Signal detection and estimation using atomic decomposition and information-theoretic criteria

An algorithm to detect and estimate a linear mixture of signals corrupted by white Gaussian noise is presented. The number of signals is assumed to be unknown. The algorithm is based on a type of information theoretic criterion, capable of adjusting the probability of false alarm, and uses atomic decomposition refined by the expectation maximization method to efficiently compute the maximum likelihood estimate of the signal parameters. Signals are modeled as chirplets. The algorithm consistency and efficiency are shown by simulation.