Deep data hiding‐based channel state information feedback within audio signals

Abstract

AbstractDeep learning‐based channel state information (CSI) hiding within images has been introduced to eliminate the downlink CSI feedback overhead in frequency division duplexing systems. In this letter, a deep data hiding‐based CSI feedback framework (named Au_EliCsiNet), which hides/superimposes downlink CSI within the transmitted audio signals, is proposed. Convolution neural networks are adopted to extract CSI features, hide CSI within audio signals, and reconstruct CSI from the audio signals. Simulation results show that the proposed Au_EliCsiNet can feed back downlink CSI accurately with no (or fewer) effects on the original audio transmission.