Wideband Spectrum Sensing based on Collaborative Multi-Task Learning

To deal with the complex wireless conditions in cognitive radios, data-driven learning technologies have been advocated for spectrum sensing. While the most existing learning-based methods are designed for basic single-band and narrow-band circumstances, they may not work well in practical wide-band regimes. Due to the limited sensing capability and hardware constraints of practical secondary users (SUs) devices, individual SUs can only observe a portion of the entire wideband spectrum pool. It is also known as the issue of partial observations, which leads to a heterogeneous multi-task learning problem. To overcome these challenges, this work proposes a novel framework of wideband spectrum sensing via collaborative learning among distributed SUs. Capitalizing on the hierarchical nature of feature extraction in deep neural networks (DNN), we design a novel multi-task DNN architecture to detect wideband spectrum occu-pancy accurately and efficiently. By decoupling the large DNN into smaller band-specific sub-networks, these sub-networks can be jointly trained among distributed SUs even with heterogeneous local data. Simulation results indicate that our proposed method outperforms existing benchmarks by achieving higher learning accuracy at faster convergence speed.