Integrated Access and Backhaul (IAB)

• IAB is a networking paradigm that integrates access and backhaul functions using shared radio spectrum for cost-effective, flexible 5G deployments.
• It employs TDM and joint optimization of resource allocation, routing, and relay selection to maximize fairness and throughput for users.
• Mesh-based IAB designs improve latency and load balancing over traditional spanning trees, enabling incremental fiber deployment and robust performance.

Integrated Access and Backhaul (IAB) is a network architecture and resource management paradigm in which the same wireless infrastructure, spectrum, and protocol stack are jointly used for both user access and backhaul connectivity among base stations. This unified approach enables multi-hop wireless relay of backhaul traffic for 5G and beyond networks, drastically reducing the need for extensive fiber deployment, lowering costs for ultra-dense small cell deployments, and increasing coverage and flexibility. IAB networks typically operate in millimeter-wave (mmWave) frequency bands, leverage 3GPP New Radio (NR) standards, and require sophisticated joint optimization of resource allocation, relay (backhaul) topology formation, and fairness measures to deliver high quality of service under diverse physical and deployment constraints.

1. Architectural Principles and Optimization Framework

IAB networks comprise access links (between UEs and base stations) and backhaul links (between base stations and core network aggregation points or “anchor” nodes). The defining aspect is the integration of access and backhaul—both functions share the same radio technology, protocol stack, and often the same spectrum resources (e.g., in-band 5G NR). This architectural coalescence supports interoperability across vendors and offers flexibility for incremental fiber deployment where only a subset of base stations have wired backhaul.

The network performance objective is typically to maximize a network-wide utility metric that reflects both throughput and fairness among user equipments (UEs). In (Islam et al., 2018), the target is the geometric mean (GM) of UE rates, defined as:

$$\mathrm{GM} = \left[ \prod_{i \in \mathcal{UE}} \left( \sum_{j \in \mathcal{BS}} f^U_{ij} \cdot \sum_{j \in \mathcal{BS}} f^D_{ji} \right) \right]^{\frac{1}{2N}}$$

where $f^U_{ij}$ and $f^D_{ji}$ are the uplink and downlink flows between UE $i$ and BS $j$, and $N$ is the number of UEs. The multiplicative nature of GM heavily penalizes low-performing UEs, driving a strong form of fairness.

The resource allocation problem is subject to layered constraints:

• Time Division Multiplexing (TDM): Access and backhaul links share time-slotted resources. The time allocated to all links at a node must not exceed unity.
• Capacity and Flow: Each link’s flow is upper bounded by the product of its allocated time, channel capacity, and a binary activity indicator (e.g., $f^U_{ij} \leq t^U_{ij} c_{ij} a_{ij}$).
• Backhaul Routing/Relay Selection: A feasible topology must be established such that each non-anchor BS is connected (potentially multi-hop) to an anchor node.

The optimization jointly determines link time allocations, flow assignments, and the multihop routing of backhaul traffic.

2. Relay Selection and Topology Design

Two principal topology formation paradigms are contrasted:

• Signal-Strength (RSRP)-Based Spanning Tree (ST): Starting from fiber-connected (anchor) BSs, each non-anchor iteratively attaches to the anchor or parent with highest reference signal power, resulting in a directed spanning tree (each node with a unique parent). This minimizes individual hop loss but does not account for aggregate load and can produce highly asymmetric trees with poor load balancing—some donors bear disproportionate relay and UE loads, inflating hop counts and latency.
• Load-Balanced Optimal Mesh: Any BS can connect to any other to form a mesh, enabling multiple routing paths. A joint optimization over routes and resource allocations allows balanced traffic loads, minimized hop counts, and maximized user rate distribution. This mesh topology, computed with knowledge of network-wide capacities and constraints, outperforms the spanning tree in both rate and latency by distributing traffic more evenly and exploiting path diversity.

The routing and resource allocation are found by solving a convex (or mixed-integer) program, taking into account per-link constraints and topology connectivity enforcement, resulting in globally optimal or near-optimal solutions under the model assumptions.

3. Rate, Latency, and Incremental Deployment Benefits

Numerical and simulation studies reveal profound performance advantages of IAB relative to legacy access-only architectures:

• UE Rate Improvement: In canonical scenarios with 7 of 18 sites fiber-connected, the geometric mean UE rate is nearly doubled with IAB compared to access-only networks. Notably, even if top-decile UEs (closest to anchors) show marginal difference, IAB substantially uplifts the rates for the bottom 90%—those most penalized by distance or poor channel conditions.
• Latency Reduction: Mesh-based IAB topologies with optimized routing balance hops and traffic, resulting in consistently lower end-to-end latencies compared to spanning tree structures, which can create long paths for some UEs.
• Incremental Fiber Deployment: IAB’s core benefit is its ability to serve a large area with minimal initial fiber investment. As more anchors are connected over time, IAB seamlessly migrates toward a full-fiber configuration, providing an effective intermediate solution with strong continuity of user experience.

4. Performance Metric: Geometric Mean UE Rate

The geometric mean is adopted due to its sensitivity to user-level discrepancy in throughput, promoting fairness and ensuring improvements for low-rate users are not masked by a few high-rate flows. For a network with $N$ UEs, the metric is:

$$\mathrm{GM} = \left[ \prod_{i=1}^{N} (\text{UL rate}_i \cdot \text{DL rate}_i) \right]^{1/(2N)}$$

This contrasts with the arithmetic mean, which may allow the network to optimize for a small subset of UEs at the expense of others. By maximizing GM, the optimization inherently targets a more equitable and robust network-wide performance.

5. Resource Allocation and Routing Methodologies

The joint optimization approach integrates several technical mechanisms:

• TDM Resource Sharing: Time is allocated between access and backhaul per-link and per-node, ensuring no oversubscription.
• Link Activation: Binary variables indicate activation; only feasible links contribute to flows.
• Capacity and Conservation Constraints: Each link’s flow does not exceed capacity, and conservation of flow is maintained at each node.
• Routing: For multi-hop topologies, flow conservation and connectivity constraints guarantee all UEs’ traffic can reach an anchor via selected paths.

The resultant mixed-integer optimization is typically solved via convex relaxation or decomposition techniques, yielding time and flow allocations and explicit routing assignments.

6. Deployment Implications and Design Trade-offs

IAB networks deliver the most pronounced gains in environments where fiber is costly or cannot be ubiquitously deployed. The ability to flexibly route backhaul traffic over multi-hop wireless links—dynamically adjusted as fiber infrastructure is extended—gives operators enhanced agility to balance cost, capacity, and user experience.

Trade-offs inherent to IAB deployment include:

• Complexity of Mesh Construction: While mesh topologies maximize performance, their computation and management overheads are higher than simple trees, necessitating advanced controllers and/or distributed algorithms.
• Resource Partitioning: Backhaul and access share the same medium, so resource multiplexing strategies must avoid starvation and guarantee per-user QoS.
• Topology Selection: Suboptimal relay selection (e.g., pure RSRP-based spanning tree) can lead to subpar performance; mesh optimization is critical, especially in asymmetric user/geography settings.

7. Significance for 5G and Beyond

Integrated Access and Backhaul, by blending access and backhaul into a unified, wireless-first architecture, underpins the densification needed for 5G urban/heterogeneous deployments without the prohibitive cost of all-fiber backhaul. Its design, driven by joint resource and topology optimization, ensures enhanced rates, lower latency, and improved fairness, particularly during transitional deployment phases. The explicit use of geometric mean rate as a metric and the superiority of mesh-optimized topologies shape future research and industrial deployment strategies, positioning IAB as a cornerstone technology for the evolution of radio access networks.