Design and performance evaluation of enhanced congestion control algorithm for wireless TCP by using a deep learning

In a mixed wireless and wired network environment, packet loss can be caused by various types of due to network congestion as well as wireless errors. However, existing TCP decides that all packet loss is network congestion, which causes congestion control to degrade TCP performance. This paper distinguishes network congestions and wireless errors through a deep learning algorithm when a packet loss occurs. In case of loss due to network congestion, congestion control is performed in the same way as existing TCP. In case of loss due to errors, we propose an algorithm that can improve the wireless TCP performance by only retransmitting lost packets without reducing the congestion window. Through the simulation, we show that the proposed algorithm improves the TCP performance by reflecting the result of the pre-learned deep learning algorithm, compared to the "TCP Westwood" or "TCP Veno" proposed for improving the wireless TCP performance by discriminating the congestion and wireless error.