Removing Water Vapor Lines From THz-TDS Data Using Neural Networks

In this work, a new technique for the removal of water vapor absorption lines from terahertz signals based on deep neural networks is presented. The designed neural networks were trained on signals acquired under different air humidity, with and without samples. The neural networks were validated on signals of samples, which were not available to the neural networks during training. The quality of the results is comparable to different reported model-based approaches; however, the removal of water vapor absorption lines can be done at a much faster rate. Finally, this technique can be used for removal of any other impulse response from terahertz signals, having a training dataset provided.