For fifteen years many studies have used MIMO systems to increase the Shannon capacity of the traditional SISO systems. To this end, a crucial problem is the design of transmitters which are optimal w.r.t. Shannon capacity, by the use of space-time codes or of prior knowledge on the transmission channel. These problems have been addressed by many studies in the case of frequency flat MIMO channels but are really less mature for frequency selective MIMO channels. This thesis focuses in the first part on the optimization, w.r.t. the ergodic capacity, of the covariance of the vector transmitted, via the Random Matrix Theory. Using multiple transmit antennas also gives rise to diversity, which improves the receiving performance. In the second part, we thus focus on the diversity, in the specific case of a MMSE receiver. Unlike the ML receiver, this receiver is suboptimal but very simple to implement. We first study the diversity at high SNR for frequency selective channels. We then focus on a diversity factor, the use of space-time codes in block (STBC), specifically the use of the Alamouti code. Thus, we propose and analyze in the multiuser context a new MMSE receiver robust to interference thanks to its ability to use optimally the degrees of freedom available in the channel
