Deep transfer learning for constellation-based modulation recognition in inter-satellite optical wireless communication systems

Abstract Modulation format recognition (MFR) plays a significant role in the adaptive modulation scheme for inter-satellite optical wireless communication (IsOWC) systems. In this paper, a technique based on transfer learning (TL) using convolutional neural network (CNN) is proposed to achieve MFR in IsOWC systems. This approach effectively addresses the challenges associated with external interferences, such as solar scintillation, satellite vibration, and space debris, which collectively contribute to a 0.0883 dB loss at a propagation distance of 200 km. Four distinct signal constellation modulation formats 16, 32, 64, and 128-quadrature amplitude modulation (QAM) are used in the dataset, with optical signal-to-noise ratio (OSNR) ranging from 5 to 30 dB, for training the offline CNN models, including visual geometry group 19 (VGG19), InceptionV3, and residual network 152 (ResNet152). A comparative analysis of different CNN models is presented based on training and validation accuracy. The results show that the VGG19 model achieves the highest MFR accuracy of 99.84% at 47 epochs, with the help of the data augmentation technique, which includes the synthetic minority oversampling technique operation. It mainly enhances the scalability and generalization capabilities of the model, even with these external interference. The impact of various performance parameters including OSNR, external interferences, and image pixel size on MFR accuracy are thoroughly investigated. Moreover, VGG19 model achieves a lower symbol error rate of 0.002 at epoch 9 compared to other CNN models. Based on the results, it is found that the VGG19 model recognizes the signal constellations very efficiently, even at low OSNR values, and have future applications in terrestrial and underwater communication, biomedical applications, industry 4.0, military communication, internet of things devices, and quantum space communication.