Dans cette thèse, nous étudions les trois sujets suivants sur les systèmes de multiples entrées multiples sorties (MIMO) avec retour limité: Distorsion bout-à-bout: La première partie de la thèse présente l'impact conjoint des les nombres d'antenne, le ratio de bande passante de la source au canal, la corrélation spatiale et diversité dans le temps sur l'espérance de la distorsion optimale bout-à-bout dans les MIMO systèmes sans panne. En particulier, repose sur l'expression analytique pour tout les ratios du signal au bruit (SNR), l'expression asymptotique de l'espérance de la distorsion optimal bout-à-bout au SNR élevé est dérivé, composé de l'exposant de distorsion optimale et le facteur de distorsion optimale. Les résultats des simulations montrent que, à un SNR élevé pratique, l'analyse sur les impacts de l'exposant de distorsion optimale et le facteur de distorsion optimale explique le comportement de la distorsion optimale bout-à-bout. Les résultats présentés dans cette partie pourraient être les objectifs de performance pour les systèmes qui transmettent les sources analogique et en outre les directives sur la conception du système. Retour analogique du canal: Dans la seconde partie de cette thèse, nous proposons d'appliquer le codage bloc d'espace-temps orthogonal (OSTBC) sur retour analogique linéaire du canal. Considérant l'information du canal MIMO est une sorte de source vectorielle analogique, par rapport au retour du canal quantifié, le retour analogique linéaire a les avantages comme sans panne, l'adaptation automatique au canal et peu complexe. Il est prouvé que la méthode de transmission analogique linéaire OSTBC peutatteindre le borne de filtre adapté (MFB) sur SNR reçu. Par rapport en méthode analogique linéaire du codage bloc d'espace-temps circulant, la méthode linéaire analogique OSTBC obtient de meilleurs résultats à l'égard de SNR reçu et erreur quadratique moyenne (MSE) ; Par rapport en méthode de la quantification vectorielle aléatoire, les résultats des simulations montrent que sous une contrainte stricte de latence, leur performances sont proches à l'égard de l'erreur moyenne au sens et la méthode linéaire analogique OSTBC exécute encore mieux à l'égard erreur quadratique moyenne. Nous étudions également l'incidence de appliquer le retour linéaire analogique OSTBC au beamforming en liaison descendante pour multiutilisateurs MIMO et nous montrons que le retour linéaire analogique OSTBC peut rendre l'approche du système des performances optimales dans un temps de latence court. Multiplexage par couche: Dans la troisime partie de cette thèse, en considérant les systèmes avec des blocs courts, une nouvelle stratégie de multi-plexage par couches est proposée d'adapter un canal incertain par le codage couche-temps Walsh, l'annuleur d'interférence successif et HARQ signalisation. Comme l'illustrent les résultats de la simulation, en raison de ses performances proches mais beaucoup moins complexe, cette stratégie serait un bon substitut µa la stratégie de modulation QAM adaptive qui est largement utilise.

The paradigm of separate source-channel coding is inspired by Shannon's separation result, which implies the asymptotic optimality of designing source and channel coding independently from each other. The result exploits the fact that channel error probabilities can be made arbitrarily small, as long as the block length of the channel code can be made arbitrarily large. However, this is not possible in practice, where the block length is either fixed or restricted to a range of finite values. As a result, the optimality of source and channel coding separation becomes unknown, leading researchers to consider joint source-channel coding (JSCC) to further improve the performance of practical systems that must operate in the finite block length regime. With this motivation, this thesis investigates the application of JSCC principles for multimedia communications over point-to-point, broadcast, and relay channels. All analyses are conducted from the perspective of end-to-end distortion (EED) for results that are applicable to channel codes with finite block lengths in pursuing insights into practical design. The thesis first revisits the fundamental open problem of the separation of source and channel coding in the finite block length regime. Derived formulations and numerical analyses for a source-channel coding system reveal many scenarios where the EED reduction is positive when pairing the channel-optimized source quantizer (COSQ) with an optimal channel code, hence establishing the invalidity of the separation theorem in the finite block length regime. With this, further improvements to JSCC systems are considered by augmenting error detection codes with the COSQ. Closed-form EED expressions for such system are derived, from which necessary optimality conditions are identified and used in proposed algorithms for system design. Results for both the point-to-point and broadcast channels demonstrate significant reductions to the EED without sacrificing bandwidth when considering a tradeoff between quantization and error detection coding rates. Lastly, the JSCC system is considered under relay channels, for which a computable measure of the EED is derived for any relay channel conditions with nonzero channel error probabilities. To emphasize the importance of analyzing JSCC systems under finite block lengths, the large sub-optimality in performance is demonstrated when solving the power allocation configuration problem according to capacity-based formulations that disregard channel errors, as opposed to those based on the EED. Although this thesis only considers one JSCC setup of many, it is concluded that consideration of JSCC systems from a non-asymptotic perspective not only is more meaningful, but also reveals more relevant insight into practical system design. This thesis accomplishes such by maintaining the EED as a measure of system performance in each of the considered point-to-point, broadcast, and relay cases.

The paradigm of separate source-channel coding is inspired by Shannon's separation result, which implies the asymptotic optimality of designing source and channel coding independently from each other. The result exploits the fact that channel error probabilities can be made arbitrarily small, as long as the block length of the channel code can be made arbitrarily large. However, this is not possible in practice, where the block length is either fixed or restricted to a range of finite values. As a result, the optimality of source and channel coding separation becomes unknown, leading researchers to consider joint source-channel coding (JSCC) to further improve the performance of practical systems that must operate in the finite block length regime. With this motivation, this thesis investigates the application of JSCC principles for multimedia communications over point-to-point, broadcast, and relay channels. All analyses are conducted from the perspective of end-to-end distortion (EED) for results that are applicable to channel codes with finite block lengths in pursuing insights into practical design. The thesis first revisits the fundamental open problem of the separation of source and channel coding in the finite block length regime. Derived formulations and numerical analyses for a source-channel coding system reveal many scenarios where the EED reduction is positive when pairing the channel-optimized source quantizer (COSQ) with an optimal channel code, hence establishing the invalidity of the separation theorem in the finite block length regime. With this, further improvements to JSCC systems are considered by augmenting error detection codes with the COSQ. Closed-form EED expressions for such system are derived, from which necessary optimality conditions are identified and used in proposed algorithms for system design. Results for both the point-to-point and broadcast channels demonstrate significant reductions to the EED without sacrificing bandwidth when considering a tradeoff between quantization and error detection coding rates. Lastly, the JSCC system is considered under relay channels, for which a computable measure of the EED is derived for any relay channel conditions with nonzero channel error probabilities. To emphasize the importance of analyzing JSCC systems under finite block lengths, the large sub-optimality in performance is demonstrated when solving the power allocation configuration problem according to capacity-based formulations that disregard channel errors, as opposed to those based on the EED. Although this thesis only considers one JSCC setup of many, it is concluded that consideration of JSCC systems from a non-asymptotic perspective not only is more meaningful, but also reveals more relevant insight into practical system design. This thesis accomplishes such by maintaining the EED as a measure of system performance in each of the considered point-to-point, broadcast, and relay cases.