Spectral Attention-Driven Intelligent Target Signal Identification on a Wideband Spectrum

Abstract

Due to the advances of artificial intelligence, machine learning techniques have been applied for spectrum sensing and modulation recognition. However, there still remain essential challenges in wideband spectrum sensing. Signal processing in the wideband spectrum is computationally expensive. Additionally, it is highly possible that only a small portion of the wideband spectrum information contain useful features for the targeted application. Therefore, to achieve an effective tradeoff between the low computational complexity and the high spectrum-sensing accuracy, a spectral attention-driven reinforcement learning based intelligent method is developed for effective and efficient detection of event-driven target signals in a wideband spectrum. As the first stage to achieve this goal, it is assumed that the modulation technique used is available as a prior knowledge of the targeted important signal. The proposed spectral attention-driven intelligent method consists of two main components, a spectral correlation function (SCF) based spectral visualization scheme and a spectral attention-driven reinforcement learning mechanism that adaptively selects the spectrum range and implements the intelligent signal detection. Simulations illustrate that because of the effectively selecting the spectrum ranges to be observed, the proposed method can achieve > 90% accuracy of signal detection while observation of spectrum and calculation of SCF is limited to 5 out of 64 of spectrum locations.