Sliding Window Random Linear Network Coding (SW RLNC) is a promising approach for ensuring resilient communication in next-generation networks. In order to enhance both the coding efficiency and the overall performance of the coding scheme, the optimal configuration of the coding parameters should be addressed. Previous studies suggest that the trade-off between coding efficiency and overall performance in SW RLNC schemes is determined by two critical parameters, namely, the code rate and the coding window size. In this work, we address the joint optimization of the code rate and the coding window size by formulating it as a single-objective Sequential Model-Based Optimization problem and employ Bayesian Optimization (BO) models for efficiently solving it. To demonstrate proof-of-concept, we carry out an extensive evaluation campaign assessing the effectiveness of different BO models for this optimization problem. The results confirm that the optimization framework reliably finds optimal solutions with relatively few evaluations. Finally, we present a practical approach for leveraging the proposed optimization framework in real-world scenarios, allowing efficient adaptation to channel variations without requiring real-time re-optimization.

Sliding Window Random Linear Network Coding (RLNC) offers a clear path towards achieving ultra-high reliability and low latency at the same time. Such requirements are pivotal for a wide range of applications in the future Internet as well as in 5G and beyond networks. While traditional RLNC has been extensively used for some years now, its Sliding Window flavor is rather recent and extremely promising because of its implementation advantages and the high degree of customization. Probably the most essential parameter of Sliding Window RLNC is the coding depth, i.e., the extent of non-coded packets protected by a coded one. In this work, for the first time, we elaborate on properly choosing the coding depth and shed light on the related trade-offs. We, first, show, experimentally, that significant performance gains can be obtained by fine-tuning the coding depth. Then, we propose and validate an analytical framework that allows us to decide the coding depth based on a channel’s reliability profile. Finally, we introduce a dynamic algorithm that, based on our analytical findings, can improve the performance of sliding window RLNC in the presence of bursts of errors.

Random Linear Network Coding (RLNC) schemes that use a sliding window, also known as Sliding Window RLNC schemes, have proved to be efficient for reliable data transmission over an unreliable link. They combine the concept of the sliding window from (Automatic Repeat reQuest) ARQ protocols with RLNC in order to maximize the data flow and quickly recover from losses through redundant coded packets. We make the observation that, in these protocols, optimizing the data flow rate and the coding operation are two distinct objectives that cannot be met only by fine-tuning the sliding window size. To tackle the problem, we propose rapidARQ, a sliding window RLNC scheme with feedback that uses two distinct windows; the sliding window, which is used to optimize the data flow based on the link's bandwidth-delay product, and the coding window, which is used exclusively in the coding process. We experimentally show that rapidARQ outperforms other state-of-the-art sliding window RLNC schemes and achieves superior throughput-delay performance that better fits in the context of Ultra-Reliable Low-Latency Communication (URLLC). More interestingly, it does so while significantly reducing the coding complexity at the same time. The superior performance of rapidARQ is more prominent in channels with large bandwidth-delay products, a fact that renders its utility to current and future networks more essential.