Open AccessA Two-User Variational Mode Decomposition Algorithm for Blind Source Separation of Arbitrary Signals
Author(s) -
Seema Sud
Publication year - 2020
Publication title -
american journal of science and engineering
Language(s) - English
Resource type - Journals
eISSN - 2687-9581
pISSN - 2687-9530
DOI - 10.15864/ajse.1301
Subject(s) - blind signal separation , algorithm , computer science , hilbert–huang transform , mode (computer interface) , source separation , signal (programming language) , signal processing , decomposition , digital signal processing , filter (signal processing) , channel (broadcasting) , telecommunications , computer vision , computer hardware , programming language , operating system , ecology , biology
Blind source separation (BSS) is a problem wherein two unknown source signals that have been combined in some fashion when collected at a single receiver must be separated. This problem is exacerbated by source signals that are time-varying, i.e. non-stationary, in nature. All commonly used techniques, such as Fourier analysis, wavelet analysis, and adaptive filtering algorithms do not apply. First, there is no training data, the data may not be digital in nature but have a random, time-varying amplitude, and there are not enough samples to estimate the signal due to the non-stationarity. A novel technique known as empirical mode decomposition (EMD) overcomes many of the issues related to non-stationarity, but suffers in noisy channels. More recently, a technique known as variational mode decomposition (VMD) was introduced that overcomes the limitations of the other techniques to reconstruct unknown, non-stationary signals; this is termed the single user (SU) VMD algorithm. In this paper, we describe a two-user VMD algorithm that improves performance over the SU algorithm by up to an order of magnitude for the problem of BSS. The two-user VMD only assumes that an estimate of the power levels of the two users may be obtained. Performance is most improved when the powers of the two signals are close to equal.