Frequency Separation Method Based On Sparse Coding

The paper presents a sparse coding method that models a signal with amplitude modulation (AM) and frequency modulation (FM) functions. Indeed, the proposed sparse coding frequency separation (SCFS) approach is based on a mutlicomponent AM-FM modeling where each monocomponent counterpart is obtained by sparse coding using orthogonal matching pursuits, and sorted from fine to coarse depending on its frequency content. SCFS appears to be an efficient tool to properly separate the frequency content of signals, and behaves like the empirical mode decomposition. Results show neat improvements in terms of frequency separation, tone separation capability, and robustness against noise.