REPUBLIC OF SERBIA

MINISTRY OF DEFENCE

www.mod.gov.rs

MINISTRY OF DEFENCE

Material Resources Sector

Defensive Technologies Department

Military Technical Institute

www.vti.mod.gov.rs

GPU-based preprocessing for spectrum segmentation in direction finding

Marko Mišić

University of Belgrade, Belgrade, marko.misic@etf.bg.ac.rs

ivan Pokrajac

Military Technical Institute, Belgrade, ivan.pokrajac@vs.rs

nadica Kozić

Military Technical Institute, Belgrade, nadica.kozic@gmail.com

Predrag okiljević

Military Technical Institute, Belgrade, predrag.okiljevic@mod.gov.rs

Abstract: Architecture of a modern direction finder is based on interception of radio signals in instantaneous bandwidth that is considerably wider than bandwidth of the signals. In the wider instantaneous bandwidth, more signals are active simultaneously. Additional processing is necessary in order to estimate the number of active emitters and parameters of each emission. This additional processing is known as spectrum segmentation or signal pre-classification. In order to improve quality of signal pre-classification, preprocessing is needed. Preprocessing can be based on morphology operations. Nowadays, graphics processing units (GPUs) can be used for signal processing in direction finding, since they offer vast computing power and bandwidth over central processing units (CPUs). GPUs are especially suitable for problems with high regularity. In this paper, preprocessing based on morphology operations have been implemented on the GPU using NVIDIA CUDA. Those operations include spatial filtering, erosion and dilation with different structural elements. GPU and sequential implementation have been tested with different test cases and significant speedups over sequential implementation have been observed. Finally, the results of the analysis are briefly discussed.

Keywords: CUDA, direction finding, GPU, morphological operations, spectrum segmentation.