Modern telecommunication systems offer services demanding very high transmission rates. Channel identification appears as a major concern in this context. Looking forward better trade-offs between the quality of information recovery and suitable bit-rates, the use of blind techniques is of great interest. Making use of the special properties of the 4th-order output cumulants, this thesis introduces new statistical signal processing tools with applications in radio-mobile communication systems. Exploiting the highly symmetrical structure of the output cumulants, we address the blind channel identification problem by introducing a multilinear model for the 4th-order output cumulant tensor, based on the Parallel Factor (Parafac) analysis. The components of the new tensor model have a Hankel structure, in the SISO case. For (memoryless) MIMO channels, redundant tensor factors are exploited in the estimation of the channel coefficients. In this context, we develop blind identification algorithms based on a single-step least squares (SSLS) minimization problem. The proposed methods fully exploit the multilinear structure of the cumulant tensor as well as its symmetries and redundancies, thus enabling us to avoid any kind of pre-processing. Indeed, the SS-LS approach induces a solution based on a sole optimisation procedure, without intermediate stages, contrary to the vast majority of methods found in the literature. Using only the 4th-order cumulants, and exploiting the Virtual Array concept, we treat the source localization problem in multiuser sensor array processing. Exploiting a particular arrangement of the cumulant tensor, an original contribution consists in providing additional virtual sensors by improving the array resolution by means of an enhanced array structure that commonly arises when using 6th-order statistics. We also consider the problem of estimating the physical parameters of a multipath MIMO communication channel. Using a fully blind approach, we first treat the multipath channel as a convolutive MIMO model and propose a new technique to estimate its coefficients. This non-parametric technique generalizes the methods formerly proposed for the SISO and (memoryless) MIMO cases. Using a tensor formalism to represent the multipath MIMO channel, we estimate the physical multipath parameters by means of a combined ALS-MUSIC technique based on subspace algorithms. Finally, we turn our attention to the problem of determining the order of FIR channels in the context of MISO systems. We introduce a complete combined procedure for the detection and estimation of frequency-selective MISO communication channels. The new algorithm successively detects the signal sources, determines the order of their individual transmission channels and estimates the associated channel coefficients using a deflationary approach.