Cognitive wireless mesh networks are an emerging technology and they are at- tracting an always growing community of researchers thanks to the ability of creating and extending pervasive communication applications to cognitive envi- ronments considering decentralized models. The major challenge in cognitive networks is the adaptation to time and space variability of the available resources, i.e. chunks of the frequency spec- trum called channels. In particular, this problem is exacerbated in cognitive mesh networks because there is no direct communication among devices and hence they cannot establish a global common control channel to coordinate the entire network. Instead, local control channels, that vary depending on time instant and location, can be established to coordinate cognitive devices among themselves. In this paper, challenges and approaches proposed in the literature to ad- dress the absence of a static and global control channel are analyzed and a Control channel formation protocol (Connor) is proposed. Connor is a fully distributed coordination scheme where cognitive mesh devices self-organize into clusters based on similarity of available channels and on topological constraints. Compared with the existing clustering algorithms in the literature, which re- quires synchronization, Connor performs better in most cases without imposing synchronization.