Accelerating a ray launching model using GPU with CUDA

The high computational cost of accurate deterministic wave propagation models often prevent them from being used in channel modelling. In this work, we present our experience of attempts to accelerate our 2.5D ray launching model using GPUs (Graphic Processing Units), which continue to grow in popularity due to their vast computation capability. At the heart of this trial is an implementation of a ray-surface intersection detection function, which was found to be the bottleneck of serial CPU computation, using NVIDIA's CUDA (Compute Unified Device Architecture). Various optimization efforts are made to obtain the best overall performance. The intersection detection function executes seven times faster on a large urban scenario after acceleration on a modest laptop GPU. This paper details the implementation of the CUDA-based intersection detection function and presents the acceleration results for different simulated environments.