Accurate Noise Floor Calibration based on Modified Expectation Maximisation of Gaussian Mixture

An accurate estimation of the noise floor is of paramount importance for an optimum performance of spectrum sensing in Cognitive Radio (CR). The most common approach followed by existing noise floor estimation methods is to attempt to isolate a set of noise-only samples based on a given energy/power threshold. However, this approach is unreliable and in general unable to provide accurate estimations of the noise floor, in particular under low SNR conditions where the power of the Primary User (PU) signal is comparable to the noise floor of the CR device. In this context, this work considers a different approach where the power observed by the CR device is modelled as a Gaussian mixture. Based on a mathematical analysis of the relation among the parameters of the obtained Gaussian mixture, a modified version of the well-known Expectation Maximisation (EM) algorithm is proposed to fit the Gaussian mixture to the observed power values and provide an estimation of the noise floor, something that the general EM algorithm fails to achieve in this scenario. The obtained results demonstrate that the proposed method provides a highly accurate estimation of the noise floor in the presence of PU signals over the whole range of SNR values.

Authors:
López-Benítez Miguel, Lehtomäki Janne, Umebayashi Kenta, Patel Dhaval

Publication type:
A4 Article in conference proceedings

Place of publication:
2019 IEEE Wireless Communications and Networking Conference (WCNC), 15-18 April 2019, Marrakesh, Morocco

Keywords:
6G Publication

Published:
31 October 2019

Full citation:
M. López-Benítez, J. Lehtomäki, K. Umebayashi and D. Patel, “Accurate Noise Floor Calibration based on Modified Expectation Maximisation of Gaussian Mixture,” 2019 IEEE Wireless Communications and Networking Conference (WCNC), Marrakesh, Morocco, 2019, pp. 1-6. doi: 10.1109/WCNC.2019.8885661

DOI:
https://doi.org/10.1109/WCNC.2019.8885661

Read the publication here:
http://urn.fi/urn:nbn:fi-fe2019121949044