Understanding atmospheric flows in urban areas is of primary importance in the context of urban densification of the worldwide population. While the flow inside an urban-type roughness has been described statistically (MacDonald 2000), its dynamics has never been fully addressed. From a purely aerodynamic point of view, the flow inside an urban canopy can be seen as the flow developing around three-dimensional obstacles (induced by the buildings geometry) immersed into a high Reynolds number, rough boundary layer flow (the atmospheric surface layer), with complex interactions (Blackman and Perret 2013).   Despite the very large range of building size, shape, or spatial arrangement in real-life built areas, the simplified canopy consisting of a regular array of cubes is now widely accepted as a canonical representation. Several contributions investigating the turbulent flow around a single wall-mounted cube can be found in the literature (Martinuzzi and Tropea, 1993 ; Sousa 2002) and are a good starting point to understand the dynamics inside such a canopy. Further insight is given by obstacle-resolving LES or DNS of turbulent boundary layers developing over an array of cubes (Coceal et al 2007 ; Anderson 2016) : the interaction of multiple cube wakes can be addressed, and the four-dimensional nature of numerical data is invaluable to study the dynamics of the flow. However, the limitation to moderate Reynolds numbers and small boundary layer thickness (?) to cube height (h) ratio does not allow a faithful reproduction of the very large scale motions originating from the outer flow. Wind-tunnel modeling offers a better physical representation of the different turbulent scales in an atmospheric boundary layer. Due to the tight clearance and high complexity of the flow, measurements within the canopy are not straightforward, and relatively little time-resolved experimental data can be found. In that perspective, the work from Castro's group (Castro et al 2006) using Laser Doppler Anemometry (LDA) is a major step forward to study the flow dynamics in this region.   In the present contribution, the dynamics of the flow inside an urban canopy consisting of a regular, staggered array of cubes with a 25% density is investigated using two unique experimental and numerical datasets. The experimental one has been obtained using 2C-LDA in the atmospheric wind-tunnel at LHEEA, Nantes, at a high Reynolds number (h+=1230 and ?+=24000), with a ratio ?/h=19.5 large enough for engineering applications. The dynamics of the flow is recorded at 16 positions within the canopy, distributed at three altitudes with respect to the cube height : z=0.25h, 0.5h and 1h, and at longitudinal/spanwise positions representative of the canopy pattern (P1, P2, P3 as defined by Castro) as well as additional points chosen to study the flow dynamics in the vicinity of a cube. The numerical dataset consists of an obstacle resolving LES simulation at h+=500 computed with the open-source software OpenFOAM. The characteristics of the simulation are similar to the DNS of Coceal et al (2007), and gives access to 3 dimensional data with a good spatial resolution of the flow around the building (?=h/32).   Using moments up to the 4th order, the statistical picture of the flow within the canopy is first analyzed and assessed against existing data in the literature. Taking advantage of the good temporal resolution of both experimental and numerical data, we then focus our analysis on the dynamics of the flow in the vicinity of a cube. A detailed temporal spectral analysis is carried out on the LDA and LES data to extract quantitative information; in particular, it is checked whether the vortex shedding typically observed on the sides of an isolated cube is still present inside the canopy.