【正文】 microwave GaAs switches were used (MACOM SW479) [9] for a total of 4 different configurations。 these ponents had previously been measured and deembedded from the electrical connexions. In spite of their relatively large size, they were chosen for their ease of use and inclusion of all the required AC/DC decoupling circuits. The characteristics of these switches (both return loss and insertion loss) are as follows (Fig. 1). B. Description of the antenna and its parametersThe antenna is posed of a ground plane (l_gnd x w_gnd)on top of which is a radiating patch (l_patch x w_patch) located h_patch high. The patch is positioned atop the ground plane at (x_patch, y_patch). The antenna is fed through an SMA connector, located at (x_feed, y_feed) on the ground plane. Two electrical vias connect the radiating patch to two switches, located on a dielectric substrate on the other side of the ground plane. The position of the switches is given, respectively by (x_sw1, y_sw1) and (x_sw2, y_sw2). Thedielectric is mm thick, the copper layer on it is mm thick and its relative permittivity is =.III. METHODOLOGY OF OPTIMIZATION AND RESULTS? Methodology of simulation The antenna was simulated as a 5port device using CST[10]. The first port is the actual coaxial feed, while on the 4 remaining ports are plugged the two microwave switches. The whole antenna+switches assembly is simulated with ADS [11],which is also used for its builtin optimization algorithms and its ability to interface with CST. After a set of parametric sweeps, some geometrical parameters were picked, which would serve as the input variables in our optimization algorithm. The size of the ground plane is constant at 40 x 60 mm, as well as the height of the radiating patch at 5 mm. The varying parameters are the positions of the switches on the ground plane (x_sw1, x_sw2, y_sw1, y_sw2), the position of the feed (x_feed, y_feed) and the size and position of the radiation patch (x_patch, y_patch, w_patch, l_patch). That accounts for a total of 10 variables, which are also constrained together (all the shorts must connect to the patch, the switches and the feed must be on the ground plane and should not overlap).B. Objectives and Goal functionsAs previously mentioned, two switches account for a total of four available configurations. Our objective is that these configurations are as separate from each other as possible: each should match a specific frequency band, while rejecting the others. Following some manual tweaking of the geometric parameters to optimize, each configuration was assigned a frequency band. These different objectives are stated in table I. The error function that we used in our optimization process is a linear bination of eight objective functions, two for each configuration. To express matching, we use function (1), which equals zero when the antenna is matched below 10 dB in the frequency band of interest: that is to say that better matching will not reduce the error. (1)Similarly, to express rejection on the other frequencies, we use function (2) that equals zero when the signal is rejected above 2 dB. (2)Finally, to make up for the fact that there are more frequency points to reject than to match, and to also prioritize matching in the band of interest over rejection, we add weight to the adaptation objective. The eight objective functions are then summed, which leads to the final error function (3). (3)ADS is a monoobjective optimization tool, therefore careful thought must be put on the weighting of the different objectives.C. Optimization methodology and resulting antenna propertiesThe optimization process was twostep: the first step was tosimulate a high number of randomly generated antennas(approximately 1000), and pick the one with the lowest errorfunction. This antenna was then further optimized using agradient algorithm [12]. The obtained parameters are given intable II.The realized matching bands are presented in table II, and the return loss curves are as follows.As can be seen, all objectives are not fulfilled。 it is important to notice that all the bands we wanted to match are actually matched. However they are much broader than we wanted them to be, which means that the rejection objective is not very well respected. That mostly es from the fact that the weight on the matching objectives is deliberately very high。 it was observed that, if that weight is too small, the antenna is not matched in all configurations, the algorithm prioritizing rejection over matching. The rejection objectives were used to separate the different configurations to begin with, but it is not desirable that rejection es before the matching of the antenna. Figures 4 and 5 give the gain of the antenna, for different frequencies of interest。 for each frequency, the gain plotted uses the matched configuration, as presented in table III. These gain patterns take into account the presence of active ponents. Most configurations have around 6 dBi gain, with the exception of the highest frequency ( GHz), which offers around 4 dBi at most, and most notably, the lowest frequency which has negative gain everywhere (2 dBi max). This probably es from the electrically small size of the antenna at such a low frequency. AND REOPTIMIZATION? Prototyping of the switch board, measurements and cosimulationThe next step was to prototype the device. However, during previous measurements, the AsGa microwave switches proved very sensitive, both in the measurement itself, but also in the wielding process. As a result, in a first step, only the switch board was manufactured and measured: the ground plane and the dielectric board were both realized, however, the two electrical vias connecting the switches to the radiating patch were replaced by SMA connectors, thus allowing us to me