Communication engineers and network theoreticians have introduced a cross-layer scheduling framework designed to jointly optimize flow control and radio resource allocation across dual-connectivity multi-queue multi-server wireless architectures. In modern cellular systems governed by specifications from the 3rd Generation Partnership Project, dual-connectivity links a user terminal simultaneously to a Master eNodeB and a Secondary eNodeB over a non-ideal backhaul interface known as X2. While connecting to two base stations provides an avenue for expanding network capacity and enhancing user quality of experience, managing bursty data traffic across coupled queues without future arrival information presents severe analytical and algorithmic challenges. The new investigation formulates the underlying theoretical boundaries of this dual-path environment and derives online control algorithms that guarantee system stability while navigating the mathematical trade-off between network utility and queuing latency.
To formally understand the limits of multi-user wireless networks equipped with orthogonal sub-channels, such as orthogonal frequency division multiplexing sub-carriers, the researchers modeled the network as a dual-connectivity multi-queue multi-server system operating across discrete, equal-length time slots. In this setup, a set of parallel queues representing individual users is served by two distinct pools of transmission servers corresponding to the primary and secondary connections. Each wireless server can be allocated to at most one user per time slot to honor the exclusive-access constraint inherent to orthogonal channel configurations. Furthermore, physical channel states are treated under a quasi-static block fading assumption, remaining constant within any given scheduling slot but varying independently across slots according to stationary probability distributions.
A critical contribution of the research lies in the exact mathematical derivation of the network-layer capacity region for dual-connectivity multi-queue multi-server networks. Defined as the closure of all arrival rate vectors that can be stably supported by any feasible scheduling policy, the capacity region was proven to be fully characterizable by a finite set of linear inequalities. By examining the algebraic structure of the channel rate sets, the authors proved that the admissible product sets of channel rates for the secondary connectivity form a subset of those for the primary base station. This containment allowed the research team to demonstrate that the binding stability constraints of the dual-connectivity system reduce to a single family of inequalities indexed entirely by the minimal generating set of the secondary connection, establishing an explicit theoretical boundary that markedly surpasses single-connectivity limits.
The study also examined the geometric character of this capacity region under the canonical ON/OFF channel model, where transmission links fluctuate between zero and unit throughput according to independent Bernoulli processes. For this environment, the authors proved that the stability region is bounded by a collection of linear constraints whose total count equals two raised to the power of the number of users minus one. In addition, the investigation formally quantified the distance slack between an interior operating arrival vector and the exterior boundary of the capacity region. By analyzing the maximum permissible perturbation vectors across both connectivities, the team established that the aggregate decomposition slack across the two individual links never exceeds the global distance slack of the combined dual-connectivity network.
Recognizing that practical network arrival processes deviate sharply from memoryless Poisson distributions, the investigation adopted an interrupted Poisson process to capture the burstiness typical of non-full-buffer web traffic. In this traffic model, data sources alternate exponentially between active ON states and dormant OFF states. Packet arrivals follow a Poisson distribution only during active periods, creating high variance and abrupt traffic surges that severely challenge conventional queue management policies. To measure performance under these dynamic bursts, the authors incorporated System User-Perceived Throughput as a central network utility metric, which calculates the ratio of delivered service volume to the duration during which user buffers remain actively non-empty after transmission.
Building on these theoretical foundations, the researchers designed two distinct scheduling algorithms using the drift-plus-penalty technique from Lyapunov optimization theory, categorizing network behavior into two operational regimes. The first algorithm, designated Flow Control and Resource Allocation I, addresses the regime where input rates reside strictly inside the network capacity region. At each discrete time step, the algorithm minimizes a quadratic Lyapunov drift-minus-utility bound. A pivotal architectural property demonstrated in the paper is that this joint problem decouples completely into two disjoint subproblems: an upper-layer flow control routine situated at the Packet Data Convergence Protocol level and a lower-layer resource allocation routine operating at the Media Access Control level.
The decoupled flow control subproblem within the first algorithm resolves into a clean linear program that admits an exact, closed-form threshold policy. At every time slot, incoming traffic for each user is directed entirely to whichever connectivity buffer possesses the smaller instantaneous backlog. When backlogs across the primary and secondary queues are identical, the algorithm breaks the tie using a Bernoulli randomization step weighted proportionally to the number of servers installed at each base station. Concurrently, the lower-layer resource allocation subproblem maximizes a combined metric comprising current queue backlogs and a smooth, bounded surrogate function that preserves the mathematical concavity necessary to optimize user-perceived throughput without triggering combinatorial explosion.
For operational scenarios where user packet arrival rates exceed the network capacity region, the authors developed a complementary framework termed Flow Control and Resource Allocation II. In this overloaded regime, queue stability cannot be maintained without shedding load, prompting the introduction of finite buffer reservoirs that clip excess incoming data. The objective function is consequently restructured around a logarithmic throughput-fairness utility formulation. The authors showed that the resulting drift-minus-utility bound remains strictly convex, enabling exact per-slot flow control decisions via Karush-Kuhn-Tucker conditions alongside a traditional max-weight resource allocation rule at the physical and media access control interfaces.
Rigorous analytical proofs established that the first proposed algorithm achieves an explicit utility-delay trade-off governed by a tunable scalar control parameter, denoted as V. The mathematical analysis demonstrated that the time-average queue backlog across all users scales linearly with V, while the penalty gap separating the realized system utility from the theoretical optimum vanishes at a rate proportional to one divided by V. Mean rate stability was formally demonstrated across all user buffers, ensuring that average queuing delays remain strictly bounded whenever arrival vectors lie within the defined capacity region. The second algorithm was shown to maintain finite drift bounds even under sustained traffic overload through the clipping mechanism of its buffer reservoirs.
Extensive numerical simulations conducted in MATLAB over one hundred thousand scheduling slots validated the theoretical findings against traditional single-connectivity baselines. Under non-overloaded traffic with four users, the proposed dual-connectivity algorithm attained theoretical throughput limits of thirty-six packets per slot compared to twelve packets per slot for single connectivity, while balancing average traffic loads across the primary and secondary connections in close agreement with their assigned server ratios. Under severe traffic overloads, dual connectivity delivered an average throughput ranging from sixty-three to ninety-one packets per slot, roughly three times the throughput of the single-connectivity benchmark, confirming that the cross-layer Lyapunov architecture successfully expands usable capacity while providing predictable control over queuing delay.
To ground the theoretical framework before conducting large-scale multi-user network simulations, the authors detailed specific low-dimensional numerical examples that illustrate the precise geometry of capacity region expansion. When examining a baseline configuration comprising two users alongside three servers on the primary link and three servers on the secondary link, the algebraic boundary reduced to five distinct linear inequalities evaluated using Monte Carlo integration over ten thousand channel realizations. Under these conditions, the arrival rate constraints bounded individual arrivals at approximately 7.4991 packets per slot for the first user and 7.03061 packets per slot for the second user, while limiting their weighted combinations. Expanding the secondary connection to six servers substantially broadened the stable region, lifting individual user limits to 13.5016 and 12.9356 packets per slot and raising sum-rate thresholds. A three-queue scenario similarly demonstrated consistent expansion over single-connectivity boundaries across all coordinate planes.
The evaluation environment also incorporated concrete operational parameters reflecting realistic network conditions. The simulation setup examined interrupted Poisson processes configured with identical average ON and OFF period lengths of 50 time slots, generating bursty arrivals that challenge buffer stability during active states. In the non-overload simulation runs, user arrival vectors were set to [8, 4, 8, 4] under single connectivity with 10 servers and scaled to [24, 12, 24, 12] under dual connectivity with 10 primary and 20 secondary servers, maintaining an equal normalized utilization across both architectures. Individual connection throughputs converged to approximately 13.9 packets per slot for the primary connection and 22.3 packets per slot for the secondary connection, confirming practical load balancing. Under overload testing, throughput-fairness values ranged from 19.8 to 22.5 in single connectivity and from 7.1 to 8.8 in dual connectivity across parameter sweeps up to ten million.
The researchers acknowledged fundamental analytical simplifications within the modeled architecture, highlighting areas for subsequent investigation. Key assumptions included zero-delay backhaul signaling, which presumes instantaneous availability of channel and queue state information across the X2 interface despite the non-ideal latency typical of physical base station interconnects. Furthermore, the discrete-time Lyapunov drift formulation relied on quasi-static block fading where channel rates remain static within a slot but vary independently across slots. To address these operational constraints, the authors noted that future research should examine continuously time-varying wireless channels modeled through differential equations.
Subject of Research: Cross-layer scheduling and capacity region analysis for dual-connectivity multi-queue multi-server wireless communication systems
Article Title: A Scheduling algorithm for joint flow control and resource allocation for dual-connectivity MQMS systems
Article References: Li, Y., Zhang, Y., & Liang, Z. (2026). A Scheduling algorithm for joint flow control and resource allocation for dual-connectivity MQMS systems. Journal on Wireless Communications and Networking, 2026(1), Article 119. https://doi.org/10.1186/s13638-026-02656-z
Image Credits: AI Generated
DOI: 10.1186/s13638-026-02656-z
Keywords: Dual-Connectivity, Multi-Queue Multi-Server, Flow Control, Resource Allocation, Lyapunov Optimization, System User-Perceived Throughput, Wireless Networks, Capacity Region, Interrupted Poisson Process, Drift-Plus-Penalty, Packet Data Convergence Protocol, Media Access Control, Master eNodeB, Secondary eNodeB, Queue Stability, OFDM
Cite Scienmag News
Reid Dalton. (November 20, 2026). Dual-Connectivity Scheduling Framework Balances Wireless Network Throughput and Delay. Scienmag. https://scienmag.com/dual-connectivity-scheduling-framework-balances-wireless-network-throughput-and-delay/
Reid Dalton. "Dual-Connectivity Scheduling Framework Balances Wireless Network Throughput and Delay." Scienmag, 20 November 2026, https://scienmag.com/dual-connectivity-scheduling-framework-balances-wireless-network-throughput-and-delay/. Accessed 20 November 2026.
Reid Dalton. "Dual-Connectivity Scheduling Framework Balances Wireless Network Throughput and Delay." Scienmag. November 20, 2026. https://scienmag.com/dual-connectivity-scheduling-framework-balances-wireless-network-throughput-and-delay/








