Adaptive time-frequency decomposition with matching pursuits

An adaptive time-frequency decomposition is introduced. It represents signals as a linear expansion of local time-frequency waveforms, selected in order to match the signal structures. The decomposition is performed with an algorithm, called matching pursuit, whose convergence is guaranteed. A new definition of signal energy density in the time-frequency plane is derived by summing the Wigner distribution of each local time-frequency waveform. This energy density does not have any interference terms, unlike Wigner and Cohen class distributions. Numerical examples are described.<<ETX>>