Accelerating the Detection of Spectral Bands by ANN-ED on a GPU

Spectrum sensing is the most important technique used to implement cognitive radio; this approach allows opportunistic and dynamic allocation of spectral bands. Among the methods used for detection, there are Artificial Neural Networks (ANN) and Energy Detection (ED); those exploit the signals coming from a Fast Fourier Transformed block (FFT). In this work, we focus on improving the performance of these three blocks by performing parallel computing, and considering the fusion of the two detectors ANN and ED. In this context, we implement three algorithms on GPU, which consist on exploiting the large number of cores to perform parallel calculation. The experimental results are compared with those obtained for CPU implementations. Our study presents how calculations distribution on GPU cores influences the global performance, and how to reduce execution time by optimizing data transfer. Furthermore, by exploiting the fine-grained parallel processing, and using a suitable choice of parameters, we find a considerable advantage of GPUs compared to CPUs, specifically for high data volumes.