TAROT: Towards Optimization-Driven Adaptive FEC Parameter Tuning for Video Streaming

Forward Error Correction (FEC) remains essential for protecting video streaming against packet loss, yet most real deployments still rely on static, coarse-grained configurations that cannot react to rapid shifts in loss rate, goodput, or client buffer levels. These rigid settings often create inefficiencies: unnecessary redundancy that suppresses throughput during stable periods, and insufficient protection during bursty losses, especially when shallow buffers and oversized blocks increase stall risk. To address these challenges, we present TAROT, a cross-layer, optimization-driven FEC controller that selects redundancy, block size, and symbolization on a per-segment basis. TAROT is codec-agnostic–supporting Reed-Solomon, RaptorQ, and XOR-based codes–and evaluates a pre-computed candidate set using a fine-grained scoring model. The scoring function jointly incorporates transport-layer loss and goodput, application layer buffer dynamics, and block-level timing constraints to penalize insufficient coverage, excessive overhead, and slow block completion. To enable realistic testing, we extend the SABRE simulator 1 with two new modules: a high-fidelity packet-loss generator that replays diverse multi-trace loss patterns, and a modular FEC benchmarking layer supporting arbitrary code/parameter combinations. Across Low-Latency Live (LLL) and Video-on-Demand (VoD) streaming modes, diverse network traces, and multiple ABR algorithms, TAROT reduces FEC overhead by up to 43% while improving perceptual quality by 10 VMAF units with minimal rebuffering, achieving a stronger overhead-quality balance than static FECs.

💡 Research Summary

The paper addresses a fundamental shortcoming of current video streaming deployments: the reliance on static, coarse‑grained Forward Error Correction (FEC) configurations that cannot react to rapid changes in loss rate, goodput, or client buffer occupancy. Static settings either waste bandwidth with unnecessary redundancy during stable periods or fail to protect against bursty losses, leading to stalls and degraded QoE. To overcome this, the authors propose TAROT (Towards Optimization‑Driven Adaptive FEC Parameter Tuning), a cross‑layer controller that selects, for each video segment, the FEC code family (Reed‑Solomon, RaptorQ, or XOR‑based), the block size (n source symbols), the number of repair symbols (k), and the symbol size (S).

TAROT works by first monitoring real‑time telemetry: packet loss, instantaneous goodput, client buffer level, and remaining time before the segment deadline. It then prunes a pre‑computed candidate set of (n, k, S, code) configurations to those feasible under current bandwidth constraints. Each candidate is evaluated with a multi‑objective scoring function that penalizes (1) insufficient coverage (i.e., the probability that n recv + k recv < n), (2) overhead (k/n), (3) encoding/decoding latency (which grows with larger S and more complex codes), and (4) deadline violations. The weights of these penalties can be tuned to prioritize latency‑sensitive live streams or quality‑sensitive VOD. The candidate with the lowest score is selected for the upcoming segment, enabling per‑segment adaptation without heavy online optimization.

To evaluate TAROT, the authors extend the widely used SABRE simulator with two modules: a high‑fidelity packet‑loss generator that replays multi‑trace LTE and 5G loss patterns, and a modular FEC benchmarking layer that can instantiate any combination of code family and parameters, while also modeling the extra bandwidth consumption and the CPU‑induced encoding delay. This extended platform, called SABRE‑FEC, provides a reproducible environment for fine‑grained FEC research.

Experiments cover both Low‑Latency Live (LLL) and Video‑on‑Demand (VoD) modes, multiple Adaptive Bitrate (ABR) algorithms (e.g., BBA, Pensieve), and a diverse set of real‑world network traces. Compared with static FEC baselines (e.g., a fixed RaptorQ configuration with n = 20, k = 10, S = 64 B), TAROT achieves up to 43 % reduction in redundancy overhead and improves VMAF by an average of 10 points (≈1 JND). In burst‑loss scenarios with shallow buffers, TAROT dynamically shrinks block size and raises redundancy just enough to avoid rebuffering, while in stable periods it expands blocks and lowers redundancy to maximize usable goodput. Rebuffering events drop to below 0.2 % and overall average bitrate rises by ~12 %.

The paper also discusses limitations: the need to pre‑compute the candidate pool, the lightweight nature of the online scoring (which may need further optimization for ultra‑low‑latency use cases), and the current focus on a single client‑server path (future work includes multi‑path, CDN, and energy‑aware extensions).

In summary, TAROT introduces a novel, practical framework that jointly optimizes FEC code selection, block configuration, and symbolization on a per‑segment basis, integrating transport‑layer loss, application‑layer buffer dynamics, and encoding latency into a unified decision process. The extensive simulation study demonstrates that such fine‑grained, cross‑layer adaptation can substantially improve QoE while curbing bandwidth waste, making TAROT a promising solution for next‑generation 5G/Edge video streaming deployments.