|
|
|
TELEcommunications & MICrowaves
|
printer-friendly version |
|
|
|||||||
|
Network and circuit group presentations by Dagang Li, Majid Homayouni and Vahid Tavakol Dagang Li will present his work between 14.00 and 15.30. Majid Homayouni probably from 15.30 to 16.30. Vahid Tavakol probably from 16.30 to 17.30 An introduction to mm-wave imaging systems -System modeling of the millimeter wave imaging setup for concealed weapon detection -The original 1.5D method -Transition from 1.5D to 2.5D -Fast calculation method based on 2D-FFT -The Detector -Study of possible design variations for the millimeter wave detector -Initial design of the detector plane -Future Work -Measurements of the current design and investigating newer designs -Identifying a practical system Characterization, modeling, and circuit design based on GaN FETs by Xiao Dongping AlGaN/GaN FETs are very promising for high-power, high-frequency, and high-temperature application. Due to the lack of commercial availability of bulk GaN substrates, to date GaN FETs are mainly grown on Sapphire, SiC and Si substrates. Although rapid progress has been made in the development of GaN-based and material and devices technology, Aspects about RF device layer out design, small- and large- signal model and power amplifier circuit design still need to be further studied. In this presentation, we will discuss about the GaN characterization, modeling and circuit design based on our in-house processed AlGaN/GaN FETs grown on different substrates. MLFMA for boundary element problems by Ides van den Bosch It has been a little more than a decade that the multilevel fast multipole algorithm (MLFMA) has been proposed to accelerate the matrix-vector product that appears in numerical methods such as the method of moments (MoM). The computational and memory costs of a MLFMA matrix-vector multiply are O(NlogN)for a surface boundary elements problem. In this work, we have implemented the MLFMA on top of a MoM code by combining Python extended with a powerful scientific library (SciPy, http://www.scipy.org), C++ and Fortran. We present the objects used in C++ to form the octree used for the far field multiplication, namely, the Cube, the Level and the Octtree, and how the natural data locality that results from this object-oriented strategy leads to a very clear, concise and efficient code. We present the work done on the inter-levels interpolators/anterpolators. Indeed, they have been highly optimized, as they can represent a significant part of the computational cost. We chose to use Gaussian interpolators, as they are better suited for periodic functions than Lagrangian interpolators. Also, the interpolation following theta is decoupled from that following phi, thereby greatly reducing the cost compared to a simultaneous treatment. We also present how the re-computing of the radiation functions at the finest level for each iteration has provided us great memory savings, with a reasonable increase in computation time. The data structure used for both the near field and preconditioner sparse matrices are also described. It is also discussed how the slicing of these matrices in blocks of limited size and their ``dumping'' to the hard disk allows to treat problems previously untractable on shared-memory single or multi-processor computers. As a result of all these optimizations, problems involving large numbers of unknowns per computing node can be solved. Therefore, the CPU and memory scalings of the program are presented and discussed, especially for problems which have a much larger memory footprint than the system available RAM. All these developments are available through Puma-EM, an open source (GPLv3) program that implements the aforementioned features. Find the slides here (pdf format). |
||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
