ABConv: Attention Based Convolution for Automatic Modulation Recognition

Automatic modulation recognition (AMR), an intermediate process of signal detection and signal demodulation, has been widely used in various fields. We introduce ABConv, a novel machine learning (ML) filter designed for AMR tasks, leveraging Attention-Based Convolution. Unlike the feature extraction approach using deep learning in AMR tasks, we focus on signal processing methods. Drawing parallels between attention and autocorrelation functions, our model dynamically generates convolution kernels based on attention scores to perform channel blind equalization. ABConv demonstrates state-of-the-art (SOTA) performance with in AMR tasks, surpassing multiple existing advanced algorithms. Additionally, we provide a preliminary visual analysis.