Microwave imaging in cluttered media with an Ultrawideband Time Reversal-based prototype
Résumé
In any application where one wants to reconstruct the electromagnetic properties of objects embed- ded in a cluttered medium [1, 2], it is mandatory to improve the signal-to-clutter ratio as much as possible before processing the scattered field data. If this is not done, non-linear inverse algorithms - which are known to be very sensitive to noise and clutter - are more easily trapped into local minima of the cost function, giving insatisfactory results. One way to achieve this task is to illuminate the scene with a wave focusing onto the relevant targets, so that the scattered field coming from the surrounding cluttered medium is reduced. One of the Time Reversal-issued methods known as DORT method [3] allows to easily generate such waves provided the time-harmonic multistatic data matrix associated to an array of antennas is measured. In fact, via a simple Singular Value Decomposition of the matrix, it has been shown [4] that the resulting singular vector(s) contain the complex amplitude law(s) needed to properly "bias" the array antennas. The idea is then to include the response to such DORT wave(s) within the inversion process in order to increase the robustness of the algorithm with respect to clutter. To prove and explore this concept, we have built a microwave-range RADAR prototype working in the [2-4] GHz band with an array of 8 ultrawideband linearly polarized Exponentially Tapered Slot Antennas (ETSA) [5]. Since each antenna channel is equipped with a couple of digitally controlled attenuator and phase shifter, we can experimentally build and transmit the focusing wave by coding the complex amplitudes of the DORT singular vectors into the prototype. Before inversing the 2D scattered field data, a fine antenna calibration procedure based on the same experimental data is run in order to properly model the incident field of the antennas. This step is delicate since a wrong or imprecise model could undermine the inversion results, and we solve it by applying a multipolar expansion of the incident field [6] combined with a separation between transmitting and receiving antenna patterns. Finally, the non-linear, iterative Modified2 Gradient Algorithm [7] is run to retrieve the features of the targets. It is here that we introduce DORT data as a regularization term of the "traditional" cost function. At the conference we will show experimental data and results proving the whole concept.