Particle Methods for Indoor Tracking in WiFi Networks
(2013) FMS820 20131Mathematical Statistics
 Abstract (Swedish)
 This thesis treats the problem of positioning in WiFi networks and proposes a
solution using hidden Markov models and particle lters based on sequential importance
sampling with resampling. Hidden Markov models prove to be a powerful
framework for this type of problem exhibiting both an intuitive and adaptive model
structure. One of the diculties encountered when ltering this model is that it
is increasing in dimension over time. To get around this problem in this thesis, a
method called sequential importance sampling is used which makes it possible to
update the estimation sequentially using the previous estimate to calculate the new
one. In addition, parameter inference is conducted using expectationmaximizationbased
likelihood... (More)  This thesis treats the problem of positioning in WiFi networks and proposes a
solution using hidden Markov models and particle lters based on sequential importance
sampling with resampling. Hidden Markov models prove to be a powerful
framework for this type of problem exhibiting both an intuitive and adaptive model
structure. One of the diculties encountered when ltering this model is that it
is increasing in dimension over time. To get around this problem in this thesis, a
method called sequential importance sampling is used which makes it possible to
update the estimation sequentially using the previous estimate to calculate the new
one. In addition, parameter inference is conducted using expectationmaximizationbased
likelihood inference in the proposed model. To estimate model parameters
in a hidden Markov model smoothing is done to reconstruct the underlying state
sequence given all data observations. These states are then used to calculate suf
cient statistics for the sought parameters. Finding the smoothing distribution it
might be necessary to use a large number of particles in the lter to overcome degeneration
of the approximation. A method called xedlag smoothing is therefore
investigated in order to get better estimates. Finally we illustrate the model and
the estimation algorithms on data from a real world application. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/studentpapers/record/3813595
 author
 von Ehrenheim, Vilhelm
 supervisor

 Jimmy Olsson ^{LU}
 organization
 course
 FMS820 20131
 year
 2013
 type
 H2  Master's Degree (Two Years)
 subject
 language
 English
 id
 3813595
 date added to LUP
 20130614 16:19:46
 date last changed
 20130614 16:19:46
@misc{3813595, abstract = {This thesis treats the problem of positioning in WiFi networks and proposes a solution using hidden Markov models and particle lters based on sequential importance sampling with resampling. Hidden Markov models prove to be a powerful framework for this type of problem exhibiting both an intuitive and adaptive model structure. One of the diculties encountered when ltering this model is that it is increasing in dimension over time. To get around this problem in this thesis, a method called sequential importance sampling is used which makes it possible to update the estimation sequentially using the previous estimate to calculate the new one. In addition, parameter inference is conducted using expectationmaximizationbased likelihood inference in the proposed model. To estimate model parameters in a hidden Markov model smoothing is done to reconstruct the underlying state sequence given all data observations. These states are then used to calculate suf cient statistics for the sought parameters. Finding the smoothing distribution it might be necessary to use a large number of particles in the lter to overcome degeneration of the approximation. A method called xedlag smoothing is therefore investigated in order to get better estimates. Finally we illustrate the model and the estimation algorithms on data from a real world application.}, author = {von Ehrenheim, Vilhelm}, language = {eng}, note = {Student Paper}, title = {Particle Methods for Indoor Tracking in WiFi Networks}, year = {2013}, }