Ziyan Wang, Enmao Diao, Qi Le, Pu Wang, Minwoo Lee, Shu-ping Yeh, Evgeny Stupachenko, Hao Feng, Li Yang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Integrated Access and Backhaul (IAB) represents a fast and cost-efficient network deployment technology that enhances the coverage of millimeter-wave (mmWave) 5G networks. In addition to the conventional challenges of wireless multi-hop relaying such as, e.g., increased interference and packet delays, traffic asymmetry can lead to significant delay degradation. While centralized coordination can mitigate these challenges, it may also lead to unnecessary overheads. In this paper, we propose an effective delay-aware distributed solution for joint access and backhaul link scheduling and route selection designed to function with limited information, which relies only on the knowledge collected from immediate neighbors. We formulate the joint upstream and downstream routing and scheduling problem, which is solved in a distributed manner for the IAB system with diverse delay requirements. To effectively tackle this problem, we employ deep reinforcement learning (DRL) algorithms. Our numerical results demonstrate that the proposed distributed solution provides improved scalability as compared to the centralized approach without a significant performance loss.

The use of millimeter-wave (mmWave) frequencies by 5G/5G+ technology results in increased signal attenuation naturally requiring dense network deployments. However, traditional fiber-based backhauling proves costly for network operators. To address this issue, 3GPP proposed the Integrated Access and Backhaul (IAB) concept to enable wireless backhaul and reduce deployment costs. However, system dynamics such as user mobility and traffic variations challenge system optimization and may shift the performance from its optimized state. On top of this, in-band mmWave IAB networks are subject to the half-duplex constraint, which prevents simultaneous transmission and reception. These limitations present challenges in optimizing the IAB network. Therefore, the goal of this study is to provide a computationally-efficient methodology for resource allocation and user scheduling in mm Wave IAB networks considering the aforementioned system limitations and constraints. Moreover, we evaluate the influence of system-specific factors and dynamics on the optimization of IAB networks and the time that it takes for the system to deviate from its optimized state. Our results show that by employing an optimally-parametrized scheduler, the throughput gain is 55% as compared to the baseline, where the radio resources are split equally among the users. The cell size is the primary parameter affecting the optimization gain, i.e., smaller cell sizes result in diminishing benefits when utilizing optimized algorithms.

The recent development of reinforcement learning (RL) has boosted the adoption of online RL for wireless radio resource management (RRM). However, online RL algorithms require direct interactions with the environment, which may be undesirable given the potential performance loss due to the unavoidable exploration in RL. In this work, we first explore the use ofofflineRL algorithms in solving the RRM problem. We evaluate several state-of-the-art offline RL algorithms for a practical RRM problem that aims at maximizing a linear combination of total rates and 5-percentile rates via user scheduling. Our findings indicate that the performance of offline RL for the RRM problem is heavily contingent upon the behavior policy deployed for data collection. We propose an innovative offline RL approach utilizing heterogeneous datasets from various behavior policies. This method demonstrates that a strategic mixture of datasets enables near-optimal RL policy generation, even with suboptimal behavior policies. Additionally, we introduce two enhancements: an ensemble-based policy to augment dataset mixture training efficiency, and a novel offline-to-online strategy for seamless adaptation to new environments. Our data mixture approach achieves over 95% efficiency of an online RL agent in the absence of expert data. The ensemble algorithm notably reduces training duration by half compared to the data mixture method. Furthermore, our model, when applied with offline-to-online fine-tuning, surpasses existing benchmarks by approximately 5% in our user scheduling problem.

Recently proposed by 3GPP, Integrated Access and Backhaul (IAB) technology promises to deliver a cost-efficient and flexible solution for network densification in 5G/6G systems. Since IAB architecture is based on multi-hop topology and advanced functionalities, such as multi-connectivity transmission and multi-routing, the potential utilization of IAB systems raises an issue of efficient system design. In this paper, we develop an optimization framework capable of jointly selecting transmission paths and allocating radio resources in compliance with half-duplexing and interference constraints. The presented numerical results illustrate that directional mm Wave beams employed at the wireless backhaul are essential for capacity boosting, thus allowing to fully exploit the radio resources in self-backhauled systems. We also establish that the multi-hop IAB topology provides advantages in terms of end-to-end user throughput as compared to single-hop systems.

Modern wireless infrastructures provide multiple options for a transmitter to utilize multiple independent data paths. Examples include simultaneous connections via multiple radio access technologies (multi-RAT), dual/multi-connectivity and carrier aggregation in 5G, Integrated Access and Backhaul (IAB) network, etc. Such network redundancies can be utilized to provide enhanced reliability and/or delay performances over the wireless media, e.g., to support ultra-reliable low-latency (URLLC) services. However, traditional reliability enhancement techniques fall short of making efficient use of such multi-path transmission environments. Packet-level coding, in this case, can be a better candidate to support URLLC since it provides enhanced reliability with higher spectral efficiency and low latency by proactively adding coded redundancy, which also enables the effective treatment of all paths as a single data pipe. As the multi-path scenarios are oftentimes heterogeneous in terms of supported data rate, packet-level reliability, transmission delay, etc., how to optimally design the coding parameters to achieve URLLC requirements becomes an issue. In this paper we study the problem of minimizing the transmission delay while meeting the required reliability target in the multi-path environment. We assume the packets arrive in bursts and linear packet-level coding is used to enhance reliability. We propose a fast bisection algorithm to determine the optimal code rate and path traffic distribution rule for the encoded packets, which provably achieves the minimum delay with the required reliability under very general conditions.

The high propagation losses and sensitivity to link blockage naturally require dense deployments of millimeter-wave (mmWave) 5G New Radio (NR) systems. One of the inherent challenges in these deployments is cost-efficient backhauling. Addressing this issue, 3GPP has recently proposed the concept of integrated access and backhaul (IAB) to reduce the deployment costs by enabling wireless backhaul. The efficient utilization of spectrum in these systems is conditional on the ability of IAB nodes to simultaneously receive signals on their sectoral antennas. In this paper, we investigate the interference caused by this functionality and identify countermeasures including angular and spatial diversities. Our numerical results demonstrate that the angular distance of 25° between the user equipment (UE) served by adjacent sectoral antennas is sufficient to efficiently mitigate interference. A comparable reduction in the interference level can also be achieved by utilizing spatial diversity with antenna separation of at least 20 m. By combining these methods, one can identify the target levels of angular and spatial diversities suitable for the particular deployment restrictions.

With multiple radio access technologies such as Wi-Fi, LTE, 5G available at the edge, increasingly prevalent multi-radio end devices can establish multiple concurrent connections with the edge server to deliver data traffic with more bandwidth, lower latency and higher reliability. To fully harvest multi-connectivity benefits, the edge network requires intelligent traffic management to efficiently utilize radio resources as well as to optimize quality-of-service (QoS) metrics. This paper presents a general framework for multi-connectivity traffic management at the edge, and provides algorithms for routing traffic over multiple paths to enhance the latency QoS metric. Simulation results demonstrate that, with two concurrent radio connections such as LTE and WiFi, our proposed edge traffic management algorithm can achieve significant latency reduction by more than 4× for the 95-th packet latency, and 23× in delay violation rate reduction given 10 ms latency target when compared to state-of-the-art client-based traffic management solutions.

The support of unmanned aerial vehicles, also known as drones, in cellular networks has become very important to enable a wide range of new applications for the next generation wireless systems. However, the cellular networks have been traditionally designed to serve terrestrial users, and thus are encountering many challenges to support drone wireless communication. Particularly, drones experience increased interference and channel fluctuation and consequently suffer more frequent handover, higher handover failure rate and ping-pong rate while in motion. In this paper, we propose enhanced mobility management for drones by exploiting their pre-configured flight path information. Our route-aware handover algorithm will instruct the base station to trigger the handover procedure to minimize handover failure and reduce unnecessary handover. We further provide a practical implementation where the network can make online decisions under realistic modeling assumptions. Based on 3GPP-compliant handover evaluation methodology, we demonstrate that our algorithm can effectively reduce the number of handover and handover failure by upto 32× and 24×, respectively, compared to the existing approach. This significantly reduces handover signalling overhead and service interruption time. The simulation results also show that our algorithm can completely eliminate ping-pong effect in certain cases.

Recent achievement in self-interference cancellation algorithms enables potential application of full-duplex (FD) in 5G radio access systems. FD communication promises to double the spectral efficiency by enabling the same time-frequency uplink and downlink transmissions. Yet for cellular access network with FD small cell base stations (BS) serving multiple user equipment (UE), additional BS-to-BS and UE-to-UE interference due to FD operation could diminish the performance gain if not tackled properly. Existing works all strive to achieve close to 2× FD gain at network scale, in this paper, we demonstrate that much higher FD gain (≫ 2×) can be achieved under realistic nonfull buffer traffic due to significant latency reduction. We develop practical interference mitigation strategies such as adaptive power control, joint scheduling and inter-cell interference coordination to cope with the newly introduced interference in FD cellular systems. We validate the flexible traffic adaption of FD system under bursty traffic model by our LTE-based system level simulator, where significant FD gain of upto 13v in mean throughput and 19× in cell-edge throughput can be achieved under different traffic loads over an existing LTE half-duplex system.

The use of extremely high frequency (EHF) or millimeter-wave (mmWave) band has attracted significant attention for the next generation wireless access networks. As demonstrated by recent measurements, mmWave frequencies render themselves quite sensitive to “blocking” caused by obstacles like foliage, humans, vehicles, etc. However, there is a dearth of analytical models for characterizing such blocking and the consequent effect on the signal reliability. In this paper, we propose a novel, general, and tractable model for characterizing the blocking caused by humans (assuming them to be randomly located in the environment) to mmWave propagation as a function of system parameters like transmitter-receiver locations and dimensions, as well as density and dimensions of humans. Moreover, the proposed model is validated using a ray-launcher tool. Utilizing the proposed model, the blockage probability is shown to increase with human density and separation between the transmitter-receiver pair. Furthermore, the developed analysis is shown to demonstrate the existence of a transmitter antenna height that maximizes the received signal strength, which in turn is a function of the transmitter-receiver distance and their dimensions.

As fourth generation communications technology is already being deployed, research efforts are now being shifted to what comes beyond state-of-the-art wireless systems. Driven by the anticipated acceleration in mobile traffic demand, the wireless industry is specifically focused on improving capacity and coverage of current networks through aggressive reuse of the cellular spectrum. Together with deploying an increasingly dense overlay tier of smaller cells, mobile network operators are beginning to rely on unlicensed-band WLAN technologies to leverage additional spectrum and relieve congestion on their networks. Consequently, the emerging vision of heterogeneous networks exploits the potential of a diverse range of devices requiring connectivity at different scales to augment available system capacity and improve the user connectivity experience. In this paper, we seek to meet this important trend with our novel integrated methodology for assisted (managed) radio network selection capturing spatial randomness of converged cellular/WLAN deployments together with dynamic uplink traffic from their users. To this end, we employ tools coming from stochastic geometry to characterize performance of macro and pico cellular networks, as well as WLAN, mindful of user experience and targeting intelligent network selection/assignment. We complement our analysis with system-level simulations providing deeper insights into the behavior of future heterogeneous deployments.

Multi-radio heterogeneous networks (Het-Nets) are an important focus area for next generation cellular standards, with the 3GPP community actively developing WiFi/LTE interworking solutions for small-cell deployments. This paper explores coordinated usage of multiple radios (e.g., LTE and WiFi) to improve quality of service (QoS) in multi- radio heterogeneous networks. We focus on heterogeneous network deployments based on co- located WiFi-LTE small cells that allow for tighter multi-radio coordination. To evaluate QoS enhancements, we consider on-time throughput, a metric that captures the deliverable data rate for traffic with delay deadlines. A QoS-aware scheduling algorithm is designed to optimize on-time throughput. Cross-RAT coordination schemes are also explored to further enhance QoS. Through combining the QoS-aware scheduling algorithm and intelligent radio link assignment, we observe significant improvement in per user on-time throughput. Our results indicate that tightly coupled integrated LTE-WiFi small cell architectures can significantly increase the number of user achieving their targeted QoS.

We develop cognitive resource allocation scheme to mitigate co-tier and cross-tier interference in overlay femtocell networks. By exploiting statistical models for characterizing multitier interference, our scheme avoids prohibitive exchange of realtime interference statistics in the network. The proposed scheme is independently implemented at each femtocell and allocates resources distributedly in response to the probabilistic interference conditions in the network. This “self-organizing” framework can be useful to address interference management in dense, ad-hoc, and consumer-deployed femtocell networks. Simulation results show that the proposed scheme can improve the throughput of femtocell links while simultaneously reducing cross-tier interference.

This paper analyzes the capacity of a wireless relay network composed of a large number of nodes that operate in an amplify-and-forward mode and that divide into a fixed number of levels. The capacity computation relies on the study of products of large random matrices, whose limiting eigenvalue distribution is computed via a set of recursive equations.