ICon: How Can We Intelligently Predict TCP Congestion?

To enhance the wireless network performance, several new wireless standards, such as IEEE 802.11ac/ax, are introduced, where the objective is to increase the throughput in the network. However, Transmission Control Protocol (TCP) congestion degrades the network performance due to limited network capacity. The standard congestion control mechanism focuses primarily on acknowledgment (ACK) packets, which may result in inappropriate handling of congestion, and lead to the increase in packet drop rate and packet transmission delay. As a consequence, overall network efficiency is degraded. In this paper, we propose Intelligent Congestion control (ICon), a learning-based congestion control mechanism, that uses Deep Neural Network (DNN) to smartly predict the occurrence of congestion based on the present channel condition in the network and the last congestion window size (old window size). We use Signal-to-Noise Ratio (SNR) as the measurement of channel condition, and thus the impact of signal quality on the volume of transmitted packet is implicitly introduced in the prediction of congestion in ICon. For the generation of the dataset to train the proposed DNN, an intensive simulation is used. Through simulation, we analyze the performance of ICon, where it is noted that the proposed scheme significantly improves overall network performance.