Sleep plays a crucial role in our overall health and well-being. Analyzing sleep stages and the frequency of arousals can enhance our understanding of sleep quality and help protect individuals’ sleep health. This study delves into the application of deep learning for the simultaneous tasks of sleep staging and sleep arousal identification using the single-lead electrocardiogram (ECG) signal. A deep learning model was developed and evaluated on the Multi-Ethnic Study of Atherosclerosis (MESA) dataset, demonstrating promising performance. The model achieved comparable or improved results compared to existing approaches. It was able to predict sleep stages with an average accuracy of 0.79 ($\kappa=0.68$) and arousal with an area under receiver operating characteristic (AUROC) of 0.935. Notably, the model’s fairness across different cohorts based on sex, age, and ethnicity was explored, revealing consistent performance regardless of these factors. The model’s ability to automatically extract relevant features, such as instantaneous heart rate and heart rate variability, was investigated, shedding light on its interpretability.