MU-MIMO Uplink Timely Throughput Maximization for Extended Reality Applications

In this work, we study the cross-layer timely throughput maximization for extended reality (XR) applications through uplink multi-user MIMO (MU-MIMO) scheduling. Timely scheduling opportunities are characterized by the peak age of information (PAoI)-metric and are incorporated into a network-side optimization problem as constraints modeling user satisfaction. The problem being NP-hard, we resort to a signaling-free, weighted proportional fair-based iterative heuristic algorithm, where the weights are derived with respect to the PAoI metric. Extensive numerical simulation results demonstrate that the proposed algorithm consistently outperforms existing baselines in terms of XR capacity without sacrificing the overall system throughput.

💡 Research Summary

This paper tackles the pressing challenge of supporting extended reality (XR) applications over the uplink of future cellular networks, where users must continuously stream high‑resolution raw video to the base station. Recognizing that conventional scheduling solutions focus mainly on downlink traffic and rely on application‑level metadata exchange—incurring prohibitive signaling overhead for the stringent latency budgets of XR—the authors propose a cross‑layer design that integrates a freshness metric, the peak Age of Information (PAoI), directly into the scheduling decision process.

The system model assumes a single gNB equipped with multiple transmit chains serving N XR users in the uplink via MU‑MIMO. Each user can transmit up to λₙ(t) spatial streams, with the total number of streams constrained by a hardware limit Λ̄. The physical layer is modeled with block‑fading channels, and the achievable per‑user throughput Qₙ(t) is approximated by the sum of log‑rate terms over allocated resource elements, scaled by the SINR. Users periodically report a Buffer Status Report (BSR) indicating pending bits, and the gNB provides binary scheduling decisions βₙ(t) together with stream allocations and MCS values. No retransmissions are considered; instead, successful reception is signaled by a one‑bit feedback sₙ(t).

Age of Information (AoI) is defined as the elapsed time since the last successfully received packet. The peak AoI (PAoI) captures the worst‑case freshness over a horizon and is directly linked to the packet delay budget (PDB) D̄: a user is deemed satisfied if its time‑averaged PAoI stays below D̄, which corresponds to the XR KPI of ≥99 % of a packet being delivered within the deadline. The authors formalize the scheduling problem (P1) as a maximization of the long‑term sum of α‑fair utilities of instantaneous throughputs (with α=1, i.e., proportional fairness) subject to (i) the PAoI constraint for each user, (ii) the per‑slot spatial‑layer budget, and (iii) binary scheduling variables. This yields an integer non‑linear program with both instantaneous and time‑averaged components, proven to be NP‑hard due to intra‑cell interference coupling.

To obtain a practical solution, the paper introduces a per‑TTI iterative heuristic that augments the classic proportional‑fair (PF) metric with PAoI‑derived weights. The PF metric seeks to maximize Σₙ Qₙ(t)·Qₙ,Avg(t), where Qₙ,Avg(t) is an exponential moving average of past throughputs (time constant τ≈1000 TTIs). The PAoI‑based weight Wₙ,PAoI(t) is defined piecewise: if a weighted AoI measure Δₙ,wa(t) ≤ D̄, the weight is 1 − Δₙ,wa(t) − κ (boosting the user’s priority); otherwise it is 1 / (1 − Δₙ,wa(t) − κ), which gradually reduces the influence, reverting to the standard PF behavior. Δₙ,wa(t) blends the instantaneous AoI Δₙ(t) and the historical PAoI average via a parameter θ∈