Accelerating MATLAB based 5G Link Level simulator with MEX functions

Abstract

In this thesis, the acceleration of a MATLAB based 5G link level simulator with MATLAB Executable (MEX) functions is studied. The MEX functions written with C++ programming language are implemented and compared against MEX functions implemented with MATLAB Coder. Several techniques are explored, such as OpenMP parallelization, SIMD vectorization, compiler optimizations and overall compatibility and performance of available APIs. Other means of speeding up MATLAB code are also studied, with main focus on MATLAB page-wise functions. The performance of different implementations are compared by benchmarking the functions and simulations. The simulation benchmarking consists of calculating relative execution times of functions. With the functions accelerated with C++ in this thesis, it was possible to accelerate the execution by up to 12 times. If the original MATLAB implementation is optimized with page-wise functions, the acceleration is reduced to 6 times. In some other cases, the acceleration isn't as significant and the page-wise function may result in better performance. The findings of this thesis indicate that MEX acceleration still has a role in link level simulation acceleration, but careful analysis on which functions or sections of code should be converted to MEX is needed. Maintenance and update of MEX files versus pure MATLAB implementation requires always extra effort. Thus, not only the achieved simulation time reduction but also the maturity of the code impact the decision on choosing MEX acceleration.