Source-linked AI summary
An Overview of Load Balancing in HetNets: Old Myths and Open Problems
Jeffrey G. Andrews, Sarabjot Singh, Qiaoyang Ye, Xingqin Lin, Harpreet Dhillon
TL;DR
Dense small-cell networks require proactive load balancing to meet applications’ rate and latency needs. The paper examines optimization-based load balancing and design questions including interference, static cell range expansion, and switching off macrocells, while recognizing implementation and association constraints.
Problem
Dense small-cell networks require proactive load balancing while jointly delivering the rate and latency that users’ applications require.
Method
The paper examines optimization-based load balancing and design choices involving load utility, co-channel interference, and cell operation.
Results
The paper reports that static cell range expansion can differ from the globally optimal solution and that macrocells should be shut off about half the time.
Takeaways & Limitations
Load balancing in small-cell networks requires proactive design attention to time and space, interference, and cell activation decisions.
Takeaways & Limitations
Accurately stating load remains constrained by implementation considerations, including cancellation-related constraints and the association definition.
Abstract
from arXiv · showhide
Matching the demand for resources ("load") with the supply of resources ("capacity") is a basic problem occurring across many fields of engineering, logistics, and economics, and has been considered extensively both in the Internet and in wireless networks. The ongoing evolution of cellular communication networks into dense, organic, and irregular heterogeneous networks ("HetNets") has elevated load-awareness to a central problem, and introduces many new subtleties. This paper explains how several long-standing assumptions about cellular networks need to be rethought in the context of a load-balanced HetNet: we highlight these as three deeply entrenched myths that we then dispel. We survey and compare the primary technical approaches to HetNet load balancing: (centralized) optimization, game theory, Markov decision processes, and the newly popular cell range expansion (a.k.a. "biasing"), and draw design lessons for OFDMA-based cellular systems. We also identify several open areas for future exploration.
1 Myth One: Signal Quality is the Main Driver of User Experience
The paper challenges the assumption that signal quality alone predicts user experience, arguing that base-station load and resource allocation over time are critical in heterogeneous networks.
- Network heterogeneity: HetNets complicate wireless systems through diverse base stations, access points, technologies, and frequency bands.Users may be within range of many jointly available network resources.
- The signal-quality myth: Myth 1 treats received SINR as the primary predictor of user experience or link reliability.Traditional relationships connect SINR with bit error rate, detection probability, outage, and achievable rate.
- The signal-quality myth: User-perceived rate depends on instantaneous rate multiplied by the fraction of resources allocated to that user.Under proportional-fair or round-robin scheduling, the resource share is about 1/K, where K is the number of other active users on the base station in that band.
- Load awareness: Base-station load varies spatially and temporally, making it difficult to determine a priori and difficult to model accurately.Load relates to coverage area but also depends on user distribution, traffic models, and other extrinsic factors.
- Load awareness: Load-aware models provide approaches for analyzing cellular networks while exposing the limitations of load-blind models.The paper presents these approaches as a main goal.
2 Myth 2: The “Spectrum Crunch”
The paper disputes the idea that wireless broadband faces primarily a spectrum crunch, emphasizing infrastructure deployment and proactive load balancing as central capacity concerns.
- The spectrum-crunch myth: Myth 2 claims that regulators must urgently release much more spectrum to improve wireless broadband user experience.This claim is framed as the “spectrum crunch” myth.
- The spectrum-crunch myth: In 2012, global mobile data traffic more than doubled for the fifth consecutive year, while projected spectrum additions remained comparatively limited.The FCC was considering 500 MHz of new spectrum by 2020, less than twice the amount available in 2010.
- The spectrum-crunch myth: The paper reports a shortfall of more than 500x between projected demand and available broadband spectrum.This observation motivates the paper’s reassessment of the spectrum-crunch explanation.
- Infrastructure shortage: The paper argues that the central problem is an infrastructure shortage rather than a spectrum shortage.Small cells are presented as a key element for increasing cellular capacity.
- Infrastructure shortage: Because small cells are fixed and opportunistically deployed while users and resource demands move and fluctuate, their offered load varies across time and space.The paper therefore calls for more proactive load balancing to use newly deployed infrastructure effectively.
- Design implication: Load modeling and optimization should receive status comparable to spectrum and SINR, although doing so rigorously is not straightforward.The paper identifies technical difficulty as a continuing issue.
3 Technical Approaches to Load Balancing
HetNet load balancing requires approaches that account for coupled association, scheduling, interference, and uneven base-station capabilities. The paper surveys optimization, MDPs, game theory, biasing, and stochastic-geometry methods, highlighting trade-offs between optimality, complexity, overhead, and adaptability.
- Motivation: Natural SINR or RSSI association can create major load imbalance because heterogeneous base stations have disparate transmit powers and capabilities.In the illustrated three-tier HetNet, max-SINR association leaves some small base stations idle while macro base stations serve most users.
- Optimization: Jointly optimizing user association and scheduling is combinatorial, NP-hard, and exponentially more complex as the network grows.Dynamic traffic further complicates this long-standing load-balancing problem.
- Relaxed Optimization: Relaxing binary association and applying dual decomposition can produce a low-complexity distributed algorithm that converges to a near-optimal solution.The relaxation assumes a fully loaded model and allows users to associate with multiple base stations, which upper-bounds binary-association performance.
- Stochastic Geometry: A 3.5x rate gain for cell-edge users and a 2x rate gain for median users is reported versus maximum received-power association.Stochastic geometry models user and base-station locations with point processes to obtain tractable SINR and rate expressions and study system-level parameters.
- Markov Decision Processes: MDPs model sequential decisions under uncertainty, but large heterogeneous networks and continuous state spaces make exact solution difficult.They have been used for handoff, cellular-to-WiFi offloading, and association assisted by broadcast load information.
- Game Theory: Game theory analyzes decentralized association decisions, but convergence is not generally guaranteed and convergent algorithms need not be optimal.The paper presents game theory as potentially insightful for uncoordinated users and base stations, despite possible overhead and inefficiency.
- Cell Range Expansion: Cell range expansion uses biased received power to proactively offload users to lower-power cells and can nearly achieve optimal load-aware performance when biases are chosen carefully.The method expands small-cell coverage through tier-specific bias values and is part of 3GPP standardization efforts.
4 System Design Principles
HetNet load balancing requires biasing and interference management to be designed together, with optimal settings depending on deployment type, density, and blanking. The resulting rules challenge assumptions that adding small cells or shutting down macrocells necessarily harms network performance.
- Bias Values: Co-channel biasing typically uses 5–10 dB, whereas out-of-band offloading can require 20 dB or more.Out-of-band offloading avoids the strong co-channel interference source in the new band, allowing more aggressive biasing.
- Blanking: Almost blank subframes let offloaded users avoid co-channel macro-tier interference during muted macrocell slots.Macro BSs are periodically muted, and offloaded users can be scheduled in the blanked slots.
- Blanking: With five picocells per macrocell, the optimal bias rises from about 6 dB to 20 dB as blanking increases, reaching roughly 16 dB at the optimum.This result assumes offloaded users are served only in blanked time slots.
- Blanking: When offloaded users can also use normal slots, optimal blanking grows with small-cell density and is approximately one half for plausible deployments.Macrocells should be muted about half the time because they are also the biggest interferers.
- Biasing as Small Cell Density Increases: As small-cell density increases, optimal bias decreases for out-of-band offloading but is unchanged for co-channel offloading.Greater density improves nearby small-cell connectivity in both cases; in co-channel deployments, added interference affects all users equally.
- Myth 3: Adding randomly deployed base stations does not decrease SIR under max-SIR association, and adding base stations can increase the rate CDF even when SIR falls after biasing.The paper identifies a qualification: biasing can depart from max-SIR association and thereby lower SINR, increasing the value of interference management.
5 Open Challenges
HetNet load balancing remains incompletely understood, with simplified models and implementation constraints limiting current analyses. Open challenges span biasing robustness, mobility, uplink/downlink asymmetry, inter-network coordination, and deployment policy.
- Modeling assumptions: Existing HetNet load-balancing analyses often simplify user distributions, antenna configurations, transmit power, and scheduling to remain tractable.These assumptions may not be realistic, so the sensitivity of biasing and its gains requires deeper study.
- Biasing robustness: Biasing may need to adapt to user distributions and current base-station load rather than rely only on per-tier values.The paper specifically suggests per-BS bias values if optimum biasing is sensitive to spatio-temporal user distributions.
- Implementation constraints: Backhaul constraints can reduce desired offloading once small cells exceed a backhaul-dependent load threshold.This motivates incorporating backhaul limitations into the associated bias value.
- Mobility: Mobility makes handovers a load-balancing trade-off because short small-cell sojourns may not justify costly procedures into and out of the cell.Supporting seamless handovers among heterogeneous cell types remains essential.
- UE capability and interference: Biasing lowers SINR and can become infeasible when user equipment cannot decode the lowest-rate modulation, while offloaded users may face much higher interference.LTE-A throughput with a typical codeset is zero below about -6.5 dB; backhaul and interference constraints also limit practical offloading.
- Uplink and downlink: Downlink-optimal association need not be uplink-optimal because HetNets have asymmetric powers, coverage areas, and traffic volumes.The paper calls for extending downlink work to uplink scenarios and ideally studying both jointly.
- Network integration: Jointly exploiting small-cell and D2D offloading remains unresolved because D2D mode selection is coupled with access-point association.The paper also identifies seamless cellular–WiFi handoff and improved WiFi MAC efficiency as open issues.
- Deployment and policy: Future progress also depends on making small-cell deployment easier through spectrum, backhaul, access, compensation, and incentive policies.The paper highlights open-access deployment for femtocells and WiFi and a more cellular-like OFDMA-based WiFi scheduler.