Implementation of a high-throughput OFDM system using Graphics Processing Units

To verify the performance of communication systems, software and hardware prototypes are widely used. However, with the increasing complexity of communication systems, software prototypes cannot meet the needs of real-time operations while hardware prototypes are expensive and time-consuming to develop. Nowadays, Graphics Processing Unit (GPU) shows great prospect as a coprocessor and has the capability of 1 Tera Floating-Point Operations Per Second (TFLOPS). Besides, programs of GPU are developed in C-language and can be debugged easily with an integrated development environment. This paper presents a novel procedure of using multiple GPUs and multi-thread technology to implement a high-throughput Orthogonal Frequency Division Multiplexing (OFDM) system. Experimental results show the GPU-based prototype is capable of real-time processing and is nearly 512 times faster than the traditional one using Central Processing Unit (CPU). Especially in some operations, GPU can even achieve a significant speedup of more than 1 000 times.