Near-infrared optical communication via deep learning-enhanced computational temporal ghost imaging in fog scattering environments

Fog scattering is one of the major factors limiting the performance of near-infrared optical communication (NIRC) systems, since it causes severe attenuation of optical signals. In this work, a NIRC system employing computational temporal ghost imaging (CTGI) is established based on a Mie-scattering channel model, and a controllable fog-scattering experimental platform is constructed to validate the theoretical analysis. Both simulation and experimental results consistently demonstrate that stronger fog scattering leads to significant degradation of communication performance, accompanied by a pronounced increase in the bit error rate (BER). To compensate for the degradation of temporal signals caused by fog scattering, a U-net-LSTM-based deep learning framework is introduced into the CTGI-based NIRC system. The proposed method significantly reduces the BER to approximately 10% of that obtained without deep-learning enhancement, enabling high-fidelity signal recovery. The results may provide technical support for the development of robust NIRC systems operating in adverse atmospheric scattering environments.