time performance while consuming optimized resources. Majority of these
efforts have focused on reducing the number of ooding recipients at each hop so that the
additional resource consumption can be minimized and QoS constraints such as real-time can
also be satisfied. But even though the number of recipients is reduced, the redundancy
introduced by this approach remains due to its probability-based recipient selection
mechanism. In fact, it is possible that network congestion or disconnection is not significant
enough throughout the entire path from source to sink to warrant multipath forwarding for
each hop. Therefore, it is sometimes unnecessary to transmit redundant data copies generated
by multipath forwarding protocols, which often consumes additional energy and bandwidth.

1.4. Prototype Implementation: Finally, we implement a prototype of the proposed algorithms
and protocols designed for the three dissertation problems, and evaluate their efficiency with
field experiments in a real deployment. The implementation was motivated by the following
three advantages of field experiments, compared to the simulation:

1) During the implementation, the rationality of the dissertation's algorithm/protocol design
can be verified.


Future Scope:

We find that an important research track in geographic and geometric-based schemes is handling
distances and nodes positions uncertainties in tree and/or cluster construction. Any error in
nodes coordinates may propagate during the computation process which may lead to wrong
results. We believe that taking into account these uncertainties may improve the decisions about
the target position predictions and the data reporting especially in multimedia WSN.


Any Technical query -9555476716(Ashish )

You might also like