<p>A collaborative paradigm, federated learning enables devices to train machine learning models locally, without exposing raw data. The server initiates the training by transmitting the global model parameters to the clients, where it is trained using the local data. The model updates are communicated to the server, which in turn aggregates the model parameters and generates the global model. Multiple federated learning cycles are enacted until the model converges. Although data retention occurs at the client, the frequent exchange of local model parameters with the server poses notable privacy concerns and communication overhead. To address these challenges, an array of state-of-the-art mechanisms have been developed to safeguard user privacy and improve communication efficiency. In particular, privacy preserving measures can induce communication latency. Despite considerable advancements in this field, inherent trade-offs between accuracy, communication overhead, and privacy protections persist in hindering the development of FL frameworks. In this survey, we optimize our focus on approaches to improve communication efficiency while preserving the privacy of user data. This underscores the importance of exploring hybrid methods that effectively mitigate the dual concerns of privacy and communication. Additionally, we presented a decision framework for choosing communication efficient and privacy preserving FL strategies based on system limitations and threat models, as well as a unified compatibility matrix of these techniques in FL.</p>