Sporadic machine-to-machine communication requires a signaling efficient medium access strategy, which can be achieved by compressed sensing based multi-user detection (CS-MUD). This novel application for Compressed Sensing requires adapted reconstruction algorithms considering typical communications assumptions. In order to get fast signal reconstruction, we utilize a smoothed version of the $\ell_{1}$-regularization, namely, the elastic-net. Further, we connect this approach with a final projection step to incorporate finite alphabets and introduce a new parameter choice strategy for the elastic-net: the so-called C-curve.