CNN-Based Automatic Modulation Classification Under Phase Imperfections

Dynamic spectrum allocation for diverse future applications is anticipated to be supported by sixth-generation (6G) wireless networks. Specifically, automatic modulation classification (AMC) has been highlighted as a technique to enhance spectral utilization. However, its accuracy is influenced not only by additive white Gaussian noise and channel fading but also by phase imperfections (PI) coming from unsynchronized local oscillators and imperfect channel state information (CSI), leading to degraded classification performance. To solve this problem, we propose a convolutional neural network (CNN)-based scheme that transforms the received data to improve the classification accuracy under generalized PI conditions. Moreover, we also modify the kernel dimensions of the CNN layers to further improve the performance based on the geometry of the modulated schemes after the proposed transformation is applied to the received data. Finally, through simulations, we verified the effectiveness of the method in elevating AMC accuracy, even in intense PI conditions.