Les tĂ©lĂ©communications sans fil ont connu ces dernières annĂ©es un immense succès Ă  tel point que le spectre des frĂ©quences est dĂ©sormais surchargĂ© et nĂ©cessite la disponibilitĂ© de nouvelles ressources. Pour rĂ©pondre Ă  ce besoin, des techniques de rĂ©utilisation dynamique du spectre ont alors vu le jour sous la dĂ©nomination de radio cognitive. Elles consistent Ă  partager de manière opportuniste et efficace certaines frĂ©quences ayant Ă©tĂ© initialement allouĂ©es Ă  d’autres systèmes. Cette thèse se place dans le contexte de rĂ©seaux sans fil tactiques hĂ©tĂ©rogènes comportant des segments de radios cognitives. La difficultĂ© provient alors de la garantie de qualitĂ© de service de bout en bout : respect du dĂ©bit nĂ©gociĂ©, du dĂ©lai et de la gigue. Nous nous sommes tout d’abord intĂ©ressĂ©s au contrĂ´le d’admission dans ce type de rĂ©seaux en proposant une mĂ©thode de calcul de bande passante rĂ©siduelle de bout en bout s’appuyant sur un algorithme de complexitĂ© polynomiale et pouvant ĂŞtre implantĂ© de manière distribuĂ©e. Nous nous sommes ensuite concentrĂ©s sur le routage en proposant une nouvelle mĂ©trique tenant compte des particularitĂ©s de ce type de rĂ©seaux. Enfin, nous nous focalisons sur la thĂ©matique du routage Ă  contraintes multiples en Ă©tudiant et implantant en environnement rĂ©el des algorithmes d’approximation proposĂ©s dans la littĂ©rature. ABSTRACT : The unprecedented success of wireless telecommunication systems has resulted in the wireless spectrum becoming a scarce resource. Cognitive Radio systems have been proposed as the enabling technology allowing unlicensed equipments to opportunistically access the licensed spectrum when not in use by the licensed users. The focus of this thesis is on heterogeneous tactical networks deploying cognitive radios in parts or in their entirety. Such networks can be organized in multiple sub-networks, each characterized by a specific topology, medium access scheme and spectrum access policy. As a result, providing end-to-end Quality of Service guarantees in terms of bandwidth, delay and jitter, emerges as a key challenge. We first address the admission control in multi-hop cognitive radio networks and propose a polynomial time algorithm that can be implemented in a distributed fashion for estimating the end-to-end bandwidth. Then, we focus on routing and propose a new metric that takes into account the specifics of such networks. Finally, as quality of service requirements can be expressed using multiple metrics, we turn our attention to multi-constrained routing and implement on a real testbed low complexity approximation algorithms.