band, this phenomenon does not have much practical significance.If computation works sensibly, within the PBG frequency range, transmission and reflection should add up to unity in the 2D case. Thus, during the optimization process (for example), it suffices to study the transmission power only. Because reflection is not studied, short input WG can be used, which limits the computation space. Also, the post-processing time is halved, because only one observation plane is used. Assuming that one is only interested in the transmission of the bend structure shown in Figure 7, one could approximately halve the computation space and use a single observation plane. Then, in the 2D case, one optimization step (simulation plus post-processing) with 4000 FDTD time steps, 2*27 E-points, 27 H-points, takes about 100 sec using a PC with 700 MHz/770-MB RAM (the speed of field computation is 1000 time steps/18 sec).
CONCLUSIONModelling PBG waveguide components has been considered, with field computation using FDTD. As an example, a 120°b end has been studied more closely, and power transmission results have been shown for various cases. The true necessity of bend geometry optimization has become apparent. Considering optimization, an approximative result for bend transmission is enough, that is, the spectrum does not have to fully converge. Then, with a 700-MHz PC, one optimization step (field computation and post-processing) takes a few minutes when using a 2-D model in field computation. Taking the finite-PBG plate thickness into account by using a 3D model will increase the needed time to tens of minutes.With 2D bend structures, the convergence of the spectrum may be very slow in certain frequencies. With 3D structures this "ringing problem" is less severe, due to the radiation losses in the z direction. In general, the cell size ⌬ ϭ a/10 seems to be sufficient if studying these structures at frequencies fa/c Յ 0.3. This is an important point if one strives for quick optimization. However, there may be structures with special frequencies, which can be very sensitive to the cell size and the relative hole size d/a.
Key words: wide tuning range; millimeter-wave VCO; P-HEMT technology
INTRODUCTIONAs demand for higher-frequency systems such as high-speed optical communication networks and automotive radar systems has increased, millimeter-wave component technology has become very important. Tunable low-phase noise oscillators are key components in millimeter-wave systems. Several oscillators operating at millimeter-wave frequencies using HEMT or HBT technology have been reported [1][2][3][4][5]. Although HBT-based oscillators have exhibited better phase-noise performance than HEMT-based oscillators, the output power available from HBT-based oscillators cannot compete with that of HEMT-based oscillators. Also, the HBT process is not compatible with the P-HEMT process, which is the most favored millimeter-wave circuit technology. For highlevel integration of monolithic VCOs with other MMIC components using the mature GaAs P-HEMT...