Bandwidth estimation in real-time communications

Develop robust algorithms to accurately estimate the time-varying available bandwidth between a sender and receiver in real-time communication systems despite rapidly evolving network architectures, increasingly complex protocol stacks, and the challenge of defining Quality of Experience (QoE) metrics that reliably improve user experience.

Background

In real-time video communication systems, the sender’s transmission rate is governed by an estimate of the available bandwidth on the bottleneck link. Overestimation causes congestion, while underestimation wastes capacity and degrades user-perceived quality. The problem is compounded by the heterogeneity and dynamism of modern networks (e.g., cellular, Wi‑Fi, wired) and the partial observability inherent in using packet-level statistics at the receiver.

A core difficulty is aligning optimization with user-perceived Quality of Experience (QoE). Traditional Quality of Service (QoS) metrics (throughput, delay, loss) do not reliably reflect perceptual quality across contexts, making it challenging to define metrics and estimators that consistently enhance QoE. The paper proposes a data-driven, human-in-the-loop and offline RL approach, but explicitly states that accurate bandwidth estimation for real-time communications remains an open challenge due to evolving architectures, complex protocol stacks, and QoE metric definition.