A high performance FPGA-GPU-CPU platform for a real-time locating system

A heterogeneous Software Defined Radio (SDR) cluster platform that handles highly demanding processing algorithms in real-time is proposed. The solution based on a combination of FPGA, GPU and CPU offers the best balance between performance, cost, and flexibility. The key feature of our heterogeneous platform is achieving the required performance by assigning the tasks according to the technology characteristics. The FPGA in the proposed system does not only acquire external data but perform initial acquisition. This process aids in facilitating parallelism on the GPU side and optimizing the data transfer. The performance of the platform is demonstrated for an intensive real-time localization application. The overall costs are kept extremely low when compared to other solutions that can provide similar performance.