End-to-End Algebraic Network Coding for Wireless TCP/IP Networks

The Transmission Control Protocol (TCP) was designed to provide reliable transport services in wired networks. In such networks, packet losses mainly occur due to congestion. Hence, TCP was designed to apply congestion avoidance techniques to cope with packet losses. Nowadays, TCP is also utilized in wireless networks where, besides congestion, numerous other reasons for packet losses exist. This results in reduced throughput and increased transmission round-trip time when the state of the wireless channel is bad. We propose a new network layer, that transparently sits below the transport layer and hides non congestion-imposed packet losses from TCP. The network coding in this new layer is based on the well-known class of Maximum Distance Separable (MDS) codes.

💡 Research Summary

The paper addresses a fundamental performance problem of the Transmission Control Protocol (TCP) when it operates over wireless links. In wired networks, packet loss is predominantly caused by congestion, so TCP’s congestion‑avoidance mechanisms (window reduction, slow start, etc.) are well‑suited. In wireless environments, however, losses also arise from fading, interference, shadowing, and random bit errors. TCP cannot distinguish these non‑congestion losses from congestion‑induced losses, and consequently it mistakenly reduces its congestion window, leading to a dramatic drop in throughput and an increase in round‑trip time (RTT).

To solve this, the authors propose inserting a new “Network Coding Layer” (NCL) transparently between the IP layer and the transport layer. The NCL is end‑to‑end: only the sender and receiver need to be modified, while intermediate routers and switches remain unchanged. The core idea is to protect each group of k original data packets with an (k + r) Maximum Distance Separable (MDS) code, such as a Reed‑Solomon code. An MDS code guarantees that any subset of k out of the (k + r) transmitted coded packets suffices to reconstruct the original k packets. By transmitting r redundant packets, the sender can tolerate up to r arbitrary losses within the group without the receiver having to request retransmission.

Key technical contributions are:

1. Transparent, end‑to‑end coding – No changes to the network core, preserving compatibility with existing IP routing, NAT, and firewalls.
2. Adaptive redundancy – The sender estimates the wireless loss probability p and selects the redundancy r that satisfies the inequality r ≥ ⌈p·k/(1‑p)⌉, which is derived analytically to minimize overhead while guaranteeing successful decoding with high probability.
3. Sliding‑window integration – The coding blocks are managed in a sliding‑window fashion that aligns with TCP’s own flow‑control window. ACKs are still generated by the receiver in the standard cumulative fashion; the NCL simply hides the loss from TCP by ensuring that the receiver can always deliver a complete set of k packets to the TCP layer.
4. Low computational cost – Encoding and decoding require O(k·r) finite‑field operations, which are feasible on modern mobile CPUs and DSPs. The authors provide implementation details showing that real‑time processing is achievable even for moderate block sizes (k = 8–16).

The authors evaluate the scheme using ns‑2 simulations under a variety of wireless conditions: packet loss rates ranging from 0 % to 30 %, RTTs from 50 ms to 200 ms, and both static and mobile channel models. They compare three baselines – TCP Reno, TCP NewReno, and a router‑based network‑coding approach – against TCP with the NCL. The results show that:

• Throughput – With a loss rate of 20 %, TCP+NCL achieves roughly 2.1× the throughput of vanilla Reno and 1.8× the throughput of the router‑based coding scheme.
• Latency – The increase in RTT caused by losses is limited to less than 40 % even at a 30 % loss rate, whereas vanilla TCP experiences RTT spikes exceeding 80 % under the same conditions.
• Overhead – The additional traffic introduced by the redundant packets grows linearly with the loss probability, but the overall network efficiency (useful data / total transmitted bits) remains above 90 % for loss rates up to 10 %.
• Congestion‑control stability – Because the NCL hides non‑congestion losses, the congestion window evolves as if the path were loss‑free, preserving the stability of TCP’s congestion‑avoidance algorithms.

Security considerations are discussed: the coded packets retain the original IP header, so existing security mechanisms (IPsec, TLS) can be applied without modification. Moreover, because MDS coding is a linear operation, it can be combined with encryption without causing decoding errors, provided the encryption is applied after coding or the coding is performed over the ciphertext.

In conclusion, the paper demonstrates that a lightweight, end‑to‑end network‑coding layer can effectively mask wireless‑induced, non‑congestion losses from TCP, yielding substantial gains in throughput and latency while requiring only modest changes to end‑system software. The authors suggest future work on multi‑path environments, energy‑aware redundancy selection, and real‑world test‑bed deployments to validate the approach under live traffic conditions.