Optimization of ARQ Protocols in Interference Networks with QoS Constraints

We study optimal transmission strategies in interfering wireless networks, under Quality of Service constraints. A buffered, dynamic network with multiple sources is considered, and sources use a retransmission strategy in order to improve packet delivery probability. The optimization problem is formulated as a Markov Decision Process, where constraints and objective functions are ratios of time-averaged cost functions. The optimal strategy is found as the solution of a Linear Fractional Program, where the optimization variables are the steady-state probability of state-action pairs. Numerical results illustrate the dependence of optimal transmission/interference strategies on the constraints imposed on the network.

💡 Research Summary

The paper addresses the problem of designing optimal transmission strategies for wireless interference networks that must satisfy Quality‑of‑Service (QoS) constraints. The authors consider a dynamic, buffered network with multiple sources that employ an Automatic Repeat reQuest (ARQ) mechanism to improve packet delivery reliability. Each source maintains its own queue, and in every time slot a source can decide whether to transmit a packet or remain silent. Because transmissions from different sources interfere with each other, the success probability of a packet depends on which other sources are transmitting simultaneously.

To capture the stochastic evolution of the system, the authors model the network as a Markov Decision Process (MDP). The state of the MDP consists of the queue lengths of all sources together with the current transmission decisions, while the action is a binary vector indicating which sources transmit in the current slot. Transition probabilities are derived from packet arrival processes (modeled as independent Bernoulli streams), the interference‑dependent success probabilities, and the ARQ rule (failed packets are retained for possible retransmission).

Two families of cost functions are introduced. The first family measures resource consumption, such as average transmit power or spectrum usage. The second family quantifies QoS metrics, namely average packet delay (or queue length) and packet loss probability (e.g., when a packet exceeds a maximum number of retransmissions). The performance objective is expressed as a ratio of time‑averaged costs – for instance, the average number of successfully delivered packets divided by the average power consumption – which reflects transmission efficiency. QoS requirements are also expressed as ratios, e.g., average delay ≤ Dmax and loss probability ≤ Lmax.

Because both the objective and the constraints are fractional, a direct linear programming formulation is impossible. The authors therefore cast the problem as a Linear Fractional Program (LFP). The key insight is that the steady‑state probabilities of state‑action pairs, denoted π(s,a), serve as decision variables. These probabilities satisfy the usual balance equations πP = π and Σπ = 1, which are linear. By applying the Charnes‑Cooper transformation, the fractional objective and constraints are converted into linear functions of new variables y(s,a) = π(s,a)/γ, where γ = Σ_{s,a} c₁(s,a)π(s,a) is the denominator of the objective ratio. The transformed problem is a standard linear program that can be solved efficiently with off‑the‑shelf solvers (e.g., Gurobi, CPLEX).

After solving the LP, the optimal steady‑state distribution π* is recovered, and a stationary randomized policy is obtained by normalizing π*(s,a) over actions for each state. In practice, the policy can be implemented either as a true randomized scheduler or, after a deterministic approximation (choosing the most probable action), as a simple rule‑based scheduler.

Numerical experiments are conducted for networks with two to four sources. Parameters such as packet arrival rates, baseline success probabilities, and an interference attenuation factor are varied. The results illustrate several important trends: (i) When QoS constraints are relaxed, the optimal policy tends to allow more simultaneous transmissions, thereby increasing the efficiency ratio; (ii) Tight delay or loss constraints force the policy to reduce concurrent transmissions, often inserting idle periods to mitigate interference and relying on ARQ to recover lost packets; (iii) In some regimes, the optimal solution is counter‑intuitive – the policy may deliberately keep all sources silent for a fraction of time, which minimizes interference and satisfies stringent power or loss limits. The study also shows that the structure of the optimal policy depends heavily on traffic intensity: high arrival rates favor aggressive transmission control, whereas low traffic allows more opportunistic simultaneous transmissions.

Overall, the paper contributes a rigorous methodology that combines MDP modeling with linear fractional programming to obtain globally optimal transmission policies under complex QoS constraints. The approach is analytically tractable, yields policies that can be interpreted in terms of transmission probabilities, and provides a benchmark for more scalable heuristics. Future work suggested by the authors includes state‑space reduction techniques (e.g., mean‑field approximations), online learning extensions using reinforcement learning, and extensions to multi‑channel, asynchronous, or non‑Bernoulli traffic scenarios.