Accelerating Radio Wave Propagation Algorithms by Implementation on Graphics Hardware

Radio wave propagation prediction is a fundamental prerequisite for planning, analysis and optimization of radio networks. For instance coverage analysis, interference estimation or channel and power allocation all rely on propagation predictions. In wireless communication networks optimal antenna sites are determined by either conducting a series of expensive propagation measurements or by estimating field strengths numerically. In order to cope with the vast amount of different configurations to select the best candidate from and to avoid expensive measurement campaigns, numerical predictions have to be both accurate and fast. In this chapter we focus on accelerating techniques for radio wave propagation algorithms in dense urban environments with the target frequency range of common mobile communication systems, i.e., several hundred MHz up to few GHz. One important aspect in radio wave propagation is the prediction of the mean received signal strength which can be simulated by taking complex interactions between radio waves and the propagation environment (see Figure 1) into account. Thus, the simulation of radio waves for propagation predictions becomes a computationally intensive task. A promising approach is the use of ordinary graphics cards, nowadays available in every personal computer. With over 1000 Gigaflops, modern graphics hardware offers the computational power of a small-sized supercomputer. This is achieved by a strict parallel many-core architecture which can be accessed by a high level of programmability. The main challenge of utilizing graphics hardware for scientific computations is to trick the graphics processors into general purpose computing by casting problems as graphics: Input data is transformed into images and algorithms are turned into image synthesis. However, in the last couple of years a growing support of so-called ”General Purpose Computation on Graphics Hardware” has led to recent changes in this architecture, allowing more common ways of parallel programming. Much effort and interest has been put on the acceleration of ray optical approaches, since most ray tracing algorithms tend to be computational intensive and exhibit run times up to hours. Therefore, we focus on the efficient implementation of wave guiding effects on graphics hardware. Among the most time consuming tasks in ray tracing is the problem of visibility between objects, i.e., the identification of all possible interaction sources for diffracted or reflected propagation rays. The algorithms we will present here are specifically designed to reduce the computational cost of the visibility computations by exploiting special features of the graphics card.