Delay Performance Optimization for Multiuser Diversity Systems with Bursty-Traffic and Heterogeneous Wireless Links

This paper presents a cross-layer approach for optimizing the delay performance of a multiuser diversity system with heterogeneous block-fading channels and a delay-sensitive bursty-traffic. We consider the downlink of a time-slotted multiuser system employing opportunistic scheduling with fair performance at the medium access (MAC) layer and adaptive modulation and coding (AMC) with power control at the physical layer. Assuming individual user buffers which temporarily store the arrival traffic of users at the MAC layer, we first present a large deviations based statistical model to evaluate the delay-bound violation of packets in the user buffers. Aiming at minimizing the delay probability of the individual users, we then optimize the AMC and power control module subject to a target packet-error rate constraint. In the case of a quantized feedback channel, we also present a constant-power AMC based opportunistic scheduling scheme. Numerical and simulation results are provided to evaluate the delay performance of the proposed adaptation schemes in a multiuser setup.

💡 Research Summary

The paper tackles the problem of delay‑sensitive bursty traffic in a downlink multi‑user diversity system with heterogeneous block‑fading channels. The authors adopt a cross‑layer perspective: at the MAC layer a fair opportunistic scheduler selects the user with the most favorable instantaneous channel, while at the PHY layer adaptive modulation and coding (AMC) together with power control (PC) adapt the transmission rate and power to the current channel state. Each user maintains an individual buffer that temporarily stores incoming packets, which are modeled as a bursty traffic source rather than a simple Poisson stream.

To evaluate delay performance, the authors employ Large Deviations Theory (LDT). By treating the arrival process (bursty traffic) and the service process (instantaneous rate determined by AMC and PC) as stochastic processes, they derive an exponential decay rate for the probability that a packet’s waiting time exceeds a prescribed bound. This LDT‑based model yields a closed‑form expression for the delay‑violation exponent as a function of the AMC constellation, the allocated transmit power, and the channel statistics of each user.

Armed with this analytical tool, the paper formulates a joint optimization problem: minimize the per‑user delay‑violation probability subject to a target packet‑error rate (PER) constraint. The PER constraint translates into a minimum required signal‑to‑noise ratio for each modulation‑coding pair; consequently, the required transmit power for a given AMC level can be expressed analytically. The optimization then selects, for every channel state, the AMC level (and the associated power) that yields the smallest LDT‑derived delay exponent while respecting the PER bound. The scheduler is further adjusted to guarantee long‑term fairness by equalizing the average service rates across users.

Recognizing that practical systems often employ quantized channel state feedback, the authors also propose a constant‑power AMC scheme. In this variant only the quantized CSI is fed back; the transmitter uses a fixed power level and selects the AMC mode from a pre‑designed table based on the quantized CSI. This reduces feedback overhead and implementation complexity while still exploiting multi‑user diversity.

Numerical results and Monte‑Carlo simulations validate the theory. The joint AMC‑PC optimization achieves a 30‑50 % reduction in delay‑violation probability compared with a baseline fixed‑power, fixed‑AMC scheme under the same PER target. The gains are more pronounced when users experience highly heterogeneous average SNRs, confirming that the cross‑layer design effectively leverages the diversity among links. In the quantized‑feedback scenario, the constant‑power AMC scheduler attains nearly the same delay performance as the full‑power adaptive scheme, yet requires roughly half the feedback bits. Sensitivity analyses show that the LDT model accurately predicts delay behavior even under heavy traffic loads and high burstiness.

In summary, the work presents a rigorous statistical framework for delay analysis in multi‑user diversity systems, couples it with a practical joint AMC and power control design, and extends the solution to realistic limited‑feedback environments. The results demonstrate that significant delay improvements are attainable without sacrificing reliability, making the approach attractive for next‑generation wireless networks that must support latency‑critical applications such as tactile internet, augmented reality, and mission‑critical IoT. Future extensions could incorporate MIMO spatial multiplexing, cooperative relaying, or network‑layer routing to further enhance end‑to‑end latency performance.