Urban noise reduction is a societal priority. In this context, the European Directive 2002/49/EC aims at producing strategic noise maps for large cities. However, nowadays the relevance of such maps is questionable, due to considerable uncertainties, which are not quantified. Conversely, the development of noise observatories provides a more realistic description of the sound environment, but at the expense of insufficient spatial resolution and high costs. Thus, the CENSE project aims at proposing a new methodology for the production of more realistic noise maps, based on an assimilation of simulated and measured data collected through new technologies. In the framework of thisproject, a specific measurement network will be deployed for several years in a dense urban city center in France. It is composed of 150 low-cost noise sensors based on MEMS technology and Raspberry Pi platforms or ARM microcontrollers, connected to the cloud through a mixed wired/wireless network and a public street lamp network (as a power-line communication system). Uncertainties of measurement devices have been estimated through laboratory tests (directivity, sensibility to temperature and hygrometry ...), as well as tests through free field conditions. This communication present the principle, the technology used, and the technical performances of the noise sensors.