14. Kernel Methods for RSS-Based Indoor Localization

  1. Seyed A. (Reza) Zekavat2 and
  2. R. Michael Buehrer3
  1. Piyush Agrawal and
  2. Neal Patwari

Published Online: 6 SEP 2011

DOI: 10.1002/9781118104750.ch14

Handbook of Position Location: Theory, Practice, and Advances

Handbook of Position Location: Theory, Practice, and Advances

How to Cite

Agrawal, P. and Patwari, N. (2011) Kernel Methods for RSS-Based Indoor Localization, in Handbook of Position Location: Theory, Practice, and Advances (eds S. A. (. Zekavat and R. M. Buehrer), John Wiley & Sons, Inc., Hoboken, NJ, USA. doi: 10.1002/9781118104750.ch14

Editor Information

  1. 2

    Michigan Technological University, Houghton, MI, USA

  2. 3

    Virginia Tech, Blacksburg, VA, USA

Author Information

  1. University of Utah, Salt Lake City, UT, USA

Publication History

  1. Published Online: 6 SEP 2011
  2. Published Print: 16 SEP 2011

ISBN Information

Print ISBN: 9780470943427

Online ISBN: 9781118104750



  • Gaussian kernel localization algorithm;
  • indoor localization;
  • kernel methods;
  • LANDMARC algorithm;
  • linear signal-distance map localization algorithm;
  • radial basis function-based localization algorithm;
  • received signal strength (RSS)


This chapter explores the features and advantages of kernel-based localization. Kernel methods simplify received signal strength (RSS)-based localization by providing a means to learn the complicated relationship between RSS measurement vector and position. The chapter discusses their use in self-calibrating indoor localization systems. It reviews four kernel-based localization algorithms, namely LANDMARC algorithm, Gaussian kernel localization algorithm, radial basis function-based localization algorithm and linear signal-distance map localization algorithm. The chapter presents a common framework for their comparison. It shows results from two simulations and from an extensive measurement data set, which provide a quantitative comparison and intuition into their differences. Results show that kernel methods can achieve a root mean square error (RMSE) up to 55% lower than a maximum likelihood estimator.

Controlled Vocabulary Terms

indoor communication; position measurement