CSCNN: Lightweight Modulation Recognition Model for Mobile Multimedia Intelligent Information Processing

A new round of global technological revolution and industrial transformation is accelerating. The era of 6G is on the horizon, envisioning the creation of an integrated network spanning across regions, airspace, seas, and land. This integration aims to achieve a truly global seamless coverage. Simultaneously, the importance of modulation recognition technology is continuously growing in tandem with the advancements in communication technology during this 6G vision. But faced with the vast amount of electromagnetic data and limited computing resources of terminal devices, previous technologies were unable to meet the realtime decision-making requirements for processing short observations or sudden bursts of signals in deployable systems. In this paper, to better leverage the correlation in the I/Q data of communication signals, our approach proposes the Deep Complex Separable Convolution (DCSC) operation by combining separable convolution operation and complex convolution operation. At the same time, to better preserve coupling information between channels and minimize the model size, we propose the Multilevel Separable Convolutional Residual Block (MSCRB). Based on the above two methods, we constructed the Complex Separable Convolutional Neural Network (CSCNN). This neural network significantly reduces the complexity of the deep learning model. We conducted experiments on RML2016.10a dataset and a dataset we created using signals collected through software-defined radio platform. On the RML2016 dataset, the smallest network we constructed, CSCNN-Tiny, has a model size of 3.04M, only 24.6% of the size of MobileNet. With 1.361M Flops, only 6% of MobileNet. However, it achieved a recognition accuracy of 52.45%, which is 0.54% higher than MobileNet.