Lightweight deep convolutional neural network in the photonic receiving system of broadband dual-chirped waveform

To address the performance degradation issues caused by the nonlinearity of optical components and optical transmission media in dual chirped microwave waveform (DCMW) photonic receiving system, we have proposed a lightweight convolutional neural network (LwCNN)-assisted photonic reception solution. The system architecture includes a microwave photonic link to complete the photoelectric conversion and long-distance transmission of signals, as well as an intelligent signal processing module based on deep learning to recover signal distortion. The DCMW data input to the intelligent signal processing module undergoes time-frequency transformation, and features are extracted through LwCNN to achieve signal restoration. The time-frequency diagram processing results show that the noise and harmonic interference are suppressed, and the signal trajectory is clarified. The third-order spurious-free dynamic range (SFDR3) is improved from 78.3 dBꞏHz2/3 to 94.2 dBꞏHz2/3, which optimizes the linearity of the receiving system and provides an effective path for high-precision reception of double chirped waveform.