This paper presents an approach for low-cost simulation modeling for
application development for wireless sensor networks. Computational complexity
of simulating wireless sensor networks can be very high and as such must be
carefully managed. Application-level code prototyping with reasonable accuracy
and fidelity can be accomplished through simulation that models only the
effects of the wireless and distributed computations which materialize mainly
as delay and drop for the messages being exchanged among the motes. This
approach employs the abstraction that all physical or communication and
protocol level operations can be represented in terms of their effects as
message delay and drop at the application level for a wireless sensor network.
This study proposes that idea of empirical modeling of delay and drop and
employing those models to affect the reception times of wirelessly communicated
messages. It further proposes the delay and drop to be modeled as random
variables with probability distributions empirically approximated based on the
data reported in the literature. The proposed approach is demonstrated through
development of a neural network application with neurons distributed across the
motes of a wireless sensor network. Delay and drop are incorporated into
wireless communications, which carry neuron output values among motes. A set of
classification data sets from the Machine Learning Repository are employed to
demonstrate the performance of the proposed system in a comparative context
with the similar studies in the literature. Results and findings indicate that
the proposed approach of abstracting wireless sensor network operation in terms
of message delay and drop at the application level is feasible to facilitate
development of applications with competitive performance profiles while
minimizing the spatio-temporal cost of simulation.
[Journal_ref: ]
