Practical Issues Concerning the Use of Bicoherence for the detection of quadratic phase coupling

It is often diicult to apply the techniques of Higher Order Statistics (HOS) in practice, since many of the techniques require unrealistic signal assumptions to hold. In this paper a set of realistic restrictions on the experimental data is stipulated at the outset, and the use of the bispectrum as a detector of Quadratic Phase Coupling (QPC) is investigated. It is found that under these restrictions the squared bicoherence is not a good measure of QPC, since it shows peaks even if there is no QPC present. A new visualisa-tion of the problem is presented in the complex bispectral plane, and a new QPC detector is proposed based on the joint conditions of signiicant bicoherence magnitude and zero biphase. This detector, which is based on the empirical results of Sebert and Elgar 1989] performs well over a variety of SNR's, although some problems still persist. The main points of the paper are illustrated with simulated data and experimental speech data.