ML Assisted Feedback Mechanism for TCP Congestion Control in Next Generation Wireless Networks

5G and beyond delivers high data throughput, in turn, requiring proactive mechanism in core and Radio Access Network (RAN) fast path for delivering congestion feedback to Congestion Control (CC) algorithm of Transmission Control Protocol (TCP). Mobile handovers between cells with higher variance in Bandwidth Delay Product (BDP) especially in a heterogeneous network present multitude of challenges for transport layer protocols such as TCP and QUIC. Inability of existing CC mechanisms in TCP and QUIC to adjust to sudden changes in BDP result in packet queueing delays, packet drops leading to degraded user experience. In this paper, we propose a novel method to implement feedback mechanism in RAN fast path using Machine Learning (ML) model to predict impending congestion event and notify TCP’s CC. This paper also discuss about implementation of a proactive bandwidth regulation function in User Plane Function (UPF) by using ML to predict an optimal maximum TCP Receive Window (RWND) size. We demonstrate the effectiveness of our proposed method in both RAN and core network with Key Performance Indicators (KPIs) collected from live air network. The results from conducted experiments demonstrate 17% reduction in Flow Completion Time (FCT) and 15% reduction in packet loss end to end.