Abstract
Traditional Transmission Control Protocol (TCP) relies on the assumption of bidirectional, coupled paths for data and acknowledgments. However, emerging technologies such as Visible Light Communication (VLC) and 5G millimeter-wave often utilize high-speed unidirectional links, creating significant path asymmetry. This paper proposes TCP Decoupling, a transport-layer modification that enables the independent management of forward and reverse paths. By expanding the standard 4-tuple to a 6-tuple and modifying the TCP handshake, the proposed system allows data and acknowledgments to traverse separate heterogeneous links. Experimental evaluations demonstrate that TCP Decoupling achieves high bandwidth utilization (up to 95%) and maintains stability under typical wireless packet loss conditions.
Introduction
The rapid evolution of 5G and beyond has introduced a variety of asymmetric network technologies. Visible Light Communication (VLC) and massive MIMO millimeter-wave systems provide exceptional downlink capacity but are often impractical for symmetric uplink implementation due to cost or hardware constraints. In such heterogeneous environments, a mobile terminal might receive high-speed data via a VLC link while being forced to return acknowledgments (ACKs) through a lower-capacity WiFi or cellular connection.
Standard TCP is ill-equipped for these scenarios because it "couples" the forward and reverse paths. When these paths exhibit wildly different characteristics, it can lead to performance degradation, such as the "data pendulum" effect. While link-layer aggregation can mitigate some issues, it adds complexity and cost. This paper argues for a transport-layer solution: TCP Decoupling.
Methodology: TCP Decoupling Architecture
The core innovation of this work is treating the forward (data) and reverse (ACK) directions as independent logical paths at the transport layer.
From 4-Tuple to 6-Tuple
Standard TCP identifies connections using a 4-tuple (Source IP, Source Port, Destination IP, Destination Port). To support decoupling, the authors introduce a 6-tuple by adding a Complementary Source IP and a Complementary Destination IP. This allows the TCP stack to explicitly manage separate IP addresses for the return path.
Modified Handshake and State Machine
During the initial three-way handshake, a new TCP Option is used to exchange these complementary addresses. This ensures both endpoints are aware of the specific interfaces designated for data versus ACKs. The researchers implemented this in the Linux kernel, updating the TCP state machine to handle interface handovers based on packet direction.
Experimental Results
The implementation was tested on a custom hybrid network testbed featuring a VLC downlink and an Ethernet uplink. Key findings include:
- High Efficiency: In a VLC/Ethernet hybrid setup with no packet loss, the decoupled TCP achieved a 95% bandwidth utilization ratio.
- Resilience to Loss: The protocol remained remarkably stable under minor network instability. Even with packet loss rates of 0.5%, 1.0%, and 2.0%, the bandwidth utilization stayed above 92%.
- Breaking Point: Performance only began to degrade significantly once the packet loss ratio exceeded 5%, indicating a robust implementation for typical wireless environments.
Conclusion and Future Work
TCP Decoupling provides a flexible, transport-layer solution for aggregating heterogeneous links without requiring specialized hardware at every intermediate router. By moving decoupling to the transport layer, the system gains end-to-end flexibility while maintaining compatibility with existing application layers. Future research will focus on handling extreme latency differences between paths and integrating these techniques with Multipath TCP (MPTCP) to combine path splitting with redundancy for even greater reliability.
Summary updated