Signal Decomposition with Discontinuous and Continuous Bases

Signal decomposition is usually used as a preprocessing step in signal processing. After decomposing an original signal, we can process each decomposed bases individually. Different decomposition methods result in different bases. Less number of bases could reduce the complexity of further processing. This paper shows a new method, which decomposes a signal into a mixture of continuous and discontinuous bases. The method decomposes signals into continuous and discontinuous bases separately first, and then select most significant bases by using sparse Bayesian learning. Our experiment results show that it can reduce the number of bases and improve the accuracy