NDA maximum-likelihood waveform identification by model selection in digital modulations

In this paper, the problem of blind waveform identification of overlapped replicas is addressed. The proposed method departs from the unconditional maximum likelihood (UML) criterion and it makes use of the information regarding the signal subspace decomposition of the received signal. For the low-SNR regime, the paper shows that the UML criterion can be understood as a correlation matching approach in the transformed domain of the signal subspace. In addition, it is found that the initial set of unknowns is compressed into an smaller number of unknowns in this transformed domain. As a consequence, the solution space is reduced and the overall stability of the identification method is improved in front of the noise and possible ill-conditioned scenarios.