A Stochastic Geometry Framework for Analyzing Pairwise-Cooperative Cellular Networks

I. INTRODUCTION

• Full cooperation requires channel information at all transmitters, but the resulting feedback burden can consume considerable system resources.Grouping BSs into cooperative clusters may also require a centralized controller with network-wide channel-quality information.Source: §I · p. 2

A. Related Literature

• Prior work largely studied small cooperating-transmitter models, leaving network-level coverage analysis with inter-cluster interference insufficiently investigated.This work targets coverage in a network with an infinite number of nodes.Source: A. Related Literature · p. 3

B. Work Presentation and Major Contributions

• The paper develops a stochastic-geometry framework for pairwise cellular cooperation, with each user choosing its two geographically closest BSs rather than relying on a central controller.It derives exact closed-form or integral expressions for network performance when nodes are randomly distributed over the infinite plane.Source: B. Work Presentation and Major… · p. 4
• The analysis limits exchanged channel information to phase, enabling in-phase reception without amplitude exchange and avoiding the often-unrealistic feedback required for full channel knowledge.Full channel knowledge and beamformer adaptation are left for future investigation and comparison.Source: B. Work Presentation and Major… · pp. 4–5
• User-optimal geometric policies determine whether to cooperate, and varying ρ changes coverage-region shapes while favoring areas near cell boundaries.The policies are based on the relative distances to the two nearest BSs; ρ controls the width of the cooperation zone.Source: B. Work Presentation and Major… · pp. 5–6
• The work specifies the scenario’s SINR and derives explicit typical-point coverage expressions for threshold T as a function of ρ.Source: B. Work Presentation and Major… · p. 5
• A cooperation-aware shot-noise model provides the interference random variable, its Laplace transform, and its expected value; schemes share the same mean but differ in higher-order behavior.Source: B. Work Presentation and Major… · pp. 5–6
• The coverage expressions with and without DPC are evaluated numerically, and simulations are compared with theory to validate the results.Source: B. Work Presentation and Major… · p. 6
• Cooperation substantially benefits network coverage with limited channel adaptation and two neighboring BSs, while DPC significantly increases those benefits.DPC can eliminate the second-neighbor signal when its interference is measured and known to both transmitters.Source: B. Work Presentation and Major… · p. 6
• The framework combines k-Voronoi geometry for identifying optimal antenna clusters with Poisson stochastic geometry, and can extend to larger cooperating groups and varied schemes or assumptions.Source: B. Work Presentation and Major… · p. 6

C. Organization of the work

• The paper first presents the cooperation model, power splitting, and geometric policies, then derives coverage-probability expressions using stochastic geometry.The coverage analysis includes the full-cooperation signal, interference, and a variant eliminating second-neighbor interference with DPC.Source: C. Organization of the work · pp. 6–7
• The later sections compare cooperative policies with the non-cooperative model and evaluate analytical expressions against simulations.The paper then presents its conclusions; supplementary material and omitted proofs are in the Appendix.Source: C. Organization of the work · p. 7

A. Geometry and Model Assumptions

• Base stations lie at points of a planar Poisson process, and one randomly located user occupies each base station’s 1-Voronoi cell.A user may be served by its nearest base station alone or by its first and second closest base stations.Source: A. Geometry and Model Assumptions · pp. 7–9
• A 2-Voronoi cell groups points for which a given pair of base stations are the two closest; such a cell is non-empty exactly for Delaunay neighbors.The 2-Voronoi cells partition the plane, up to intersections of zero Lebesgue measure.Source: A. Geometry and Model Assumptions · p. 8

B. Downlink Cooperative Transmission in Pairs with Power Splitting

• The cooperating pair has channel-phase information and, in the DPC case, second-base-station interference, but cannot adapt fully to instantaneous channel conditions.The stated purpose is to analyze cooperation with limited information exchange.Source: B. Downlink Cooperative Transmission… · pp. 10–11
• Each user's signal is split into a private part from its first base station and a common part shared by both serving base stations over backhaul.The second base station sends no private part for that user.Source: B. Downlink Cooperative Transmission… · p. 11
• The serving base stations divide the user's total power p between them, with the first using (1 − ai)p and the second using aip.The power-split ratio may differ across users.Source: B. Downlink Cooperative Transmission… · p. 11
• Equal phases at the two serving links make the common signal arrive in phase and maximize the beneficial signal for the user.The model assumes these phases are available at the transmitters.Source: B. Downlink Cooperative Transmission… · p. 13
• Averaging the interference over the other users' channel phases produces coverage results almost identical to those from the original SINR formulation.The extra interference term disappears under this expectation because the phase-difference cosine has expectation zero.Source: B. Downlink Cooperative Transmission… · p. 13

C. A Family of Geometric Policies with either Full or No Cooperation

• The geometric policy chooses Full Coop or No Coop according to a global parameter ρ and the user’s relative distances to its two closest base stations.For ρ = 0, Full Coop applies throughout the 2-Voronoi cell; for ρ = 1, No Coop applies throughout it.Source: C. A Family of Geometric Policies with… · p. 14
• For ρ ∈ [0, 1), the locus satisfying r1 ≤ ρr2 is a disc, and the Full Coop region lies outside it within the 2-Voronoi cell.As ρ grows, the Full Coop region becomes thinner and tends to disappear near the 1-Voronoi boundary.Source: C. A Family of Geometric Policies with… · p. 15

III. STOCHASTIC GEOMETRY AND COVERAGE

• The analysis models base stations with stochastic geometry and defines coverage as the probability that the SINR at a typical planar location exceeds threshold T.The authors note that this typical-location SINR differs from that of a typical user because the base-station and user point processes are not independent.Source: §III · p. 16

A. Distribution of distance to the two closest neighbours

• The probability of No Coop under the rule r1 ≤ ρr2 is ρ^2.The section also derives the joint distribution of distances to the first and second closest base stations.Source: A. Distribution of distance to the two… · p. 17

B. Cooperative Channel Fading Distribution and Properties

• The cooperative fading analysis derives a Laplace transform, density, and expectation, and provides stochastic-order comparisons with the noncooperative fading case.It also introduces Laplace-Stieltjes transform ordering as a comparison tool.Source: B. Cooperative Channel Fading… · pp. 17–18

C. Interference as Shot Noise

• The interference model approximates the distances from an interfering user’s two cooperating base stations to the typical location as equal, then treats their emissions as shot noise outside radius r2.The model assigns No Coop with probability ρ^2 and Full Coop with probability 1 − ρ^2.Source: C. Interference as Shot Noise · pp. 18–20
• The interference Laplace transform is non-decreasing in ρ and r2 and non-increasing in s, while the expected interference is independent of ρ.The analysis states that larger ρ corresponds to smaller interference in Laplace-Stieltjes transform order.Source: C. Interference as Shot Noise · pp. 20–21

D. Second Neighbour Interference Elimination - Dirty Paper Coding

• Dirty Paper Coding can eliminate the second-closest base station’s interference under Full Coop, but requires the first neighbor to know the interfering signal and possible secondary-user signals.The paper frames this as an ideal scenario and notes that No Coop cannot use this cancellation.Source: D. Second Neighbour Interference… · pp. 21–22

E. General Coverage Probability

• Theorem 2 gives coverage probability as a function of ρ for specified system values, using the interference and fading Laplace transforms.For the DPC case, the interference transform is replaced by the transform for interference with second-neighbor cancellation.Source: E. General Coverage Probability · p. 22

IV. PROS AND CONS OF BS-PAIR CONFERENCING

• Cooperation can improve cell-edge coverage, but splitting power between two base stations can reduce received signal power when the second neighbor is far away.For p = 1 and equal distances r1 = r2 = 1, the expected cooperative signal power is 1 + π/4 ≈ 1.8, versus 1 for No Coop.Source: §IV · p. 23
• For some thresholds, No Coop everywhere is optimal, while intermediate ρ values can combine No Coop near base stations with cooperation at cell borders.The authors state that the coverage expression is not shown to be concave in ρ, so local optima cannot be excluded.Source: §IV · pp. 24, 26

V. NUMERICAL EVALUATION AND SIMULATIONS

• The analytical coverage results match simulations relatively well, but the gap grows as β decreases because the finite simulation window omits distant interference.The gap diminishes when the window and expected number of base stations increase.Source: §V · p. 27
• The two simulated SINR models produce almost identical results when the expectation over θ is taken in the interference term.The authors conclude this approximation simplifies analysis without considerably affecting results for the model studied.Source: §V · p. 27

A. Coverage q (ρ) Versus Threshold T, β = 4

• Geometric cooperation gains reach a maximum of over 10% for T = 0.1 to 0.5, then disappear when threshold T > 2, where No Coop is optimal throughout the plane.Source: A. Coverage q (ρ) Versus Threshold T,… · p. 28

B. Coverage q (ρ) Versus Threshold T, β = 3.2 β = 2.5

• At β = 2.5, results are similar to β = 4; reduced β explains the analytical–simulation difference, and E[N] = 20 simulated atoms with the given planar area do not sufficiently approximate analytical curves.The authors suggest considering a larger area and note that simulations may provide erroneous results due to edge effects.Source: B. Coverage q (ρ) Versus Threshold T,… · p. 28

C. Second Neighbour Interference Elimination - Dirty Paper Coding (DPC)

• The analytical and simulated curves are almost identical under the interference approximation, supporting its use for the plotted case.The approximation assumes the distances to the first and second neighbouring BSs are approximately equal; it does not hold for Full and Optimal Cooperation simulations.Source: C. Second Neighbour Interference… · p. 30
• With second-neighbour interference information, coverage gains exceed 16% for T between 0.2 and 1.The information may eliminate negative interference effects from the second (and first) neighbouring BS.Source: C. Second Neighbour Interference… · pp. 28, 30

VI. CONCLUSIONS AND FUTURE WORK

• The analyzed scheme uses limited channel information to align the two transmissions in phase, rather than exhaustive channel feedback and full beamformer adaptation.The framework is presented as extensible to other cooperation types, including cases with full instantaneous-channel knowledge.Source: §VI · p. 31
• A user chooses cooperation according to the ratio of its distances to the two closest BSs and design parameter ρ; cooperation zones lie near cell borders.The parameter ρ determines the width of these zones, where absolute coverage gains are concentrated.Source: §VI · pp. 31, 33
• Without interference information, power-splitting gains are visible for T < 2 and exceed 10%; with DPC, gains reach 17% across the T domain.The conclusion describes DPC as adapting to interfering signals from the serving pair.Source: §VI · p. 33
• The proposed geometric user policies are not necessarily optimal; the authors expect better, potentially non-geometric policies when channel fading or neighbouring-cell effects inform the choice.Choosing the number of cooperating BSs and the cooperation type is left for future investigation; this work uses two cooperating transmitters.Source: §VI · p. 33
• The framework evaluates coverage for cooperation between pairs of cellular nodes, using stochastic geometry for random node positions.The paper focuses on a particular cooperation type: power-splitting a user message between two transmitters, with data exchange and linearly dependent encoding after conferencing.Source: §VI · p. 33

A. Geometry aspects

• A 2-Voronoi cell identifies points whose two nearest stations are its associated pair, characterized by an expanding-ball test.After the first station meets the expanding ball’s boundary, the second is found when one station is inside and another meets the boundary; if the pair is equidistant, both lie on the boundary of an empty ball.Source: A. Geometry aspects · p. 38
• A 2-Voronoi cell can be constructed from 1-Voronoi cells by removing one member of the pair and combining the resulting subregions.The construction takes the union of two regions formed by intersecting a station’s 1-Voronoi cell with the other station’s cell in a diagram where the first station has been removed; the roles are then interchanged.Source: A. Geometry aspects · p. 39
• The cited passage states that the 1-Voronoi cells of zi and zn share a 1-Voronoi edge.Source: A. Geometry aspects · p. 39
• In case 3, d(zi, z) = d(zm, z) and d(zn, z) < d(zm, z) imply z ∈ V1(zn), contradicting (43).Here m differs from i and n.Source: A. Geometry aspects · p. 40
• In case 4, d(zi, z) < d(zm, z) and d(zn, z) = d(zm, z) place z in V1(zn) and int(V1(zi)), which is impossible.Here m differs from i and n.Source: A. Geometry aspects · p. 40
• The proof concludes that ∂V2(zi, zn) = ∅ implies V2(zi, zn) = ∅.This follows from the preceding geometric case analysis.Source: A. Geometry aspects · p. 40

B. Special Case for User Position

• The special scenario assigns each base station one primary user in its own 1-Voronoi cell and one secondary user in a neighboring cell.Only one of the possible 2-Voronoi cells associated with a base station is nonempty.Source: B. Special Case for User Position · p. 41
• It is feasible when every Delaunay-graph vertex has one incoming and one outgoing arc, making the problem a traveling-salesman cycle over Delaunay neighbors.Dillencourt shows that finding Hamiltonian cycles in Delaunay triangulations is NP-Complete.Source: B. Special Case for User Position · p. 41

C. The Multiple Access Channel with Conferencing Encoders

• The conferencing channel models two transmitters sending to one receiver over a multiple-access channel, with directed links of finite capacities C12 and C21.The links carry information between the encoders.Source: C. The Multiple Access Channel with… · p. 42
• With infinite-capacity conferencing links, splitting each transmitter’s power between private and common messages and using superposition coding is capacity achieving.The split is based on κ_i.Source: C. The Multiple Access Channel with… · pp. 43–44

D. Motivation on User-optimal Geometric Cooperation Policies

• For fixed choices by other users, the two nearest base stations’ optimal choice between full cooperation and no cooperation depends on their relative channel-power ratio.The decision is also described as depending on the user’s position in the 2-Voronoi cell and fading characteristics.Source: D. Motivation on User-optimal… · p. 45
• With β = 4, full cooperation is optimal when the distance to the nearest base station is at least 0.6436 times the distance to the second nearest.The result assumes no fast fading and path loss d^-β with β > 2.Source: D. Motivation on User-optimal… · p. 45
• The full-cooperation result assumes an unbounded base-station power budget; many secondary users can make full cooperation costly under power constraints.The text raises the optimal scheme under per-base-station power constraints as an open question.Source: D. Motivation on User-optimal… · p. 45

E. Geometric Locus of the Cooperation Policies

• For ρ in (0, 1), points satisfying r_i1 ≤ ρr_i2 form a disc; at ρ = 1, the locus becomes the 1-Voronoi boundary line.Here r_i1 and r_i2 are the distances to the first and second nearest base stations.Source: D. Motivation on User-optimal… · p. 45, E. Geometric Locus of the Cooperation… · p. 46

F. Proofs and Supplementary material for distance p.d.f.

• The appendix derives the joint probability density of distances from a typical location to its first and second closest base stations, and gives the first-distance density.It also states that the expected second-neighbor distance exceeds the expected first-neighbor distance.Source: F. Proofs and Supplementary material… · pp. 47–48

G. Proofs and Supplementary Material for Channel Fading Distributions

• For μ1 = μ2 = 1/2, the expected value of Z is π + 4; the paper also plots its density for μ1 = μ2 = 2.Source: G. Proofs and Supplementary Material… · p. 49, Figure 14 · p. 50
• For μ < 1, the stochastic-ordering inequality holds without a level crossing; for μ > 1, it changes at a finite crossing t*.For μ > 1, the inequality holds below t* and reverses above it.Source: G. Proofs and Supplementary Material… · p. 50

• The interference Laplace transform is non-decreasing in ρ and r2 and non-increasing in s.Source: H. Proofs and Supplementary Material… · p. 52
• At ρ = 1, no cooperation is applied throughout each interferer's Voronoi cell and the interference Laplace transform is maximal.Source: H. Proofs and Supplementary Material… · p. 53
• Increasing ρ or r2 corresponds to less interference; a larger r2 guarantees a larger empty ball around the typical location.Source: H. Proofs and Supplementary Material… · p. 53
• The expected interference is independent of the cooperation parameter ρ.Source: H. Proofs and Supplementary Material… · p. 53
• As r2 approaches zero, expected interference diverges for β > 2, irrespective of transmit power p.Source: H. Proofs and Supplementary Material… · p. 53

• Coverage probability is expressed as an expectation over the two nearest-base-station distances and evaluated through two integrals involving the fading and interference Laplace transforms.Source: §I · p. 54
• The general-fading coverage expression requires the desired-signal variable to have a finite first moment and square-integrable density, and the interference to have a square-integrable density.Source: §I · p. 54
• For ρ = 1, the coverage expression reduces to the cited expression [33, eq. (2)] with an interferer at distance r2.Source: §I · p. 54