A signal detection and classification scheme based on Atomic Decomposition (AD) is presented that allows the detection and modulation analysis of simultaneous signals with good performance at low SNR. An AD-based detector is used to detect the components of the signal acting as the feature extractor and its performance is compared to Fourier-based detectors. The classification stage consists of signal separation by clustering, reconstruction of the signal instantaneous frequency using the previous AD and modulation analysis. The system is applied to the identification of radar signals.