Reinforcement Learning based Adaptive Congestion Control for TCP over Wireless Networks

Abstract

Over the years, it has been observed that standard protocols designed for wired networks do not perform adequately when used for wireless networks. Researchers have proposed various protocols to enhance the functionalities of wireless network layers. TCP-Transmission Control Protocol is a transport layer protocol that experiences significant performance degradation in wireless networks. This is primarily because TCP considers any packet loss as a cause of network congestion, leading to an unnecessary reduction in transmission rate even when losses occur due to other reasons. This research work focuses in reviewing the existing approaches for improvement of TCP Congestion Control for wireless networks along with proposing TCP-RLACC (TCP with Reinforcement Learning based Adaptive Congestion Control). TCP-RLACC explores the network to select the most appropriate growth (linear, quadratic polynomial or exponential) of Cwnd - Congestion Window for adjusting transmission rate. TCP-RLACC is implemented in NS-3 simulator and evaluated with TCP Westwood+ for large number of wireless network scenarios. TCP-RLACC has shown significant improvements in terms of average throughput and end to end packet delivery ratio.