Lightweight OFDM-AMR: Representing Continuous Signal With Attention Enhanced LTC Network

With the advancement of 5G and 6G communication systems, automatic modulation recognition has gained increasing significance in the contexts of cognitive radio and spectrum monitoring. However, existing AMR algorithms encounter substantial challenges when deployed in base stations with hardware constraints, highlighting the urgent need for the development of lightweight models. This paper introduces a novel lightweight AMR approach that integrates the Liquid Time-Constant (LTC) network with an attention mechanism. The proposed network, grounded in Euler discrete Ordinary Differential Equation (ODE) is capable of capturing complex temporal dynamics. By refining the formula of network, it is adapted to effectively handle the discrete nature of communication time-series data. The inclusion of the attention mechanism enhances model performance while simultaneously mitigating the computational burden. Comprehensive experiments were conducted using the RML2016.10a, RML2016.10b, and EVAS datasets. The results demonstrate that the proposed method outperforms existing models in terms of computational efficiency, as it is the lightest among all evaluated approaches. The LTCA model achieves superior average accuracy, particularly excelling in recognition tasks involving the OFDM dataset, thereby achieving an optimal balance between model complexity and recognition accuracy.