Multiuser broadcast erasure channel with feedback - capacity and algorithms

We consider the $N$-user broadcast erasure channel with $N$ unicast sessions (one for each user) where receiver feedback is regularly sent to the transmitter in the form of ACK/NACK messages. We first provide a generic outer bound to the capacity of this system; we then propose a virtual-queue-based inter-session mixing coding algorithm, determine its rate region and show that it achieves capacity under certain conditions on channel statistics, assuming that instantaneous feedback is known to all users. Removing this assumption results in a rate region that asymptotically differs from the outer bound by 1 bit as $L\to \infty$, where $L$ is the number of bits per packet (packet length). For the case of arbitrary channel statistics, we present a modification of the previous algorithm whose rate region is identical to the outer bound for N=3, when instant feedback is known to all users, and differs from the bound by 1 bit as $L\to \infty$, when the 3 users know only their own ACK. The proposed algorithms do not require any prior knowledge of channel statistics.

💡 Research Summary

The paper studies a broadcast erasure channel with N users, each having its own unicast data stream, and investigates how regular ACK/NACK feedback from the receivers can be exploited to achieve the channel’s capacity. The authors first derive a generic outer bound on the achievable rate region for this multi‑session setting. The bound is expressed in terms of virtual‑queue stability conditions: for each user i, the average arrival rate λ_i must not exceed the effective service rate μ_i that depends on the erasure probabilities and the feedback information. This outer bound generalizes the classic single‑user erasure‑channel capacity (1 – ε_i) to a multi‑user, multi‑session scenario and serves as a benchmark for any coding scheme.

To approach this bound, the authors propose a novel “virtual‑queue‑based inter‑session mixing” algorithm. For each user i a virtual queue Q_i stores the information packets that have not yet been successfully delivered to that user. At every time slot the transmitter selects a subset of the queues and forms a linear combination (XOR) of the head‑of‑line packets. The selection rule maximizes the “coding gain”: the transmitted linear combination should provide new information simultaneously to as many users as possible, given the current ACK/NACK feedback. After transmission, the feedback is used to update the queues: packets that were successfully decoded are removed, while those that were erased remain for future mixing. This dynamic mixing replaces the traditional repeat‑until‑success ARQ approach with a network‑coding‑style operation that reuses already delivered symbols to create new useful combinations.

The paper analyzes two feedback models. In the first model, “global instantaneous feedback,” every user instantly knows the ACK/NACK outcomes of all other users as well as the transmitter. Under this assumption the proposed algorithm can be shown to be throughput‑optimal: the virtual‑queue stability region exactly matches the outer bound, meaning the scheme achieves the full capacity region without any prior knowledge of the erasure probabilities. The algorithm adapts in real time solely based on the observed feedback, making it robust to time‑varying channels.

In the second model, “local feedback,” each user only learns its own ACK/NACK status, while the transmitter still receives the individual feedback messages. This creates an information asymmetry that prevents the transmitter from perfectly coordinating the mixing decisions across all users. To mitigate the loss, the authors introduce a modest modification: occasional “helper” transmissions that target a single user, together with a refined queue‑synchronization rule. With this adjustment they prove that for N = 3 the modified scheme again attains the outer bound exactly. For arbitrary N, the achievable rate region differs from the outer bound by at most one bit per packet when the packet length L → ∞. Because the gap shrinks as L grows, the scheme is essentially capacity‑achieving for realistic packet sizes (e.g., L ≥ 1000 bits).

A notable aspect of the work is that the algorithm does not require any statistical knowledge of the channel (the erasure probabilities ε_i). All necessary parameters are inferred on‑the‑fly from the instantaneous feedback, which eliminates the need for offline channel estimation or learning phases. The authors also provide extensive simulations that confirm the theoretical results: the average queue lengths remain bounded, the per‑user throughput follows the predicted region, and the delay performance is comparable to that of optimal policies derived from the outer bound.

The paper’s contributions can be summarized as follows:

1. General Outer Bound – A concise, analytically tractable capacity outer bound for the N‑user broadcast erasure channel with independent per‑user erasures and feedback.

2. Virtual‑Queue Mixing Algorithm – A practical, low‑complexity coding scheme that dynamically forms XOR combinations based on real‑time ACK/NACK information, achieving the outer bound under global feedback and staying within one bit of it under local feedback.

3. Feedback‑Model Analysis – Rigorous proofs that the algorithm is optimal with global instantaneous feedback and near‑optimal (≤ 1 bit gap) with only local feedback, including a special exact‑capacity result for the three‑user case.

4. Statistical‑Independence – The scheme works without any a priori knowledge of the erasure statistics, making it suitable for time‑varying wireless environments where channel parameters are unknown or rapidly changing.

5. Practical Relevance – The 1‑bit gap shrinks with packet length, implying that for typical network packets the performance loss is negligible. This makes the approach attractive for modern wireless standards (e.g., 5G/6G) that already support fast ACK/NACK signaling and require efficient multi‑user broadcast mechanisms.

In conclusion, the authors demonstrate that by treating each user’s pending information as a virtual queue and by intelligently mixing these queues using instantaneous feedback, one can essentially close the gap between achievable rates and the information‑theoretic capacity of the broadcast erasure channel. The work bridges the gap between network coding theory and practical ARQ‑style retransmission protocols, offering a robust, capacity‑approaching solution that is both analytically sound and implementable in real wireless systems.