A technique to produce images from heterogeneous structures or objects buried in a medium - such as ground or walls - relies on the measurement of diffracted field from inhomogeneities under incident electromagnetic illumination. It requires the operation of a microwave acquisition system with a linear or planar sensor array. Its conception criteria are dictated by imaging algorithms which need, most of the time, multifrequency datas upon large observation window. This thesis work lies in the characterization, calibration and the experimental validation of an ultra wideband multistatic imaging prototype. A description of current microwave imaging systems has firstly fulfilled the initial objectives upon chosen criteria and has presented considerations regarding subsurface detection problem. Then the SIMIS (french acronym of Synthetic Impulse Microwave Imaging System), developped in LEAT, is characterized by its theoretical performances, with a specific attention towards dynamic range optimization. In order to correct systematic errors, the radar calibration is achieved on each module successively. Several studies on drift and random errors are also proposed. Next, the system is operated in anechoid chamber to detect canonical targets - a metallic cylinder and a dielectric parallelepiped - in a 2D-TM configuration. Presence of these two targets is confirmed from the observation of diffraction hyperbolae on B-scan images. A modification of the system is conducted and validated to reduce coupling in the array between antennas. Therefore, if needed, detection is achieved only with the total field measurement. First results in qualitative imaging are obtained by the time reversal mirror technique. In frequency domain, the D.O.R.T. method (Decomposition of the Time Reversal Operator) performs detection of targets when the wavelength is comparable to their tranversal dimension.
