Research Description

There is a gap between theoretical achievements and high-speed hardware implementation in the field of positioning, tracking and navigation. Also, there is a need for the improved positioning, tracking and navigation solutions in many applications including aircraft positioning, target tracking for radar and sonar applications, car collision detection, and positioning and tracking in homeland security. Recently, a lot of research has been performed on the algorithms that can handle various tracking models using the same approach and that have better performance than the traditional algorithms. These algorithms are called particle filtering and they rely on Bayesian signal processing and Monte-Carlo sampling and as such, are very computationally intensive. Particle filters have not been used in commercial and military applications mainly because of their high computational complexity. Current work in this project is related to developing and simulating algorithms for different tracking and navigation applications in order to define a library of functions and operations needed for these applications. Papers published as a result of this work are shown below.

Particle filtering algorithms require floating-point arithmetic and they are much more complex than traditional signal processing algorithms. For example, in order to process one input sample for the simple two-dimensional bearing-only tracking problem, several thousands of exponential and arctan functions have to be executed. Parallel particle filters have been simulated in systemsC (2), but they have not been synthesized and implemented. We would like to extend our work to modeling, designing and implementation of domain-specific architectures for tracking and navigations. This is a 4-year project funded by Canada's NSERC Discovery Grant. We believe that multiprocessing architecture based on configurable processors will be the right approach to tracking the real-time implementation of particle filters and to allow for the flexibility of changing parameters and operational modes of the filters. Hence, we will try to develop an architecture based on Tensilica processors for tracking applications.

References and Published Papers:

"Simplifying Physical Realization of Gaussian Particle Filters with Block Level Pipeline Control"
S. Hong, J. K. Lee, M. Bolic, P. M. Djuric, HYPERLINK
EURASIP Journal of Applied Signal Processing, No. 4, pp. 575-587, 2004.

http://scholar.google.com/scholar?q=EURASIP+Journal+of+Applied+Signal+Processing,+No.+4,+pp.+575-587,+2004.&hl=en&um=1&oi=scholart

"Resampling Algorithms and Architectures for Distributed Particle Filters"
M. Bolic, P. M. Djuric, S. Hong, HYPERLINK
IEEE Transactions on Signal Processing, 2004

http://www.site.uottawa.ca/~mbolic/Miodrag_Bolic_files/published/bolic_T-SP-2004.pdf

"Resampling Algorithms for Particle Filters: A computational Complexity Perspective"
S. Hong, M. Bolic & P. M. Djuric, HYPERLINK
EURASIP Journal of Applied Signal Processing, 2004

http://www.site.uottawa.ca/~mbolic/Miodrag_Bolic_files/published/bolic_jasp_2004.pd

"An Efficient Fixed-Point Implementation of Residual Systematic Resampling Scheme for High-Speed Particle Filters"
S. Hong, M. Bolic, & P. M. Djuric, HYPERLINK
IEEE Signal Processing Letters, vol. 11, No. 5, May 2004

http://www.site.uottawa.ca/~mbolic/Miodrag_Bolic_files/published/bolic_SPL_2004.pdf

Submitted Papers

"A Study of Gaussian Particle Filters for Practical Physical Implementation"
M. Bolic, A. Athalye, P. M. Djuric, & S. Hong
Submitted to IEEE Transactions on Circuits and Systems I, 2004

"Design and Implementation of flexible Resampling Mechanism for High-Speed Parallel Particle Filters"
S. Hong, S. S. Chin, M. Bolic, & P. M. Djuric
Submitted to Journal of VLSI Signal Processing, 2003

University of Ottawa's URL link:

http://www.uottawa.ca/welcome.html

University of Ottawa's School of Information Technology and Engineering URL link:

http://www.site.uottawa.ca/eng/