Knowledge-and-Data-Driven Learning for Multi-Task Signal Classification

Signal classification plays a pivotal role in modern digital communications by ensuring efficient and secure data transmission across critical tasks, such as automatic modulation classification (AMC) and wireless technology classification (WTC). In recent years, the advent of Deep Learning (DL) has revolutionized signal classification by automating feature extraction and significantly improving classification accuracy in dynamic and noisy environments. Despite these advancements, effectively addressing the challenges of multi-task signal classification in complex scenarios remains an open problem. To tackle this, this paper presents a novel knowledge-and-data-driven multi-scale gated multi-layer perceptron (KDD-MSGMLP) framework, specifically designed to concurrently handle AMC and WTC tasks. The framework combines environmental side information as prior knowledge with advanced deep features, thereby substantially enhancing both robustness and accuracy in classification. At its core, the KDD-MSGMLP framework is powered by an innovative MSGMLP backbone network. This backbone effectively integrates multi-scale convolution with residual connections to enable shared feature embedding across tasks, while its gated multi-layer perceptron modules focus on refining task-specific features. Such a design ensures robust and efficient signal classification, making it well-suited for complex wireless communication environments. Simulation results demonstrate that the proposed framework achieves performance improvements of over 1% in AMC and between 2% and 4% in WTC. Moreover, it shows significant potential for enhancing performance and efficiency in complex wireless scenarios. The codes supporting this work are available on GitHub¹.