A novel time-frequency technique for multicomponent signal denoising

Multicomponent signals, i.e. superpositions of modulated waves, arise in many physical or biological systems. Exploiting the particular structure of these signals, denoising methods based on time-frequency distributions often outperform standard techniques such as those based on diagonal estimation or sparsity approaches. Recently, a simple denoising technique based on local integration in scale of the wavelet transform was proposed. In spite of its behaviour being better compared to classical techniques for medium noise levels, it does not perform so well in other cases. We propose here a method to improve denoising behaviour based on a more accurate mode reconstruction technique. The method is detailed for time-frequency representation given by short-time Fourier and continuous wavelet transforms, with the emphasis placed on their differences.