SOLVING A PROBLEM OF RESOURCE-INTENSIVE MODELING OF DECODERS ON MASSIVELY PARALLEL COMPUTING DEVICES BASED ON VITERBI ALGORITHM

摘要

In this paper, we consider the problem of resource-intensive simulation of coding/decoding which corrects errors made at the preliminary stages of modern telecommunication system development. We propose to use the technology of parallel computing on GPU (GPGPU) to solve the problem of the process acceleration. We discuss the aspects of encoding/decoding simulation, which corrects errors in heterogeneous systems. The results of this technology applying in the convolutional codec parameters simulation, decoded by Viterbi algorithm, are given as well. Another problem concerned with limitation of the interaction speed with the computing device tail part and a random access to memory is also considered. We propose a solution by communication minimization at host-computing device level, as well as the use of caching. The simulation tools are described in the paper, including the use of computing technique of general purpose on GPU allowing to reduce the time required to optimize the noiseless coding system and thus for the development and implementation of telecommunication devices. We describe the solutions of tasks on codecs characteristics research using massively parallel computing, differing by simplified initialization of flow pseudorandom-number generator (PRNG) ensuring high performance with sufficient accuracy of calculations by reducing the number of calls to an external status register.