The paper presents a systematic review of differentially private federated learning (DPFL), addressing the growing concern of privacy and security in machine learning. It highlights the emergence of differential privacy as the standard for privacy protection in federated learning due to its rigorous mathematical foundation and provable guarantees. The authors propose a new taxonomy of DPFL, categorizing various differential privacy models and their applications in different federated learning scenarios. This taxonomy focuses on the definition and guarantee of differential privacy, distinguishing between sample-level and client-level privacy protection. The paper also explores the relationship between different DP models, such as DP, local differential privacy (LDP), and the shuffle model, and discusses their privacy guarantees and implementation methods. Additionally, it reviews over 70 recent articles on DPFL, summarizing their contributions and applications in various data types and real-world implementations. The authors conclude by identifying five promising directions for future research in DPFL.The paper presents a systematic review of differentially private federated learning (DPFL), addressing the growing concern of privacy and security in machine learning. It highlights the emergence of differential privacy as the standard for privacy protection in federated learning due to its rigorous mathematical foundation and provable guarantees. The authors propose a new taxonomy of DPFL, categorizing various differential privacy models and their applications in different federated learning scenarios. This taxonomy focuses on the definition and guarantee of differential privacy, distinguishing between sample-level and client-level privacy protection. The paper also explores the relationship between different DP models, such as DP, local differential privacy (LDP), and the shuffle model, and discusses their privacy guarantees and implementation methods. Additionally, it reviews over 70 recent articles on DPFL, summarizing their contributions and applications in various data types and real-world implementations. The authors conclude by identifying five promising directions for future research in DPFL.