The Broadband Interferometer Signal Processing And Direction Finding Algorithm Based On Gpu

The development of digital communication technology extends the signal frequency to an ever wide range, therefore, how to analyze and extract effective information from the high speed data flow has become an increasingly hot research subject in signal processing field. This paper studies the wideband direction finding system and proposes a virtual baseline based direction finding algorithm as well as the implementation of the system based on GPU.The tradeoff between accuracy and length of baseline is a general concern of traditional interferometer. The longer the baseline is, the more accurate its estimation is. However, the extension of the length of baseline will normally decrease the highest unambiguous frequency for direction finding. The correlative interferometer solves these issues by using the pre-measured sample phase difference to eliminate the phase ambiguity as well as imperfect factors introduced by array elements. Given that standard correlative interferometer is only available for one dimensional direction estimation, in this paper, we explore an improved correlative function followed by the curve fitting technique to realized two dimensional direction finding. In the case of wideband monitor and surveillance application, the concurrent data from multiple channels raise higher requirement for the computation capability of the system. When the number of channels is large enough, correlative interferometer becomes inapplicable for real time processing. With this problem this paper proposes an unambiguous algorithm based on virtual baseline transform while using the shorter baseline to solve the ambiguity of phase difference from longer baseline and thus reduce the computation significantly. Finally the probability of correctly solving ambiguity is given and its Cramer-Rao bound is deduced.In order to improve real time performance, the paper proposes an implementation of the wideband direction finding system based on graphic processing unit (GPU). The characteristics of the structure and parallelism of the digital modules within the software radio system including DDC, channelization and functionality algorithm unit are investigated. Their GPU based implementation is designed and accomplished using the CUDA programming platform by NVIDIA. Meanwhile the GPU programs are optimized with respect to memory access pattern and resource occupancy rate in order to maximize its performance. Finally timing test results from both the CPU-based and GPU-based implementation of digital modules are given to verify their effectiveness. Compared with its CPU version, GPU based implementation is able to achieve in some case an acceleration ratio as high as hundreds times.