Automatic Modulation Classification (AMC) is a critical component in cognitive radio systems and spectrum management applications. This study presents a comprehensive comparative analysis of three attention mechanisms (i.e., baseline multi-head attention, causal attention, and sparse attention) integrated with Convolutional Neural Networks (CNNs) for radio frequency (RF) signal classification. It proposes a novel CNN-Transformer hybrid architecture that leverages different attention patterns to capture temporal dependencies in I/Q samples from the RML2016.10a dataset. The experimental results demonstrate that while baseline attention achieves the highest accuracy of 85.05\%, causal and sparse attention mechanisms offer significant computational advantages with inference times reduced by 83\% and 75\% respectively, while maintaining competitive classification performance above 84\%. The analysis reveals distinct attention pattern preferences across different modulation schemes, providing insights for designing efficient attention mechanisms for real-time radio signal processing applications.