Source-linked AI summary
Joint Rate and SINR Coverage Analysis for Decoupled Uplink-Downlink Biased Cell Associations in HetNets
Sarabjot Singh, Xinchen Zhang, Jeffrey G. Andrews
TL;DR
The paper addresses limited understanding of uplink and joint uplink-downlink performance under load balancing, association, and power control in multi-tier HCNs. It develops an analytical model for uplink SINR and rate distributions and joint coverage, finding distinct uplink behavior and benefits from decoupled association. The analysis also identifies fixed parametric assumptions as a scope boundary.
Problem
Prior work had limited uplink analysis and did not characterize load balancing's impact on uplink rate distribution or joint uplink-downlink rate coverage in general HCNs.
Method
The paper develops a model for uplink SINR and rate coverage in K-tier HCNs, incorporating offloading, fractional power control, association rules, and joint uplink-downlink coverage.
Results
Decoupled uplink-downlink association significantly improves joint rate coverage over standard coupled association, while minimum path loss association is optimal for uplink rate and full channel inversion yields density-invariant uplink SIR.
Takeaways & Limitations
Uplink and downlink rate distributions respond differently to load balancing, and association strategies should distinguish uplink from downlink objectives.
Takeaways & Limitations
The analysis assumes parametric but fixed partitioning and does not analyze more dynamic, possibly load-aware partitioning.
Abstract
from arXiv · showhide
Load balancing by proactively offloading users onto small and otherwise lightly-loaded cells is critical for tapping the potential of dense heterogeneous cellular networks (HCNs). Offloading has mostly been studied for the downlink, where it is generally assumed that a user offloaded to a small cell will communicate with it on the uplink as well. The impact of coupled downlink-uplink offloading is not well understood. Uplink power control and spatial interference correlation further complicate the mathematical analysis as compared to the downlink. We propose an accurate and tractable model to characterize the uplink SINR and rate distribution in a multi-tier HCN as a function of the association rules and power control parameters. Joint uplink-downlink rate coverage is also characterized. Using the developed analysis, it is shown that the optimal degree of channel inversion (for uplink power control) increases with load imbalance in the network. In sharp contrast to the downlink, minimum path loss association is shown to be optimal for uplink rate. Moreover, with minimum path loss association and full channel inversion, uplink SIR is shown to be invariant of infrastructure density. It is further shown that a decoupled association---employing differing association strategies for uplink and downlink---leads to significant improvement in joint uplink-downlink rate coverage over the standard coupled association in HCNs.
I. INTRODUCTION
Uplink analysis in heterogeneous cellular networks is less developed than downlink analysis because uplink users, power control, and interference are spatially coupled. The paper motivates models that capture how association and load balancing affect uplink and joint uplink-downlink performance.
- Motivation: Uplink HCN analysis is limited despite the importance of uplink performance in cloud storage and video-chat services.Video chat requires quality of service in both directions because its traffic is symmetric.
- Why uplink analysis is difficult: Downlink insights cannot be directly extrapolated to uplink because uplink users, power control, and interference correlations differ fundamentally.Uplink interference comes from one associated UE per AP on a resource, rather than Poisson-distributed transmitters.
- Association and load balancing: Downlink-oriented biasing may improve downlink rate and uplink SNR, but its optimality for uplink rate remains uncertain.Offloaded users are more likely to transmit to nearby small cells, while downlink biasing is not established as optimal for uplink.
- Analytical gap: Existing uplink generative models often cover only special cases and may ignore conditioning that affects performance estimates.Prior limitations include macro-only single-tier systems, truncated full channel inversion, nearest-AP association, and PPP approximations for interfering users.
- Paper scope: The paper seeks a general analysis of uplink and joint uplink-downlink coverage under arbitrary association and tier count, including decoupled associations.Joint analysis requires characterizing the correlation between uplink and downlink path losses when different APs serve a user.
B. Contributions and outcomes
The paper develops a tractable uplink model for multi-tier HCNs and derives joint uplink-downlink rate coverage under flexible association and power-control settings. Its analysis yields design insights favoring minimum path loss for uplink rate and decoupled associations for joint coverage.
- Uplink SINR and rate distribution: The proposed generative model represents interfering UE locations as an inhomogeneous PPP whose intensity depends on association parameters.It also captures correlation between each interfering UE’s transmit power and its path loss to the tagged AP.
- Uplink SINR and rate distribution: Uplink SINR and rate CCDFs are derived for a K-tier HCN as functions of association and power-control parameters.The general analysis is simplified for selected scenarios, with upper and lower bounds also derived.
- Joint uplink-downlink rate coverage: Joint rate/SINR coverage is derived as the probability that uplink and downlink rates or SINRs exceed their respective thresholds.The derivation combines uplink coverage with the joint distribution of uplink-downlink path losses, while assuming independent uplink and downlink interference for tractability.
- Insights: The PCF maximizing uplink SIR coverage is inversely proportional to the SIR threshold, so edge users prefer a higher PCF than cell-interior users.The optimal PCF also increases with disparity in association weights across tiers.
- Insights: Minimum path loss association yields optimal uplink rate coverage, contrasting with the corresponding downlink result.With minimum path loss association and full channel inversion, uplink SIR coverage is independent of infrastructure density and is stochastically dominated by downlink SIR.
- Insights: With a static resource-allocation ratio, separately optimal uplink and downlink association weights also maximize joint coverage, making decoupled association optimal for joint rate coverage.The analysis is validated against simulations across a wide range of parameter settings.
II. SYSTEM MODEL
The system model represents a co-channel K-tier HCN with Poisson-distributed APs and UEs, Rayleigh fading, weighted path-loss associations, fractional power control, and orthogonal uplink access. It defines resource allocation and rate through per-AP user sharing under potentially different uplink and downlink association weights.
- Network and channel model: APs in tier k form independent homogeneous PPPs with density λ_k, while UEs form an independent homogeneous PPP with density λ_u.Tier-k APs transmit with power P_k.
- Network and channel model: Signals use path loss exponent α, fast fading H following unit-mean exponential distributions, and large-scale fading S in the path-loss model.Received power is modeled as transmit power times fading gain divided by path loss.
- Uplink power control: Uplink transmission uses fractional path-loss-inversion power control with power-control fraction ϵ between 0 and 1.The per-user maximum power constraint is ignored for tractability, though an extension is noted when transmit-power dependence on load is ignored.
- Uplink access and interference: Orthogonal uplink access permits only one UE per AP to transmit on a resource block, so cochannel interferers form a Poisson-Voronoi perturbed lattice rather than a PPP.The tagged AP is the serving AP of a typical UE located at the origin.
- Association model: Weighted path loss determines uplink and downlink associations, with potentially different per-tier weights for the two directions.Identical weights across tiers produce minimum path loss association, while biased association is included through transmit-power and bias weights.
- Resource allocation and rate: The typical user’s rate depends on bandwidth, the allocated uplink or downlink resource fraction, and equal sharing among associated users.The uplink resource fraction is η, while the downlink fraction is 1−η; AP downlink queues are saturated.
III. UPLINK SINR AND RATE COVERAGE
The paper derives uplink SINR and SIR coverage using a generative approximation for the structured interfering-UE process and path-loss distributions. The resulting expressions include general integral forms, bounds, and an exact special case for full channel inversion.
- A. General case: The model accounts for the fact that an interfering UE’s serving-link path loss is conditioned on its not associating with the tagged AP.This dependence is formalized through a conditional path-loss distribution.
- A. General case: The uplink coverage analysis models interfering UE propagation processes using association-dependent thinning and an inhomogeneous point process.The resulting intensity increases with path loss from the interfering UE to the tagged AP.
- A. General case: Tier-wise independence and independent path-loss assumptions make the interfering-UE process analytically tractable.The interference Laplace transform is then derived conditional on the serving tier.
- A. General case: Theorem 1 gives uplink SINR coverage, while SIR coverage follows by taking SNR to infinity.The analysis also derives corresponding coverage expressions for the proposed generative model.
- A. General case: The general coverage expression involves two folds of integrals and a Hypergeometric-function lookup table, motivating simpler bounds and special-case forms.A single-integral upper bound and closed-form lower bound are provided.
- A. General case: The coverage upper bound is exact for full channel inversion, corresponding to ϵ = 1.The desired-link path-loss distribution is characterized separately for the typical UE.
B. Special cases
Special cases characterize uplink SIR coverage under different power-control and association settings, including density-invariant behavior with minimum path loss association and full channel inversion.
- Minimum path loss association in a K-tier network has uplink SIR coverage equivalent to a single-tier network with aggregate density λ = Σ_K λ_k.
- Without uplink power control, the analysis gives a separate uplink SIR coverage expression for ε = 0.
- With full channel inversion, the analysis gives a separate uplink SIR coverage expression for ε = 1.
- With full channel inversion and minimum path loss association, uplink SIR coverage is independent of infrastructure density in HCNs.
- The downlink SIR stochastically dominates the uplink SIR under the compared association setting.
C. Uplink rate distribution
The uplink rate analysis combines SINR with serving-AP load while accounting for their correlation through association regions. A mean-load approximation simplifies the coverage expression at a small accuracy cost.
- Uplink rate depends on both SINR and the load at the tagged access point.
- Theorem 2 gives uplink rate coverage under the presented system model and assumptions.
- The tractable rate analysis ignores load–SINR dependence and thermal noise to obtain a simplified expression.
- Association regions create correlated load and SINR because larger regions imply higher load and larger user-to-AP distances.
- Corollary 11 approximates each tier's access-point load by its respective mean, ¯N_k = E[N|K = k].
- The mean-load corollary eliminates a sum from Theorem 2's rate-coverage expression while sacrificing some accuracy.
IV. JOINT UPLINK-DOWNLINK RATE COVERAGE
The paper derives joint uplink-downlink rate coverage by characterizing correlated path losses and combining rate and interference models under mean-load and interference-independence assumptions.
- Joint rate coverage is the probability that both uplink and downlink rates exceed their respective thresholds.
- For coupled association, uplink and downlink path losses are identical; under general association policies, they are correlated.
- The analysis characterizes the joint path-loss distribution for arbitrary downlink and uplink association weights.
- Joint coverage uses downlink interference Laplace transforms conditioned on the serving tier and path loss.
- Theorem 3 derives joint uplink-downlink rate coverage using mean uplink and downlink loads and independent uplink-downlink interference.
- The final joint-rate expression combines joint SIR coverage with the rate model and joint path-loss distribution.
V. VALIDATION THROUGH SIMULATIONS
Simulations validate the proposed uplink SINR, rate, and joint-rate analyses across multiple network settings. The results support the modeling assumptions and show that simplifying interference-user modeling can significantly distort coverage.
- The proposed model and analytical results are validated by simulations.
- Analysis and simulation compare uplink SINR distributions across association weights and tier-2 densities, including two simplifications for ε = 1.
- Neglecting the proposed thinning or conditioning produces significant deviation from actual coverage.
- Thermal noise has minimal impact on uplink SINR in the evaluated settings.
- Rate and joint-rate distributions show close analysis–simulation agreement across two-tier and three-tier settings, supporting the mean-load and interference-independence assumptions.
VI. OPTIMAL POWER CONTROL AND ASSOCIATION
The analysis identifies how power-control and association choices affect uplink and joint uplink-downlink coverage. Minimum path loss association is optimal for uplink rate, while decoupled association substantially improves joint rate coverage.
- Uplink association: Minimum path loss association maximizes uplink SIR coverage and uplink rate across power-control factors and densities.
- Optimal power control: The optimal PCF decreases with SIR threshold because greater channel inversion benefits cell-edge users with higher path loss.
- Optimal power control: Higher association-weight imbalance uniformly increases the optimal PCF because network path losses increase.
- Uplink rate: A higher PCF maximizes fifth-percentile uplink rate rather than median rate because edge users experience lower uplink SIR.
- Uplink association: Association-weight changes have nominal effects on uplink SIR because increased path loss and reduced interference from larger association areas compensate each other.
- Joint uplink-downlink association: Decoupled association provides an approximately 1.5x joint-rate gain over coupled association across power-control factors.
VII. CONCLUSION
The paper develops and validates a tractable model for uplink SINR and rate distributions in multi-tier HCNs with offloading and fractional power control. It derives association insights for uplink and joint uplink-downlink coverage while identifying fixed resource partitioning and architectural costs as boundaries for further analysis.
- The paper proposes a model for uplink SINR and rate coverage in K-tier HCNs incorporating offloading and fractional power control.
- The work derives and validates uplink rate distributions with load balancing and contrasts uplink and downlink behavior under load balancing.
- The tractable SINR and rate distributions enable derivation of optimal association weights for uplink and joint uplink-downlink coverage.
- The analysis assumes parametric but fixed resource partitioning between uplink and downlink, leaving dynamic load-aware partitioning for future work.
- Performance analysis incorporating the cost of architectural changes required by decoupled association remains a future investigation.
APPENDIX A
The appendix derives interference-transform and coverage-bound expressions using the assumed point-process model, change of variables, and Jensen’s inequality.
- The derivation represents interference through a Laplace transform for interfering users associated with a tagged access point.
- Independence of fading and the probability generating functional of the assumed PPP are used to obtain the interference expression.
- A change of variables and Gauss-Hypergeometric-function identity produce the final closed-form result.
- Jensen’s inequality converts the inner expectation into a bound on coverage probability.
- Neglecting conditioning yields the lower-bound derivation used for the simplified coverage expression.