Exploiting a cross-layer design for network performance improvement through Deep Reinforcement Learning

The increase in industrial IoT has brought many different connectivity requirements such as latency, packet loss and throughput. With this rise of connected devices, Quality of Service (QoS) has become more important to ensure these requirements are met by the network. However, initial QoS in Wi-Fi has only been managed by the MAC layer, limiting the application of QoS. More diverse QoS requirements must be met for Industrial Internet of Things (IIOT) networks currently not supported by QoS. Since QoS is also affected by parameters on the other layers of the OSI Stack, we deployed a cross-layer design to improve the QoS using Deep Reinforcement Learning (DRL). We achieved similar throughput, decreased latency by 9.47% and decreased packet loss by 24.90% compared to Minstrel using a DDQN DRL model.