Source-linked AI summary
Massive-MIMO Meets HetNet: Interference Coordination Through Spatial Blanking
Ansuman Adhikary, Harpreet S. Dhillon, Giuseppe Caire
TL;DR
Same-spectrum macro- and small-cell HetNets face interference that threatens the gains of small-cell densification. The paper uses massive-MIMO spatial blanking and three low-complexity coordination strategies, finding that cell-edge deployments often perform better and offloading can improve rate fairness.
Problem
In same-spectrum HetNets, cross-tier and intra-tier interference threaten the gains from deploying small cells near concentrated user hotspots.
Source: §I · p. 2Method
The paper uses massive-MIMO spatial blanking and coordinates small-cell activity, hotspot scheduling, and offloading of macrocell hotspots.
Source: §I · p. 2, B. Contributions · p. 4Results
Cell-edge small-cell deployment performs better in most cases, while offloading yields larger gains and can improve fairness across hotspot rates.
Source: §IV · p. 22Takeaways & Limitations
Aggressive offloading raises macrocell hotspot rates while lowering small-cell rates, which may remain above the small cells’ backhaul bottleneck.
Source: §IV · p. 25
Abstract
from arXiv · showhide
In this paper, we study the downlink performance of a heterogeneous cellular network (HetNet) where both macro and small cells share the same spectrum and hence interfere with each other. We assume that the users are concentrated at certain areas in the cell, i.e., they form hotspots. While some of the hotspots are assumed to have a small cell in their vicinity, the others are directly served by the macrocell. Due to a relatively small area of each hotspot, the users lying in a particular hotspot appear to be almost co-located to the macrocells, which are typically deployed at some elevation. Assuming large number of antennas at the macrocell, we exploit this directionality in the channel vectors to obtain spatial blanking, i.e., concentrating transmission energy only in certain directions while creating transmission opportunities for the small cells lying in the other directions. In addition to this inherent interference suppression, we also develop three low-complexity interference coordination strategies: (i) turn off small cells based on the amount of cross-tier interference they receive or cause to the scheduled macrocell hotspots, (ii) schedule hotspots such that treating interference as noise is approximately optimal for the resulting Gaussian interference channel, and (iii) offload some of the macrocell hotspots to nearby small cells in order to improve throughput fairness across all hotspots. For all these schemes, we study the relative merits and demerits of uniform deployment of small cells vs. deploying more small cells towards the cell center or the cell edge.
I. INTRODUCTION
The paper addresses cross-tier interference in same-spectrum HetNets with hotspot users, where small-cell densification can threaten its gains. It combines massive-MIMO spatial blanking with low-complexity coordination and compares small-cell deployment patterns.
A. Related Work
- Prior HetNet coordination commonly orthogonalizes macrocell and small-cell time-frequency resources, while multi-antenna coordination analysis remains limited and difficult with active beamforming.Spatial blanking instead uses channel directionality to mitigate interference without time- or frequency-resource orthogonalization.Source: A. Related Work · pp. 2–3
B. Contributions
- Spatial blanking lets massive-MIMO macrocells direct energy toward served hotspots while enabling small cells in other directions to transmit simultaneously.Small hotspot areas make their users appear co-located to the macrocell.Source: B. Contributions · p. 4
- Three coordination strategies address residual cross-tier interference: small-cell shutdown based on interference, hotspot scheduling for treating interference as noise, and macrocell-hotspot offloading.Spatial blanking alone may leave strong interference when a small-cell hotspot and a macro-served hotspot share a direction.Source: B. Contributions · p. 4
- Cell-edge deployment performs better in most cases, with larger gains when macrocell traffic is offloaded to small cells to improve rate fairness across hotspots.More aggressive offloading raises macrocell hotspot rates while reducing small-cell rates, which may remain above their backhaul bottleneck.Source: B. Contributions · p. 4
II. SYSTEM MODEL
- The system model assumes coexisting multi-antenna macrocells and small cells, with macrocell antenna count M much greater than small-cell count L, both much greater than one.Each base station serves fewer simultaneous users than transmit antennas.Source: §II · p. 5
A. Spatial Setup
- The single-cell spatial setup clusters users into small, co-located hotspots, assigns small cells to a subset, and leaves other groups for macrocell service or offloading.Small-cell deployment may be uniform or concentrated toward the cell interior or edge.Source: A. Spatial Setup · pp. 5–6
B. Macrocell Subsystem
The model represents macrocell users as compact, covariance-sharing hotspots and analyzes macro and small-cell transmission under massive MIMO. It also assumes locally fixed channel covariances and specifies how they must be tracked in practice.
B. Macrocell Subsystem
- The macrocell schedules G angularly separated user groups and spatially multiplexes users within each group, with Sg determined by covariance rank and a common design parameter β.The model assumes equal power across macrocell downlink streams.Source: B. Macrocell Subsystem · pp. 7–8
- A one-ring scattering model gives each user group a shared rank-rg channel covariance, decomposed into eigenvectors and eigenvalues alongside independent small-scale fading.The scatterer-ring radius is assumed much smaller than the macrocell radius.Source: B. Macrocell Subsystem · pp. 6–8
C. Small Cell Subsystem
- Small cells use zero-forcing to serve a random set of users at peak power; their channels are isotropic with distance-dependent path loss.The number served is βL, and the path-loss model includes distance and wall penetration.Source: C. Small Cell Subsystem · pp. 8–9
- Treating channel covariances as fixed is locally valid over tens of seconds, but practical operation requires adaptive learning and tracking as network conditions change.The paper specifically notes user mobility as one source of non-stationary effects.Source: C. Small Cell Subsystem · p. 8
D. Received signals
The section specifies received-signal models for both tiers. Macrocell transmission uses JSDM, while small cells use zero-forcing and experience cross-tier and inter-cell interference.
D. Received signals
- JSDM splits beamforming into a covariance-based pre-beamformer and an instantaneous effective-channel precoder, reducing feedback dimensions and allowing hybrid RF/baseband implementation.The number of RF chains depends on simultaneous data streams rather than the macrocell antenna count.Source: D. Received signals · pp. 9–10
- With per-group processing, the macrocell designs zero-forcing precoders independently from effective channels, while each group’s received signal includes active-small-cell interference and noise.The precoder separates the statistics-dependent pre-beamformer from processing based on instantaneous projected channels.Source: D. Received signals · p. 10
- Each small cell serves βL users with zero-forcing at peak power and equal stream power; users also receive interference from other active small cells and the macrocell.The small-cell precoding matrix is formed from the users’ channel matrix.Source: D. Received signals · p. 11
E. Expressions for Received SINR
The section approximates received SINR in both tiers using deterministic equivalents, with the macrocell approximation becoming common across a group as antenna count grows.
E. Expressions for Received SINR
- The macrocell-user SINR is approximated by a deterministic equivalent that is common to users in a group and holds almost surely as M tends to infinity.The approximation relies on users in a group sharing the same channel covariance matrix.Source: E. Expressions for Received SINR · p. 11
- The section also gives a deterministic-equivalent expression for the SINR of a user served by its small cell.The expression uses nf,g as defined in equation (12).Source: E. Expressions for Received SINR · p. 12
III. INTER-TIER COORDINATION STRATEGIES
The section presents inter-tier coordination options and a no-coordination baseline. The baseline exploits macrocell directionality, but the toy example shows why explicit coordination can still be needed.
III. INTER-TIER COORDINATION STRATEGIES
- The coordination strategies turn small cells off based on cross-tier interference, leave macrocell beam directions open for small cells, or offload nearby macrocell groups.The section outlines four explicit schemes, including cases where a small cell serves multiple nearby groups.Source: §III · pp. 12–13
A. No Coordination
- Without explicit coordination, the macrocell schedules G groups fairly using second-order channel statistics and favors groups with approximately orthogonal covariance eigenspaces.The algorithm updates priorities to give groups that were not recently served a higher priority.Source: A. No Coordination · pp. 13–14
- When G is not too large, JSDM directionality implicitly suppresses interference at small cells outside the served groups’ beam directions, unlike a naive isotropic scheme.All small cells remain active on every transmission resource block in this baseline.Source: A. No Coordination · pp. 15–16
- In the toy example, increasing angular separation from θ=0 nearly removes cross-tier interference and raises small-cell throughput, while reducing group distance significantly lowers macro-user rate.The distance-related macro rate drop motivates adding explicit coordination to spatial blanking.Source: Figure 2 · p. 16, A. No Coordination · pp. 16–17
B. Coordination Scheme 1: ON/OFF
The macrocell schedules user groups, then small cells use cross-tier interference and useful-signal strengths to decide whether to transmit in that slot.
B. Coordination Scheme 1: ON/OFF
- Small cells decide whether to transmit by weighing interference received from the macrocell and interference caused to scheduled macrocell groups against useful-signal strengths.The macrocell first selects the user groups using the Section III-A algorithm.Source: B. Coordination Scheme 1: ON/OFF · p. 17
- The strategy defines the macrocell-caused interference to a small-cell user and useful-signal strengths for users served by either tier.The supplied passages identify these quantities but do not provide their complete expressions.Source: B. Coordination Scheme 1: ON/OFF · p. 17
- The small-cell ON/OFF criterion uses the defined signal and interference quantities to determine whether a cell transmits or shuts down.Its design parameters ε1 and ε2 set the desired tradeoff between macrocell and small-cell throughputs.Source: B. Coordination Scheme 1: ON/OFF · p. 17
C. Coordination Scheme 2: OFFLOAD
OFFLOAD associates some macrocell-served bottleneck groups with nearby small cells, which then serve the groups using TDMA. It targets cell-edge macrocell groups but reduces small-cell user rates when cells serve multiple groups.
C. Coordination Scheme 2: OFFLOAD
- Cell-edge macrocell groups near small cells can receive weak macrocell signals and block neighboring small-cell transmissions, motivating offload to a nearby small cell.The small cell serves both groups using TDMA, reducing the interference burden imposed by the bottleneck group.Source: C. Coordination Scheme 2: OFFLOAD · pp. 17–18
- OFFLOAD associates selected macrocell-served bottleneck groups with the nearest small cell, which serves them alongside its own group using TDMA.The small cell decides whether to absorb a group according to a condition based on direct-link gains.Source: C. Coordination Scheme 2: OFFLOAD · p. 18
- Serving two or more groups in TDMA lowers small-cell users’ average throughput by a factor equal to the number of associated groups.Source: C. Coordination Scheme 2: OFFLOAD · p. 18
- Small cells can choose groups in advance and inform the macrocell not to schedule them, reducing per-slot decision-making compared with ON/OFF coordination.The information is conveyed on a timescale of a few seconds under the stated low-mobility assumption, without significant protocol overhead.Source: C. Coordination Scheme 2: OFFLOAD · pp. 18–19
D. Coordination Scheme 3: TIN
The TIN scheme greedily selects user groups whose scheduled links satisfy the TIN-optimality condition, producing a maximal but not necessarily optimal set. Across the evaluated strategies, ON/OFF coordination gives the strongest macrocell-throughput tradeoff, while cell-edge deployment performs better in most setups.
D. Coordination Scheme 3: TIN
- The TIN Selection Algorithm greedily adds groups only when their links preserve TIN optimality, terminating with a maximal set that may not be optimal.The network is modeled as interfering links whose direct and cross gains depend on allocated powers and beamforming vectors; links within a group have identical gains under deterministic-equivalent SINR approximations.Source: D. Coordination Scheme 3: TIN · pp. 19–20
IV. NUMERICAL RESULTS
- ON/OFF coordination gives good macrocell-throughput gains at the expense of small-cell throughput, followed by TIN; cell-edge deployment generally performs better.OFFLOAD also provides a good total-throughput tradeoff, especially for high G, and edge deployment gains are particularly large when macrocell traffic is offloaded.Source: §IV · pp. 21–22
- Cell-edge small-cell deployment generally yields higher rates because path loss weakens interference to small-cell users and cell-interior macrocell groups.Macrocell group rates tend to decrease when deployment shifts from cell interior toward the edge, reflecting weaker signal strength for macrocell users farther from the macrocell.Source: §IV · pp. 22–23
- ON/OFF coordination best protects macrocell group rates but lowers small-cell rates when cells shut down, whereas TIN raises macrocell rates with only minor small-cell-rate degradation.Uncoordinated transmission performs worst for macrocell group rates because all small cells transmit and increase cross-tier interference.Source: §IV · pp. 22–23
- Lower γ increases offloading, particularly for edge deployments; it raises rates of offloaded macrocell groups while reducing rates for small-cell groups sharing resources.The reduced small-cell rates may remain above the small cells’ limited-backhaul bottleneck, so aggressive offloading may still be desirable.Source: §IV · pp. 24–25
V. CONCLUSIONS
The paper studies shared-spectrum macro/small-cell networks with hotspot users, using massive-MIMO directionality for spatial blanking and proposing three explicit coordination strategies. It also identifies deployment insights and extensions involving hotspot sizes, isolated users, multiple macrocells, and uplink.
V. CONCLUSIONS
- Massive-MIMO spatial blanking focuses macrocell energy toward served hotspots, creating transmission opportunities for small cells in other directions.The paper also proposes ON/OFF coordination and TIN-based scheduling, plus offloading macrocell hotspots to small cells for throughput gains.Source: §V · p. 25
- The analysis compares uniform, cell-interior, and cell-edge small-cell deployments, while noting cell-edge is not an obvious default under directional channels and spatial blanking.Source: §V · p. 25
- Future work includes varying hotspot sizes, adding isolated users who share spatial resources with hotspots, coordinating across multiple macrocells, and analyzing the uplink.Source: §V · p. 25
TIN SELECTION ALGORITHM
The TIN selection algorithm builds a set of macrocell and small-cell user groups that satisfies TIN optimality conditions. It iteratively considers eligible groups and terminates when no further groups can be added.
TIN SELECTION ALGORITHM
- The algorithm starts from highest-priority groups and selects the group with the strongest direct link gain.It forms separate candidate sets for macrocell and small-cell groups.Source: Appendix A · p. 26
- At each iteration, groups are admitted to candidate sets only when adding them preserves the TIN optimality conditions.The algorithm applies corresponding conditions to macrocell and small-cell groups.Source: Appendix A · pp. 26–28
- The algorithm stops when no eligible macrocell or small-cell groups remain and outputs the selected groups.Source: Appendix A · pp. 28–29
- Among eligible groups, the selection heuristic maximizes the product of κ terms, which represent direct-link strength relative to interfering-link strength.Source: Appendix A · p. 28