In this paper, we propose two blind decoding approaches for multi-input single-output (MISO) communication systems. We first introduce a nonlinear precoding scheme that allows viewing the received signal as a Volterra-like model with the following properties: the input solely depends on the coding sequence, assumed to be known to the receiver, while the kernel is a multiway array depending on informative data and on the channel parameters. We show that such a kernel admits a PARAFAC tensor model. After estimating the kernel by using the coding sequence, the data symbols are then recovered. For this purpose, two methods are proposed. The first one directly computes the PARAFAC loading factors by means of an alternating least squares method. The second one solves the problem by means of a joint diagonalization of matrices constructed with the slices of the tensor. The performance of the proposed methods is evaluated by means of simulations.