The black quads in Figure 5 explicit the optimum interval for the referred matrix dimensions. Applying the least square method in these optimum values a linear correlation is obtained. It allows optimum interval predetermination to achieve best computational performance with different size of the matrix dimensions in the interval from 3 3 10 3 to 7 3 10 4 .
Scalability AnalysisThe scalability is the runtime simulation performance over the analysis with different matrix dimensions. In this approach, the performance can be expressed by the coefficients in (1). It was chosen the traditional iterative method was chosen with the option of fixed wavelength interval to analyze the scalability.In (1), x is the matrix dimension, a and b are the unknown coefficients. Figure 6 presents the linearization of these results, which make possible to predict the behavior of this method. The results presented in this section were obtained in a computer with Athlon X2 4400, 2GB RAM, with Linux operating system under the Ubuntu distribution, and gfortran compiler for Fortran 90. Figure 6 and Table 6 emphasize the differences in linear coefficients in these two codes, which are counterbalanced by the angular coefficient. These results show that the best performance is achieved when the linear coefficients are decreased.
CONCLUSIONSThis work presents a successful computational performance version of the FEM, called fFEM. It integrates preconditioned Bi-CGSTAB and two important approaches to eliminate many ILUT decompositions. These FEM improvements has, consequently, decreased the runtime simulations and improved the numerical scalability performance successfully demonstrated in some computational performance tests of these strategies in fix wavelength intervals and smallest linear coefficients problems.In addition, the best computational performance results were obtained using the iterative method, applying local assembling/ disassembling matrix and with the integration of some strategies to process a ILUT preconditioner, achieving performance improvements up to 70% when compared with the FEM reference runtime. The contribution of this sequential FEM solution provides significant impact when applied to electromagnetic device iterative optimizations, such as the Metaheuristics based. Some of these optimizations have been under study by this work team group.
ACKNOWLEDGMENTThe authors wish to thank CAPES and FAPESP, for the partial financial. Key words: power detector; radio frequency integrated circuits; receiver
INTRODUCTIONA power detector (PD) circuit is an important building block in the square kilometre array (SKA), the next-generation high-sensitivity radio telescope [1,2]. As signals received by a radio telescope are adjusted in its receivers to an optimum level for digitization by an analog-to-digital converter (ADC) [3], the PD is used to preserve the power level of the incoming signal for subsequent data processing. Ahead of the PD, an ultra-low-noise amplifier (LNA) [4,5] followed by additional gain stages [6] is used to amp...