Parallelization of spectrum sensing algorithms using graphic processing units

Abstract

Cognitive radio (CR) is the next-generation communication system with high spectrum utilization and efficiency. It is very crucial for CR to sense the environment spectrum holes quickly and accurately. In this paper, we implement two kinds of spectrum sensing algorithms: waveform-based detection and cyclostationary feature extraction methods. Both of these algorithms are capable to separate the signal of interest from the noise or interference. In order to lower the computation time required by these complex algorithms, we parallelize these algorithms on a Graphic Processing Unit (GPU). Our methods show up to an average of 30× speedup in waveform preamble detection and an average of 39× speedup in cyclostationary feature extraction on a NVIDIA GTS 450 compared with the sequential implementation on a 2.94GHz Intel Core 2 CPU.