Combined use of graphics processing unit (GPU) and Central Processing Unit (CPU) for passive radar signal & data elaboration

Modern passive mobile radar systems equipped with circular arrays, require to elaborate simultaneously and in real time a certain number of passive signals. The signals are corresponding to, as an example, eight element of the circular array and ten carrier frequencies for each element in FM band. Further data are acquired if the system is intended for deploying at the same time DAB and/or DVBT signals. This sizing implies that the processing architecture has to sustain the load corresponding to, at least, more than ten adaptive cancellers (in time or space or both), pulse compressors, detection logics and bistatic and Cartesian trackers. Some of those algorithms map pretty well on GPU, thus giving a sizeable advantage in computational power with respect to standard Central Processing Unit (CPU)-based devices. This paper contains a rationale for the selection of algorithms to be mapped on GPU, some rules followed for the mapping strategy and comparisons of computational time obtained with GPU and CPU-based devices operating on the same input data. A computational architecture is proposed that consists in a wise mixture of GPU and CPU devices that exploits, at their best, the available equipment.