Abstract-It is known that the conventional algorithm (CA) of hybrid finite element-boundary integral-multilevel fast multipole algorithm (FE-BI-MLFMA) usually suffers the problem of slow convergence, and the decomposition algorithm (DA) is limited by large memory requirement. An efficient twofold iterative algorithm (TIA) of FE-BI-MLFMA is presented using the multilevel inverse-based incomplete LU (MIB-ILU) preconditioning in this paper. It is shown that this TIA can offer a good balance of efficiency between CPU time and memory requirement. The tree-cotree splitting technique is then employed in the TIA to further improve its efficiency and robustness. A variety of numerical experiments are performed in this paper, demonstrating that the TIA exhibits superior efficiency in memory and CPU time to DA and CA, and greatly improves the computing capability of FE-BI-MLFMA.

No abstract

IEEE 802.16 standard suite defines a reservation-based bandwidth allocation mechanism. A SS (Subscriber Station) has to be polled to request bandwidth reservation before transmits uplink data to a BS (Base Station). In this mechanism exist two main polling modes: the unicast polling mode and the contention-based polling mode. The different polling operations in MAC (Medium Access Control) result in different PHY (PHYsical layer) frame structure that deeply affect the performance. Therefore, there should be an optimal scheme to adopt these two polling modes in order to optimize the performance. Although the standard defines five service classes to adaptively use the polling modes to fit the QoS (Quality of Service) requirements of different applications, it does not specify exactly a scheme to adopt these two polling modes efficiently and fairly during the polling process. In this paper, we investigate the polling mechanisms in IEEE 802.16 networks, and focus the attention on the performance caused by different adoption schemes. We also propose a simple but efficient polling mechanism to optimize the performance. The simulation results verify 123 636 F. Yin et al. that the performance is conditioned to the fulfillment of the polling mechanisms and our proposed optimal polling scheme can allocate bandwidth more efficient and achieve better performance.

No abstract