Capacite et bornes de performances des canaux a entrees-sorties multiples et diversite spatio-temporelle

One of the main characteristics shared by all communication channels is the dependence between the maximum data rate they can support and the physical constraints to which they are subjected, the main constraints being of spatial, temporal and frequential order. A general model allowing for the simultaneous evaluation of the impact and respective interactions of these constraints is the multivariable gaussian channel where N inputs are transmitted through a multivariate linear channel having M outputs affected by colored gaussian noise sources.;This thesis proposes a method to evaluate the impact of spatial and temporal diversity and of frequency constraints on the admissible data rate in the context of linear equalization and matched filter bounds for the multivariable gaussian channel and to compare the corresponding results to the Shannon capacity of the channel. To achieve these objectives, we present the expression of an upper bound on the bit error rate for a linear equalizer using optimum combining. An explicit dependency of the performance bound on the system parameters is then obtained together with its relationship with the so-called “optimum” bounds corresponding to the matched filter and the information theoretic Shannon's capacity.;These analytical results are used to evaluate the performance of the digital radiomobile channel operating on a frequency selective channel linking N mobiles to an Mth order space diversity receiver. Because of the time varying nature of this channel, the bound is used in conjunction with a Monte-Carlo simulation to emulate the time varying realizations of the channel and evaluate the corresponding outage rate. Our results seem to indicate that, in general, equalization alone falls short of exploiting the full potential of the time diversity coding gain available with such time dispersive channels, which is particularly true for the linear equalizer. In this case however, our results tend to show that every transmitter bandwidth increment equal to the symbol transmission rate allows the interference suppression of one additional cochannel mobile, which is an extension of previous results to the frequency selective fading case.;The second application considered is the multivariable magnetic recording system where the N binary input data streams are longitudinally recorded by saturation over the corresponding number of tracks of a planar magnetic recording medium (e.g. magnetic tape, hard disk, etc.), the output streams being taken as the output of M magneto-inductive read heads. Our results tend to confirm the existence of optimal geometric configurations (heads-tracks) maximizing the capacity when the heads and tracks are sufficiently close to generate substantial intersymbol interferences from adjacent tracks. In the same operating conditions however, a significant performance degradation is seen for the linear equalizer and the matched filter bound. In this context, the interferences can be interpreted as the result of implicit redundancy within the channel which improve the value of the capacity but also shows the inability of the equalizers to exploit the available coding gain.