Improvement of bispectrum estimation based on one new 2-D combined lag window

Higher-order spectrum (HOS) has been used as a strong tool for non-Gaussian and nonlinear signal processing in many fields and bispectrum is usually employed. Using windows is one important way to improve bispectrum estimation and appropriate window function can reduce the variance and suppress noise, but sidelodes in spectrum of window functions result in spectrum leak. Thus it is one urgent problem needed to be solved for application of bispectrum that how to find one appropriate 2-D window. In this paper one new 2-D combined lag window is constructed by Hanning-Poisson window without sidelobes, which then is used for non-parametric bispectrum estimation instead of rectangle window. In the end simulation is done on QPC signals encountered in fault diagnosis of rotating machinery. The results demonstrate that using the constructed window function can restrain spectrum leak and reduce the variance of estimation better than single windows, like rectangle window.