Utilização de placas de vídeo NVIDIA para otimizar o tempo de cálculo da predição de perda de percurso utilizando o método de equações parabólicas e o método das diferenças finitas

Abstract

This work is the result of a group project to optimize the calculation time in path loss predictions, using parallel computing techniques in the method of parabolic equations and finite differences to approximate the solution of the electromagnetic wave equation. With the growing expansion of telecommunications networks, the agility of project implementa tion is a differential for any company, for example, for cellular communication. To do this, it is also important to optimize the project time of these wireless communication systems. One of the project stages is the coverage analysis or prediction of path loss of the system. This analysis can be done in two ways: a faster way (found in some commercial software), but that assumes a simplified modeling and a more detailed way that requires complex methods and with high computational times. One of the methods that offers very good results is the parabolic equations, which can be solved by finite differences. However, its execution can be costly. Therefore, this method is not usually used in practical projects. Thus, it is interesting to look for ways to optimize its execution, making it possible to use it in practical applications. To solve the problem of the long time spent doing calculations during the simulation, the GPU was used to perform the calculations in parallel mode, making better use of the resources available on the computer and delivering the results with insignificant error to the error of the sequential simulation. Currently, the project is ongoing and only the first stage has been completed, which consists of generating simulations with an acceptable precision. The results obtained so far are promising, but it is still necessary to correct some details so that the error does not exceed the floating point error 10−14. The next step is to validate and correct all the implemented features, before adding new features.