Deep Reinforcement Learning Perspectives on Improving Reliable Transmissions in IoT Networks: Problem Formulation, Parameter Choices, Challenges, and Future Directions

The majority of communication protocols used in IoT networks for caching and congestion control techniques were rule-based which implies that these protocols are dependent on explicitly stated static models. To solve this issue, techniques are becoming more adaptive to changes in the network environment by incorporating a learning-based approach using Machine Learning (ML) and Deep Learning (DL). Recent surveys and review papers have covered topics on the use of ML and DL in either caching or congestion control techniques used in various types of networks. However, there is not an article in the literature dedicated to surveying the design of caching and congestion control mechanisms in IoT networks from the perspective of a Deep Reinforcement Learning (DRL) problem. Hence, this work aimed to survey the state-of-the-art DRL-based caching and congestion control techniques in IoT networks from 2019 to 2023. It also presented general frameworks for DRL-based caching and congestion control techniques based on surveyed works as a baseline for designing future protocols in IoT networks. Moreover, this paper classified the parameter choices of surveyed DRL-based techniques and identified the issues and challenges behind these techniques. Finally, a discussion of the possible future directions of this research domain was presented.