This paper focuses on random uplink transmissions of a subset of nodes disseminated in a cell. Under the constraints of massive Machine Type Communication (mMTC) in cellular Low Power Wide Area Networks (LPWAN) and Ultra Reliable Low Latency Communications (URLLC), we assume a highly restricted coordination with the receiver and the usage of coded Non Orthogonal Multiple Access (NOMA). We then target direct data transmission and propose an optimal detector of the active users with channel state information at the receiver limited to statistical knowledge. This algorithm implements a Maximum Likelihood (ML) detector. We give the formulation of the optimal detector and we evaluate its performance, with different codelengths and for various number of base station antennas.