The thickness of ice shelves, a basic parameter for mass balance estimates, is typically inferred using hydrostatic equilibrium, for which knowledge of the depth-averaged density is essential. The densification from snow to ice depends on a number of local factors (e.g., temperature and surface mass balance) causing spatial and temporal variations in density–depth profiles. However, direct measurements of firn density are sparse, requiring substantial logistical effort. Here, we infer density from radio-wave propagation speed using ground-based wide-angle radar data sets (10 MHz) collected at five sites on Roi Baudouin Ice Shelf (RBIS), Dronning Maud Land, Antarctica. We reconstruct depth to internal reflectors, local ice thickness, and firn-air content using a novel algorithm that includes traveltime inversion and ray tracing with a prescribed shape of the depth–density relationship. For the particular case of an ice-shelf channel, where ice thickness and surface slope change substantially over a few kilometers, the radar data suggest that firn inside the channel is about 5 % denser than outside the channel. Although this density difference is at the detection limit of the radar, it is consistent with a similar density anomaly reconstructed from optical televiewing, which reveals that the firn inside the channel is 4.7 % denser than that outside the channel. Hydrostatic ice thickness calculations used for determining basal melt rates should account for the denser firn in ice-shelf channels. The radar method presented here is robust and can easily be adapted to different radar frequencies and data-acquisition geometries.